{ "nodes": [ { "id": "402-first-machine-payments-price-before-work-tool-purchases", "title": "402-First Machine Payments (Price-Before-Work Tool Purchases)", "category": "Orchestration & Control", "status": "emerging", "slug": "402-first-machine-payments", "tags": [ "payments", "micropayments", "x402", "http-402", "agent-commerce", "tool-use", "budget-guards" ], "summary": "Server returns an HTTP 402 price quote before doing work, and the agent pays the exact amount only if it fits its budget cap" }, { "id": "abstracted-code-representation-for-review", "title": "Abstracted Code Representation for Review", "category": "UX & Collaboration", "status": "proposed", "slug": "abstracted-code-representation-for-review", "tags": [ "code-review", "verification", "abstraction", "pseudocode", "intent-based-review", "explainability", "software-quality", "human-ai-interface" ], "summary": "Shows reviewers pseudocode, intent summaries, and logical diffs of code changes, with drill-down to the real code to confirm the mapping" }, { "id": "action-caching-replay-pattern", "title": "Action Caching & Replay Pattern", "category": "Reliability & Eval", "status": "emerging", "slug": "action-caching-replay", "tags": [ "caching", "replay", "regression-testing", "cost-reduction", "deterministic", "xpath" ], "summary": "Records each agent action with XPath and frame metadata so later runs replay it without LLM calls, with LLM fallback when replay fails" }, { "id": "action-selector-pattern", "title": "Action-Selector Pattern", "category": "Orchestration & Control", "status": "emerging", "slug": "action-selector-pattern", "tags": [ "prompt-injection", "control-flow", "safety", "tool-use" ], "summary": "LLM maps user intent to a pre-approved action ID with schema-validated parameters, and tool outputs never return to the selector" }, { "id": "adaptive-sandbox-fan-out-controller", "title": "Adaptive Sandbox Fan-Out Controller", "category": "Reliability & Eval", "status": "emerging", "slug": "adaptive-sandbox-fanout-controller", "tags": [ "fan-out", "adaptive", "parallel-sandboxes", "early-stopping", "controller", "variance", "prompt-refinement" ], "summary": "Starts a small batch of parallel sandboxes, then scales up, stops early, or refines the prompt based on early success, variance, and error signals" }, { "id": "agent-circuit-breaker", "title": "Agent Circuit Breaker", "category": "Reliability & Eval", "status": "emerging", "slug": "agent-circuit-breaker", "tags": [ "circuit-breaker", "fault-tolerance", "tool-reliability", "graceful-degradation", "resilience" ], "summary": "Prevents agents from wasting tokens and time on repeatedly failing tools by tracking failure rates and temporarily disabling broken tool endpoints", "maturity": "maturing", "domains": [ "coding", "ops", "research" ] }, { "id": "agent-modes-by-model-personality", "title": "Agent Modes by Model Personality", "category": "Orchestration & Control", "status": "emerging", "slug": "agent-modes-by-model-personality", "tags": [ "model-personality", "interaction-modes", "multi-model", "ux-design", "agent-behavior", "opus", "gpt-52" ], "summary": "Offers separate working modes, each with its own prompts, tools, UI, and expectations, tuned to one model's working style" }, { "id": "agent-reinforcement-fine-tuning-agent-rft", "title": "Agent Reinforcement Fine-Tuning (Agent RFT)", "category": "Learning & Adaptation", "status": "emerging", "slug": "agent-reinforcement-fine-tuning", "tags": [ "reinforcement-learning", "fine-tuning", "tool-use", "multi-step-rl", "agent-training", "exploration" ], "summary": "Trains model weights with reinforcement learning on real tool calls and custom graders to improve domain-specific tool use and multi-step reasoning" }, { "id": "agent-sdk-for-programmatic-control", "title": "Agent SDK for Programmatic Control", "category": "Tool Use & Environment", "status": "emerging", "slug": "agent-sdk-for-programmatic-control", "tags": [ "sdk", "automation", "ci/cd", "programmatic access", "scripting", "api", "headless agent" ], "summary": "Exposes agent functions through an SDK and CLI so code can run the agent headless with set tools, permissions, and resource limits" }, { "id": "agent-assisted-scaffolding", "title": "Agent-Assisted Scaffolding", "category": "UX & Collaboration", "status": "validated-in-production", "slug": "agent-assisted-scaffolding", "tags": [ "code-generation", "bootstrapping", "scaffolding", "feature-development", "ide", "initial-setup" ], "summary": "Agent generates initial files, boilerplate, and directory structure from a high-level description so developers start on core logic" }, { "id": "agent-driven-research", "title": "Agent-Driven Research", "category": "Orchestration & Control", "status": "established", "slug": "agent-driven-research", "tags": [ "research", "information retrieval", "tool use", "iterative process", "autonomous search" ], "summary": "Agent plans its own search queries, runs them across sources, reflects on gaps, and iterates until it can write a sourced report" }, { "id": "agent-first-tool-discovery", "title": "Agent-First Tool Discovery", "category": "Tool Use & Environment", "status": "emerging", "slug": "agent-first-tool-discovery", "tags": [ "tool-discovery", "mcp", "agent-search", "service-registry", "llms-txt", "api-discovery", "agent-infrastructure" ], "summary": "Build search indexes designed for agent consumers, returning structured tool metadata ranked by agent-relevant signals instead of human SEO metrics." }, { "id": "agent-first-tooling-and-logging", "title": "Agent-First Tooling and Logging", "category": "Tool Use & Environment", "status": "established", "slug": "agent-first-tooling-and-logging", "tags": [ "tool-design", "logging", "machine-readable", "observability", "agent-environment", "mcp", "structured-output" ], "summary": "Designs tools and logs for agent consumption with one unified log stream, structured JSON output, and agent-aware CLI flags" }, { "id": "agent-friendly-workflow-design", "title": "Agent-Friendly Workflow Design", "category": "UX & Collaboration", "status": "best-practice", "slug": "agent-friendly-workflow-design", "tags": [ "human-agent collaboration", "workflow design", "agent autonomy", "task decomposition", "HCI" ], "summary": "Gives agents clear high-level goals, room for implementation choices, structured I/O, and plan review before execution" }, { "id": "agent-powered-codebase-qa-onboarding", "title": "Agent-Powered Codebase Q&A / Onboarding", "category": "Context & Memory", "status": "validated-in-production", "slug": "agent-powered-codebase-qa-onboarding", "tags": [ "code-understanding", "onboarding", "q&a", "retrieval", "search", "context-awareness", "knowledge-base" ], "summary": "Agent indexes the codebase with embeddings and code graphs and answers natural-language questions about where code is and how it behaves" }, { "id": "agentic-search-over-vector-embeddings", "title": "Agentic Search Over Vector Embeddings", "category": "Tool Use & Environment", "status": "best-practice", "slug": "agentic-search-over-vector-embeddings", "tags": [ "search", "vector-embeddings", "bash", "grep", "RAG", "agentic-RAG", "maintenance" ], "summary": "Replaces vector indexes with agent-driven grep, find, and file traversal that searches current file state on demand and refines iteratively" }, { "id": "ai-web-search-agent-loop", "title": "AI Web Search Agent Loop", "category": "Tool Use & Environment", "status": "emerging", "slug": "ai-web-search-agent-loop", "tags": [ "web-search", "serp-api", "citations", "parallel-agents", "query-translation", "operators", "grounding" ], "summary": "Coordinator agent translates queries, spawns parallel search workers across domains and time ranges, refines iteratively, and answers with citations" }, { "id": "ai-accelerated-learning-and-skill-development", "title": "AI-Accelerated Learning and Skill Development", "category": "UX & Collaboration", "status": "validated-in-production", "slug": "ai-accelerated-learning-and-skill-development", "tags": [ "developer-productivity", "learning", "skill-acquisition", "iteration", "feedback", "taste-development", "education", "junior-developer" ], "summary": "Uses AI assistants as tutors that explain errors and concepts, offer alternatives, and fade support as the developer gains skill" }, { "id": "ai-assisted-code-review-verification", "title": "AI-Assisted Code Review / Verification", "category": "Feedback Loops", "status": "emerging", "slug": "ai-assisted-code-review-verification", "tags": [ "code-review", "verification", "quality-assurance", "human-ai-collaboration", "trust", "explainability", "software-quality" ], "summary": "Uses AI tools to flag issues, summarize change intent, and explain code so human reviewers focus on intent and business logic" }, { "id": "anti-reward-hacking-grader-design", "title": "Anti-Reward-Hacking Grader Design", "category": "Reliability & Eval", "status": "emerging", "slug": "anti-reward-hacking-grader-design", "tags": [ "reward-hacking", "grading", "reinforcement-learning", "adversarial-robustness", "agent-rft" ], "summary": "Design reward functions with multi-criteria evaluation and iterative hardening to prevent models from gaming graders, ensuring training rewards align with actual task quality." }, { "id": "artifact-driven-analysis-pipeline-orchestration", "title": "Artifact-Driven Analysis Pipeline Orchestration", "category": "Orchestration & Control", "status": "emerging", "slug": "multi-step-analysis-pipeline-orchestration", "tags": [ "pipeline", "multi-step", "orchestration", "report-generation", "data-analysis", "claude-code" ], "summary": "Has an agent run independent analysis scripts, read their structured reports, and merge them into one final report or visualization" }, { "id": "asynchronous-coding-agent-pipeline", "title": "Asynchronous Coding Agent Pipeline", "category": "Reliability & Eval", "status": "proposed", "slug": "asynchronous-coding-agent-pipeline", "tags": [ "asynchronous", "pipeline", "code-agent", "parallelism" ], "summary": "Splits inference, tool execution, reward modeling, and learning into asynchronous workers linked by message queues so GPUs stay busy" }, { "id": "authenticated-authority-channel", "title": "Authenticated Authority Channel", "category": "Security & Safety", "status": "emerging", "slug": "authenticated-authority-channel", "tags": [ "prompt-injection", "authorization", "provenance", "control-plane", "tool-use" ], "summary": "Preserve a distinguishable channel for authenticated intent so retrieved content can inform reasoning without granting itself authority.", "maturity": "early", "domains": [ "ops", "research", "coding" ] }, { "id": "autonomous-workflow-agent-architecture", "title": "Autonomous Workflow Agent Architecture", "category": "Orchestration & Control", "status": "established", "slug": "autonomous-workflow-agent-architecture", "tags": [ "workflow-automation", "containerization", "multi-agent", "engineering-tasks", "tmux", "error-recovery" ], "summary": "Runs multi-step engineering workflows in containers with tmux sessions, adaptive monitoring, checkpoints, and context-aware error recovery" }, { "id": "background-agent-with-ci-feedback", "title": "Background Agent with CI Feedback", "category": "Feedback Loops", "status": "validated-in-production", "slug": "background-agent-ci", "tags": [ "asynchronous", "ci", "feedback" ], "summary": "Runs the agent in the background on its own branch and uses CI results as feedback to patch failures until green or blocked" }, { "id": "black-box-skill-invocation", "title": "Black-Box Skill Invocation", "category": "Security & Safety", "status": "emerging", "slug": "black-box-skill-invocation", "tags": [ "privacy", "skill-sharing", "black-box", "schema-only", "prompt-protection", "inter-agent", "trust" ], "summary": "Shares skills through input and output schemas only and runs them on the provider side, so prompts and code never cross the boundary" }, { "id": "board-mediated-async-inter-agent-coordination", "title": "Board-Mediated Async Inter-Agent Coordination", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "board-mediated-inter-agent-coordination", "tags": [ "multi-agent", "coordination", "async", "kanban", "board", "message-routing", "inbox", "session-continuity", "audit-trail" ], "summary": "Routes agent-to-agent messages through comment threads on board cards, with a prefix that creates an inbox card to notify the target agent" }, { "id": "budget-aware-model-routing-with-hard-cost-caps", "title": "Budget-Aware Model Routing with Hard Cost Caps", "category": "Orchestration & Control", "status": "established", "slug": "budget-aware-model-routing-with-hard-cost-caps", "tags": [ "routing", "cost-control", "multi-model", "orchestration", "reliability" ], "summary": "Routes each request to the cheapest model that meets its needs under hard cost caps, and escalates only when quality gates fail" }, { "id": "burn-the-boats", "title": "Burn the Boats", "category": "Orchestration & Control", "status": "emerging", "slug": "burn-the-boats", "tags": [ "feature-killing", "forced-evolution", "courage", "focus", "self-destruct", "obsolescence", "product-strategy" ], "summary": "Removes working but obsolete features on a hard, announced deadline so the team and users move to the new approach" }, { "id": "canary-rollout-and-automatic-rollback-for-agent-policy-changes", "title": "Canary Rollout and Automatic Rollback for Agent Policy Changes", "category": "Reliability & Eval", "status": "established", "slug": "canary-rollout-and-automatic-rollback-for-agent-policy-changes", "tags": [ "canary", "rollback", "reliability", "policy", "evaluation" ], "summary": "Ships agent policy changes to a small traffic slice first, monitors guardrail metrics, and rolls back to the last stable version automatically" }, { "id": "capability-escrow-receipt", "title": "Capability-Escrow-Receipt", "category": "Orchestration & Control", "status": "experimental-but-awesome", "slug": "capability-escrow-receipt", "tags": [ "multi-agent", "coordination", "payments", "escrow", "receipts", "capabilities", "atomic-hire" ], "summary": "Binds a signed capability listing, an atomic escrow-plus-hire step, and a signed work receipt into one loop for agent-to-agent payment" }, { "id": "chain-of-thought-monitoring-interruption", "title": "Chain-of-Thought Monitoring & Interruption", "category": "UX & Collaboration", "status": "emerging", "slug": "chain-of-thought-monitoring-interruption", "tags": [ "monitoring", "intervention", "debugging", "reasoning", "ux" ], "summary": "Streams agent reasoning and tool calls in real time so a human can interrupt and redirect early when the approach is wrong" }, { "id": "classify-then-act-for-background-agents", "title": "Classify-Then-Act for Background Agents", "category": "Orchestration & Control", "status": "emerging", "slug": "classify-then-act-for-background-agents", "tags": [ "background-agents", "task-triage", "sandboxed-build", "review-queue", "autonomy-boundary" ], "summary": "Sorts each task into has-default, needs-intent, or out-of-scope, builds only has-default work in a sandbox, and queues it for human review" }, { "id": "cli-first-skill-design", "title": "CLI-First Skill Design", "category": "Tool Use & Environment", "status": "emerging", "slug": "cli-first-skill-design", "tags": [ "cli", "skills", "shell", "dual-use", "composability", "unix-philosophy" ], "summary": "Builds each skill as a standalone CLI with JSON output and exit codes so humans and agents use the same interface" }, { "id": "cli-native-agent-orchestration", "title": "CLI-Native Agent Orchestration", "category": "Tool Use & Environment", "status": "proposed", "slug": "cli-native-agent-orchestration", "tags": [ "cli", "automation", "local-dev", "headless" ], "summary": "Exposes agent capabilities as CLI commands with JSON output and exit codes so Makefiles, Git hooks, cron jobs, and CI can script and replay agent runs" }, { "id": "code-mode-mcp-tool-interface-improvement-pattern", "title": "Code Mode MCP Tool Interface Improvement Pattern", "category": "Tool Use & Environment", "status": "established", "slug": "code-first-tool-interface-pattern", "tags": [ "tool-interface", "code-generation", "sandboxing", "mcp", "mcp-improvement", "typescript", "v8-isolates", "token-optimization" ], "summary": "LLMs generate TypeScript code to orchestrate MCP tools in ephemeral V8 isolates, eliminating token-heavy round-trips and enabling efficient multi-step workflows with 10x+ token savings." }, { "id": "code-over-api-pattern", "title": "Code-Over-API Pattern", "category": "Tool Use & Environment", "status": "established", "slug": "code-over-api-pattern", "tags": [ "token-optimization", "code-execution", "data-processing", "mcp" ], "summary": "Agent writes code that calls tools and filters data inside a sandbox, so only summaries and samples return to the context window" }, { "id": "code-then-execute-pattern", "title": "Code-Then-Execute Pattern", "category": "Tool Use & Environment", "status": "emerging", "slug": "code-then-execute-pattern", "tags": [ "dsl", "sandbox", "program-synthesis", "auditability" ], "summary": "LLM writes a sandboxed program or DSL script, a static taint checker verifies data flows, and an interpreter runs it in a locked sandbox" }, { "id": "codebase-optimization-for-agents", "title": "Codebase Optimization for Agents", "category": "UX & Collaboration", "status": "emerging", "slug": "codebase-optimization-for-agents", "tags": [ "agent-first", "human-dx", "regression", "optimization", "tooling", "codebase-design", "trade-offs", "agent-native", "feedback-loops" ], "summary": "Optimizes tooling, CLIs, tests, and docs for agents first, with one-command verify loops and machine-readable output, even if human DX regresses" }, { "id": "coding-agent-ci-feedback-loop", "title": "Coding Agent CI Feedback Loop", "category": "Feedback Loops", "status": "best-practice", "slug": "coding-agent-ci-feedback-loop", "tags": [ "CI", "coding-agent", "asynchronous", "test-driven", "feedback" ], "summary": "Agent pushes a branch, polls CI for partial failures, patches the failing files within a retry budget, and notifies when all tests pass" }, { "id": "commitment-ledger-with-reality-gated-credit", "title": "Commitment Ledger with Reality-Gated Credit", "category": "Learning & Adaptation", "status": "proposed", "slug": "commitment-ledger-reality-gated-credit", "tags": [ "commitment-ledger", "outcome-feedback", "agent-memory", "credit-assignment", "reality-gating" ], "summary": "Records agent promises and credits retrieved memory only after an externally verifiable outcome settles.", "maturity": "early", "domains": [ "coding", "research", "operations" ] }, { "id": "compounding-engineering-pattern", "title": "Compounding Engineering Pattern", "category": "Learning & Adaptation", "status": "emerging", "slug": "compounding-engineering-pattern", "tags": [ "learning", "feedback-loops", "codification", "prompts", "slash-commands", "onboarding", "knowledge-sharing" ], "summary": "After each feature, codifies agent mistakes and learnings into CLAUDE.md, slash commands, subagents, and hooks so the next feature is easier to build" }, { "id": "conditional-parallel-tool-execution", "title": "Conditional Parallel Tool Execution", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "parallel-tool-execution", "tags": [ "parallel execution", "tool orchestration", "read-only tools", "stateful tools", "agent efficiency", "agent safety", "concurrency control", "task scheduling" ], "summary": "Runs a batch of tool calls in parallel when all are read-only and in sequence when any modifies state, then returns results in request order" }, { "id": "consequence-family-coverage-audit", "title": "Consequence-Family Coverage Audit", "category": "Security & Safety", "status": "emerging", "slug": "consequence-family-coverage-audit", "tags": [ "risk-classification", "policy-authoring", "coverage-testing", "admission-control", "blind-spots" ], "summary": "Audit an agent's risk policy by enumerating families of consequence and asking which rule covers each, because a hand-written risk list reliably encodes one family and stays silent on the rest", "maturity": "early", "domains": [ "ops", "security", "fintech" ] }, { "id": "context-budget-as-a-governed-resource", "title": "Context Budget as a Governed Resource", "category": "Context & Memory", "status": "validated-in-production", "slug": "context-budget-as-a-governed-resource", "tags": [ "context-budget", "token-costs", "ghost-tokens", "compaction", "scheduled-agents" ], "summary": "Measures always-loaded context per source, alerts on trend growth, and puts hard spend caps and fan-out bounds on unattended agent runs" }, { "id": "context-window-anxiety-management", "title": "Context Window Anxiety Management", "category": "Context & Memory", "status": "emerging", "slug": "context-window-anxiety-management", "tags": [ "context-anxiety", "token-management", "premature-completion", "model-behavior" ], "summary": "Enables a large context window but caps usage lower, and adds prompts that state the token budget so the model does not rush to finish" }, { "id": "context-window-auto-compaction", "title": "Context Window Auto-Compaction", "category": "Context & Memory", "status": "validated-in-production", "slug": "context-window-auto-compaction", "tags": [ "context-management", "compaction", "overflow-recovery", "token-estimation", "transcript-validation", "api-compaction" ], "summary": "Catches context overflow errors, compacts and validates the transcript with a reserve token floor, and retries the request automatically" }, { "id": "context-minimization-pattern", "title": "Context-Minimization Pattern", "category": "Context & Memory", "status": "emerging", "slug": "context-minimization-pattern", "tags": [ "context-hygiene", "taint-removal", "prompt-injection" ], "summary": "Removes untrusted user text and tool output from context after it becomes a safe structured artifact, so later steps see only trusted data" }, { "id": "continuous-autonomous-task-loop-pattern", "title": "Continuous Autonomous Task Loop Pattern", "category": "Orchestration & Control", "status": "established", "slug": "continuous-autonomous-task-loop-pattern", "tags": [ "autonomous-execution", "task-loop", "rate-limiting", "git-automation", "cli-driven", "stream-processing" ], "summary": "Runs a loop where subagents pick the next task from a todo file, execute it in fresh context, commit it, and back off on rate limits" }, { "id": "criticgpt-style-code-review", "title": "CriticGPT-Style Code Review", "category": "Reliability & Eval", "status": "validated-in-production", "slug": "criticgpt-style-evaluation", "tags": [ "evaluation", "code-review", "critique", "quality-assurance", "bug-detection", "gpt-4" ], "summary": "Runs a specialized critic model over generated code to find bugs, security flaws, and quality issues, and feeds critical findings back for regeneration" }, { "id": "cross-agent-lesson-sharing-via-git", "title": "Cross-Agent Lesson Sharing via Git", "category": "Context & Memory", "status": "validated-in-production", "slug": "cross-agent-lesson-sharing", "tags": [ "distributed-memory", "knowledge-sharing", "git-based", "swarm-memory", "lessons-learned" ], "summary": "Agents write solved problems as markdown lessons in a shared Git repo and search them before debugging, with GitHub Issues as the coordination layer" }, { "id": "cross-cycle-consensus-relay", "title": "Cross-Cycle Consensus Relay", "category": "Orchestration & Control", "status": "emerging", "slug": "cross-cycle-consensus-relay", "tags": [ "multi-agent", "state-management", "persistence", "long-running-tasks", "orchestration", "autonomous-loops" ], "summary": "Each loop cycle reads a structured relay document, does its work, and atomically writes back decisions, open questions, and the single next action" }, { "id": "cross-domain-agent-conflict-resolution", "title": "Cross-Domain Agent Conflict Resolution", "category": "Orchestration & Control", "status": "emerging", "slug": "cross-domain-agent-conflict-resolution", "tags": [ "multi-agent", "conflict-resolution", "policy-as-code", "governance", "orchestration", "drift-detection" ], "summary": "A coordination layer that cross-references recommendations from independent domain agents, detects conflicts on shared resources, and resolves them through policy-as-code." }, { "id": "cross-protocol-agent-discovery", "title": "Cross-Protocol Agent Discovery", "category": "Tool Use & Environment", "status": "emerging", "slug": "cross-protocol-agent-discovery", "tags": [ "agent-discovery", "mcp", "registry", "interoperability", "protocol-agnostic", "a2a", "agents-txt" ], "summary": "Aggregate agent metadata across fragmented registries and protocols into a unified, protocol-agnostic discovery layer." }, { "id": "cryptographic-governance-audit-trail", "title": "Cryptographic Governance Audit Trail", "category": "Security & Safety", "status": "emerging", "slug": "cryptographic-governance-audit-trail", "tags": [ "governance", "audit-trail", "compliance", "cryptographic-signing", "policy-enforcement", "eu-ai-act", "owasp" ], "summary": "Middleware checks each tool call against a policy file, then signs a receipt of the action with ML-DSA and appends it to a tamper-evident chain" }, { "id": "curated-code-context-window", "title": "Curated Code Context Window", "category": "Context & Memory", "status": "validated-in-production", "slug": "curated-code-context-window", "tags": [ "context-management", "code-agent", "file-selection", "noise-reduction" ], "summary": "A search subagent finds the few relevant files for a task and injects only top snippets or summaries into the main coding agent's context" }, { "id": "curated-file-context-window", "title": "Curated File Context Window", "category": "Context & Memory", "status": "best-practice", "slug": "curated-file-context-window", "tags": [ "code-context", "file-scope", "relevance", "memory-management" ], "summary": "Loads only the primary task files into the main context and has a search sub-agent rank and summarize secondary files before adding them" }, { "id": "custom-sandboxed-background-agent", "title": "Custom Sandboxed Background Agent", "category": "Orchestration & Control", "status": "emerging", "slug": "custom-sandboxed-background-agent", "tags": [ "background-agent", "sandboxed", "model-agnostic", "real-time", "websocket", "custom-infra", "iterative" ], "summary": "Builds an in-house background coding agent that runs in sandboxed dev environments, streams progress over WebSocket, and iterates on compiler and test feedback" }, { "id": "dead-mans-switch-for-scheduled-agent-jobs", "title": "Dead-Man's Switch for Scheduled Agent Jobs", "category": "Reliability & Eval", "status": "validated-in-production", "slug": "dead-mans-switch-for-scheduled-agent-jobs", "tags": [ "scheduled-agents", "silent-failure", "watchdog", "observability", "automation" ], "summary": "Each scheduled job writes a success sentinel to its log, and a separate checker files a task when the sentinel is missing or stale" }, { "id": "declarative-multi-agent-topology-definition", "title": "Declarative Multi-Agent Topology Definition", "category": "Orchestration & Control", "status": "emerging", "slug": "declarative-multi-agent-topology-definition", "tags": [ "multi-agent", "topology", "declarative", "orchestration", "scaffolding", "pipeline", "fan-out", "supervisor", "gate", "cross-platform" ], "summary": "Define multi-agent systems declaratively in a single topology file — agents, flows, gates, hooks, group chats — then compile to platform-specific configurations for any agentic framework." }, { "id": "democratization-of-tooling-via-agents", "title": "Democratization of Tooling via Agents", "category": "UX & Collaboration", "status": "emerging", "slug": "democratization-of-tooling-via-agents", "tags": [ "no-code", "low-code", "citizen-developer", "tool-creation", "business-users", "automation", "custom-software" ], "summary": "Non-programmers describe a tool in natural language and iterate with an AI agent that generates and fixes the code for dashboards, scripts, or small apps" }, { "id": "denial-tracking-permission-escalation", "title": "Denial Tracking & Permission Escalation", "category": "Security & Safety", "status": "emerging", "slug": "denial-tracking-permission-escalation", "tags": [ "permissions", "denial-tracking", "escalation", "background-agents", "non-interactive", "safety" ], "summary": "Track repeated tool denials and auto-escalate to blanket permission prompts or fallback strategies, preventing wasted iterations in non-interactive contexts.", "maturity": "maturing", "domains": [ "coding", "ops", "automation" ] }, { "id": "design-time-file-partition-as-concurrency-control", "title": "Design-Time File Partition as Concurrency Control", "category": "Orchestration & Control", "status": "emerging", "slug": "design-time-file-partition", "tags": [ "multi-agent", "concurrency", "decomposition", "task-queue", "shared-checkout", "headless" ], "summary": "Each task declares every path it will write; overlapping tasks get a blocking dependency so only disjoint work runs at once on a shared checkout", "maturity": "early", "domains": [ "coding", "ops" ] }, { "id": "deterministic-grader-in-the-loop", "title": "Deterministic Grader in the Loop", "category": "Feedback Loops", "status": "emerging", "slug": "deterministic-grader-in-the-loop", "tags": [ "feedback-loop", "self-critique", "evaluation", "determinism", "llm-as-judge", "reproducibility" ], "summary": "Replace the LLM-as-judge in an agent's self-review loop with a deterministic scorer that returns the raw counts behind each sub-score, so the agent has something specific to fix." }, { "id": "deterministic-security-scanning-build-loop", "title": "Deterministic Security Scanning Build Loop", "category": "Security & Safety", "status": "proposed", "slug": "deterministic-security-scanning-build-loop", "tags": [ "security", "deterministic", "build-loop", "backpressure", "static-analysis", "supply-chain" ], "summary": "Adds SAST, SCA, and secret scanners to the build target that the agent must run after each change, so scanner failures force the agent to fix the code" }, { "id": "deterministic-threat-rule-scanning", "title": "Deterministic Threat Rule Scanning", "category": "Security & Safety", "status": "emerging", "slug": "deterministic-threat-rule-scanning", "tags": [ "security", "threat-detection", "regex", "deterministic", "prompt-injection", "tool-poisoning", "mcp", "agent-safety" ], "summary": "Apply deterministic regex rules as a first-pass security layer to detect known threat patterns in AI agent tool calls and skill definitions." }, { "id": "deterministic-zero-llm-orchestration", "title": "Deterministic Zero-LLM Orchestration", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "deterministic-zero-llm-orchestration", "tags": [ "orchestration", "multi-agent", "parallel-execution", "deterministic", "test-driven", "zero-llm-overhead" ], "summary": "A deterministic code orchestrator splits goals, runs parallel coding agents, verifies with tests, and commits, spending no LLM tokens on coordination" }, { "id": "dev-tooling-assumptions-reset", "title": "Dev Tooling Assumptions Reset", "category": "UX & Collaboration", "status": "emerging", "slug": "dev-tooling-assumptions-reset", "tags": [ "dev-tools", "assumptions", "github", "tickets", "code-review", "tooling", "agent-workflows", "paradigm-shift" ], "summary": "Audits dev tools for human-effort assumptions and replaces tickets, PR ceremony, and sprints with immediate agent dispatch, variations, and automated tests" }, { "id": "discrete-phase-separation", "title": "Discrete Phase Separation", "category": "Orchestration & Control", "status": "emerging", "slug": "discrete-phase-separation", "tags": [ "orchestration", "planning", "research", "context-management", "multi-model" ], "summary": "Splits work into separate research, planning, and implementation conversations, each with fresh context, passing only distilled outputs between phases" }, { "id": "disposable-scaffolding-over-durable-features", "title": "Disposable Scaffolding Over Durable Features", "category": "Orchestration & Control", "status": "best-practice", "slug": "disposable-scaffolding-over-durable-features", "tags": [ "bitter-lesson", "temporary-tooling", "model-centric", "adaptability", "future-proofing" ], "summary": "Treats code around the model as temporary scaffolding: build the simplest thing that works now, mark it disposable, and remove it when models improve" }, { "id": "distributed-execution-with-cloud-workers", "title": "Distributed Execution with Cloud Workers", "category": "Orchestration & Control", "status": "emerging", "slug": "distributed-execution-cloud-workers", "tags": [ "distributed-systems", "parallelization", "cloud", "worktrees", "scalability", "team-coordination" ], "summary": "Runs many agent sessions in parallel on cloud workers, each in its own git worktree, with dependency-aware scheduling, merge coordination, and approval gates" }, { "id": "dogfooding-with-rapid-iteration-for-agent-improvement", "title": "Dogfooding with Rapid Iteration for Agent Improvement", "category": "Feedback Loops", "status": "best-practice", "slug": "dogfooding-with-rapid-iteration-for-agent-improvement", "tags": [ "dogfooding", "iterative-development", "feedback-loop", "agent-improvement", "internal-testing", "product-development" ], "summary": "The agent team uses its own agent for daily work, collects feedback in low-friction channels, and ships features internally first to validate or discard them" }, { "id": "dual-llm-pattern", "title": "Dual LLM Pattern", "category": "Orchestration & Control", "status": "emerging", "slug": "dual-llm-pattern", "tags": [ "privilege-separation", "quarantined-llm", "symbolic-variables" ], "summary": "Splits work between a privileged LLM that calls tools but never sees untrusted data and a quarantined LLM that reads untrusted data but has no tools" }, { "id": "dual-rail-message-delivery", "title": "Dual-Rail Message Delivery", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "dual-rail-message-delivery", "tags": [ "multi-agent", "coordination", "message-bus", "reliability", "redundancy", "fault-detection", "human-in-the-loop", "distributed-systems" ], "summary": "Sends every inter-agent message on two independent channels through one entry point, alerts when the rails diverge, and requires ACKs to confirm completion" }, { "id": "dual-use-tool-design", "title": "Dual-Use Tool Design", "category": "Tool Use & Environment", "status": "best-practice", "slug": "dual-use-tool-design", "tags": [ "tools", "ux", "slash-commands", "hooks", "human-ai-collaboration", "consistency" ], "summary": "Builds each tool with one interface and implementation that both humans and agents can call, with the same output, permissions, and logs" }, { "id": "dynamic-code-injection-on-demand-file-fetch", "title": "Dynamic Code Injection (On-Demand File Fetch)", "category": "Tool Use & Environment", "status": "established", "slug": "dynamic-code-injection-on-demand-file-fetch", "tags": [ "file-injection", "at-mention", "slash-commands", "IDE-integration" ], "summary": "Expands @file or /load tokens in the prompt into file contents, line ranges, or summaries so the agent sees code without manual copy and paste" }, { "id": "dynamic-context-injection", "title": "Dynamic Context Injection", "category": "Context & Memory", "status": "established", "slug": "dynamic-context-injection", "tags": [ "context management", "dynamic context", "lazy loading", "slash commands", "at-mention", "interactive context" ], "summary": "Lets users inject files, folders, or saved prompts into the agent's context mid-session with @-mentions and custom slash commands" }, { "id": "economic-value-signaling-in-multi-agent-networks", "title": "Economic Value Signaling in Multi-Agent Networks", "category": "Orchestration & Control", "status": "experimental-but-awesome", "slug": "economic-value-signaling-multi-agent", "tags": [ "multi-agent", "coordination", "incentives", "economic-signaling", "peer-discovery", "value-transfer" ], "summary": "Attaches a token value to inter-agent requests so recipients prioritize by value, with a peer registry for discovery and ledger settlement" }, { "id": "egress-lockdown-no-exfiltration-channel", "title": "Egress Lockdown (No-Exfiltration Channel)", "category": "Tool Use & Environment", "status": "established", "slug": "egress-lockdown-no-exfiltration-channel", "tags": [ "network-sandbox", "exfiltration", "outbound-controls", "security" ], "summary": "Puts the agent behind a default-deny egress firewall with allowlisted destinations so stolen data has no outbound channel" }, { "id": "episodic-memory-retrieval-injection", "title": "Episodic Memory Retrieval & Injection", "category": "Context & Memory", "status": "validated-in-production", "slug": "episodic-memory-retrieval-injection", "tags": [ "episodic-memory", "vector-db", "retrieval-augmented", "context-hint" ], "summary": "Writes a structured memory record after each episode and injects the top-k similar past memories as hints when a new task starts" }, { "id": "evidence-questions-not-verdict-questions", "title": "Evidence Questions, Not Verdict Questions", "category": "Security & Safety", "status": "emerging", "slug": "evidence-questions-not-verdict-questions", "tags": [ "policy-authoring", "admission-control", "decomposition", "calibration", "confidence", "guard-rails", "determinism" ], "summary": "Never ask the model for allow/ask/block. Ask it narrow, typed, observational questions and let code reduce the answers to the verdict, because the collapsed question is the low-confidence one.", "maturity": "early", "domains": [ "security", "ops", "coding" ] }, { "id": "evidence-layered-evaluation-for-interactive-agents", "title": "Evidence-Layered Evaluation for Interactive Agents", "category": "Reliability & Eval", "status": "emerging", "slug": "evidence-layered-evaluation-for-interactive-agents", "tags": [ "evaluation", "browser-agents", "computer-use", "reproducibility", "observability" ], "summary": "Scores interactive agent runs on a task outcome assertion and links each result to layered evidence: actions, screenshots, replays, network, and messages" }, { "id": "exact-action-authorization-binding", "title": "Exact-Action Authorization Binding", "category": "Security & Safety", "status": "proposed", "slug": "exact-action-authorization-binding", "tags": [ "authorization", "approval", "tool-use", "action-binding", "TOCTOU", "receipts" ], "summary": "Bind a short-lived approval to the complete action presented to the reviewer, then rederive and compare that identity at the acting boundary before execution.", "maturity": "early", "domains": [ "ops", "coding", "fintech" ] }, { "id": "explicit-posterior-sampling-planner", "title": "Explicit Posterior-Sampling Planner", "category": "Orchestration & Control", "status": "emerging", "slug": "explicit-posterior-sampling-planner", "tags": [ "RL", "PSRL", "exploration", "planning", "decision-making" ], "summary": "Embeds Posterior Sampling for RL in the LLM's reasoning: sample a task model, plan, act, observe reward, and update the posterior" }, { "id": "extended-coherence-work-sessions", "title": "Extended Coherence Work Sessions", "category": "Reliability & Eval", "status": "rapidly-improving", "slug": "extended-coherence-work-sessions", "tags": [ "coherence", "long-running tasks", "agent capability", "llm", "complex projects" ], "summary": "Combines models with long coherence windows with context compaction, prompt caching, and persisted state so agents stay on task for hours" }, { "id": "external-credential-sync", "title": "External Credential Sync", "category": "Security & Safety", "status": "validated-in-production", "slug": "external-credential-sync", "tags": [ "credentials", "oauth", "token-sync", "keychain", "cli-integration", "auth-reuse" ], "summary": "Reads AI credentials from other tools' keychains and config files and syncs them into the agent's store, with near-expiry refresh, OAuth upgrade, and dedupe" }, { "id": "factory-over-assistant", "title": "Factory over Assistant", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "factory-over-assistant", "tags": [ "parallelism", "autonomous-agents", "factory", "assistant", "sidebar", "orchestration", "asynchronous-work" ], "summary": "Spawns several autonomous agents in parallel with automated feedback loops such as tests and builds, and checks on them later instead of watching one agent" }, { "id": "failover-aware-model-fallback", "title": "Failover-Aware Model Fallback", "category": "Reliability & Eval", "status": "validated-in-production", "slug": "failover-aware-model-fallback", "tags": [ "fallback", "reliability", "error-classification", "multi-model", "failover", "resilience" ], "summary": "Classifies each model failure by reason and falls back along a model chain only for retryable errors, failing fast on auth, billing, and user aborts" }, { "id": "feature-list-as-immutable-contract", "title": "Feature List as Immutable Contract", "category": "Orchestration & Control", "status": "emerging", "slug": "feature-list-as-immutable-contract", "tags": [ "scope-control", "acceptance-criteria", "anti-scope-creep", "long-running-agents", "task-management" ], "summary": "Defines every feature up front in a JSON list with acceptance steps; the agent may only flip a feature to passing after it verifies it" }, { "id": "filesystem-based-agent-state", "title": "Filesystem-Based Agent State", "category": "Context & Memory", "status": "established", "slug": "filesystem-based-agent-state", "tags": [ "state-management", "persistence", "resumption", "long-running-tasks" ], "summary": "Agents persist intermediate results and working state to files, creating durable checkpoints that enable workflow resumption, recovery from failures, and support for long-running tasks." }, { "id": "filesystem-mediated-host-delegation", "title": "Filesystem-Mediated Host Delegation", "category": "Tool Use & Environment", "status": "emerging", "slug": "filesystem-mediated-host-delegation", "tags": [ "sandbox-escape", "host-execution", "async-rpc", "idempotency", "spool-directory", "durability" ], "summary": "Lets a sandboxed agent write request files to a shared spool directory that a host daemon runs against a whitelist, with idempotent results the agent polls for" }, { "id": "frontier-focused-development", "title": "Frontier-Focused Development", "category": "Learning & Adaptation", "status": "emerging", "slug": "frontier-focused-development", "tags": [ "frontier", "state-of-the-art", "model-selection", "product-strategy", "learning", "innovation", "no-selector" ], "summary": "Targets only state-of-the-art models, picks the best model per use case with no user model selector, and expects to rework the product every few months" }, { "id": "graph-of-thoughts-got", "title": "Graph of Thoughts (GoT)", "category": "Feedback Loops", "status": "emerging", "slug": "graph-of-thoughts", "tags": [ "reasoning", "graph-based", "problem-solving", "thought-exploration", "backtracking", "aggregation" ], "summary": "Represents reasoning as a directed graph of thoughts so the model can branch, aggregate, refine, and revisit reasoning paths" }, { "id": "hook-based-safety-guard-rails-for-autonomous-code-agents", "title": "Hook-Based Safety Guard Rails for Autonomous Code Agents", "category": "Security & Safety", "status": "validated-in-production", "slug": "hook-based-safety-guard-rails", "tags": [ "hooks", "guard-rails", "safety", "autonomous-operation", "destructive-command-blocking", "context-monitoring", "pre-tool-use", "post-tool-use" ], "summary": "Runs small shell scripts on PreToolUse and PostToolUse hooks to block destructive commands, lint edits, and warn on context use outside the agent's reasoning" }, { "id": "human-in-the-loop-approval-framework", "title": "Human-in-the-Loop Approval Framework", "category": "UX & Collaboration", "status": "validated-in-production", "slug": "human-in-loop-approval-framework", "tags": [ "human-oversight", "safety", "approvals", "risk-management", "collaboration", "slack-integration" ], "summary": "Systematically insert human approval gates for designated high-risk functions while maintaining agent autonomy for safe operations, with multi-channel approval interfaces and comprehensive audit trails." }, { "id": "hybrid-llmcode-workflow-coordinator", "title": "Hybrid LLM/Code Workflow Coordinator", "category": "Orchestration & Control", "status": "proposed", "slug": "hybrid-llm-code-workflow-coordinator", "tags": [ "hybrid", "llm-driven", "code-driven", "coordinator", "determinism", "workflow-orchestration", "progressive-enhancement" ], "summary": "Lets each workflow pick an LLM or a code script as its coordinator, so teams prototype with the LLM and move to reviewed code when determinism matters" }, { "id": "incident-to-eval-synthesis", "title": "Incident-to-Eval Synthesis", "category": "Feedback Loops", "status": "emerging", "slug": "incident-to-eval-synthesis", "tags": [ "evals", "incidents", "reliability", "feedback", "continuous-improvement" ], "summary": "Converts each production incident into executable eval cases with pass/fail criteria and gates future releases on them" }, { "id": "inference-healed-code-review-reward", "title": "Inference-Healed Code Review Reward", "category": "Feedback Loops", "status": "proposed", "slug": "inference-healed-code-review-reward", "tags": [ "reward-modeling", "code-review", "inference-healing", "quality-assessment" ], "summary": "Replaces a binary tests-passed reward with a critic that scores correctness, style, performance, and security, explains low scores, and combines them" }, { "id": "inference-time-scaling", "title": "Inference-Time Scaling", "category": "Orchestration & Control", "status": "emerging", "slug": "inference-time-scaling", "tags": [ "scaling", "inference", "compute", "reasoning", "performance", "o1-model", "test-time-compute", "search", "verification" ], "summary": "Spends extra compute at inference time with best-of-N sampling, longer reasoning, self-refinement, search, and verification to improve output quality" }, { "id": "initializer-maintainer-dual-agent-architecture", "title": "Initializer-Maintainer Dual Agent Architecture", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "initializer-maintainer-dual-agent", "tags": [ "long-running-agents", "session-handoff", "lifecycle-specialization", "project-bootstrap", "incremental-development" ], "summary": "Uses a one-time initializer agent to create the feature list, progress files, and bootstrap script, then a coding agent that resumes from them one feature per session" }, { "id": "intelligent-bash-tool-execution", "title": "Intelligent Bash Tool Execution", "category": "Tool Use & Environment", "status": "validated-in-production", "slug": "intelligent-bash-tool-execution", "tags": [ "bash", "shell", "pty", "fallback", "security", "process-management", "sandboxing" ], "summary": "Runs agent shell commands through a multi-mode executor that uses a PTY when needed, falls back to direct exec, and manages approvals, background jobs, and signals" }, { "id": "inversion-of-control", "title": "Inversion of Control", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "inversion-of-control", "tags": [ "orchestration", "autonomy", "control" ], "summary": "Gives the agent tools, a high-level objective, and guardrails, then lets it choose sequencing and recovery while humans set policy and review" }, { "id": "isolated-vm-per-rl-rollout", "title": "Isolated VM per RL Rollout", "category": "Security & Safety", "status": "emerging", "slug": "isolated-vm-per-rl-rollout", "tags": [ "isolation", "security", "reinforcement-learning", "infrastructure", "state-management", "agent-rft" ], "summary": "Spin up an isolated virtual machine for each RL rollout to prevent cross-contamination between parallel agent executions, ensuring safe training with destructive tool access." }, { "id": "iterative-multi-agent-brainstorming", "title": "Iterative Multi-Agent Brainstorming", "category": "Orchestration & Control", "status": "experimental-but-awesome", "slug": "iterative-multi-agent-brainstorming", "tags": [ "multi-agent", "brainstorming", "parallel processing", "idea generation", "sub-agents", "collaborative ideation" ], "summary": "Spawns several agents in parallel on the same problem, often with different perspectives, then merges their ideas into one set of options" }, { "id": "iterative-prompt-skill-refinement", "title": "Iterative Prompt & Skill Refinement", "category": "Feedback Loops", "status": "established", "slug": "iterative-prompt-skill-refinement", "tags": [ "refinement", "iteration", "prompts", "skills", "feedback", "multi-mechanism", "continuous-improvement", "dashboards" ], "summary": "Combines a feedback channel, editable prompt documents, log-driven skill fixes, and usage dashboards to improve agent prompts and skills continuously" }, { "id": "lane-based-execution-queueing", "title": "Lane-Based Execution Queueing", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "lane-based-execution-queueing", "tags": [ "queueing", "concurrency", "lanes", "isolation", "parallelism", "deadlock-prevention" ], "summary": "Routes work into named lanes, each with its own queue and concurrency limit, so sessions never interleave and background jobs never block users" }, { "id": "language-agent-tree-search-lats", "title": "Language Agent Tree Search (LATS)", "category": "Orchestration & Control", "status": "emerging", "slug": "language-agent-tree-search-lats", "tags": [ "search", "monte-carlo", "tree-search", "reasoning", "planning", "reflection", "evaluation" ], "summary": "Runs Monte Carlo Tree Search over reasoning steps, with the LLM generating candidate actions and scoring partial solutions to pick the best path" }, { "id": "latent-demand-product-discovery", "title": "Latent Demand Product Discovery", "category": "UX & Collaboration", "status": "best-practice", "slug": "latent-demand-product-discovery", "tags": [ "product-discovery", "extensibility", "hackable-products", "power-users", "latent-demand" ], "summary": "Builds hackable, extensible products, watches how power users repurpose them, and turns the most frequent workarounds into supported features" }, { "id": "layered-configuration-context", "title": "Layered Configuration Context", "category": "Context & Memory", "status": "established", "slug": "layered-configuration-context", "tags": [ "context management", "configuration", "scoped context", "automatic loading", "CLAUDE.md" ], "summary": "Loads context files from enterprise, user, project, and local levels automatically and merges them into the agent's baseline context" }, { "id": "lethal-trifecta-threat-model", "title": "Lethal Trifecta Threat Model", "category": "Reliability & Eval", "status": "best-practice", "slug": "lethal-trifecta-threat-model", "tags": [ "security", "prompt-injection", "threat-model", "data-exfiltration" ], "summary": "Classifies each tool by private-data access, untrusted-content exposure, and external communication, and blocks any execution path that has all three" }, { "id": "llm-map-reduce-pattern", "title": "LLM Map-Reduce Pattern", "category": "Orchestration & Control", "status": "emerging", "slug": "llm-map-reduce-pattern", "tags": [ "map-reduce", "sub-agents", "isolation", "untrusted-data" ], "summary": "Processes each untrusted document in its own sandboxed LLM call with a constrained output, then aggregates the results with deterministic code" }, { "id": "llm-observability", "title": "LLM Observability", "category": "Reliability & Eval", "status": "proposed", "slug": "llm-observability", "tags": [ "observability", "logging", "debugging", "tracing", "datadog", "langsmith", "spans", "monitoring", "llmops" ], "summary": "Sends agent runs to an LLM observability platform for span-level traces of each LLM call and tool use, plus aggregate cost, latency, and success metrics" }, { "id": "llm-friendly-api-design", "title": "LLM-Friendly API Design", "category": "Tool Use & Environment", "status": "emerging", "slug": "llm-friendly-api-design", "tags": [ "api-design", "llm-interaction", "tool-use", "system-design", "code-structure", "agent-compatibility" ], "summary": "Designs APIs with visible versions, self-descriptive names and schemas, simple calls, actionable errors, and few indirection levels so LLMs call them correctly" }, { "id": "local-first-credential-broker", "title": "Local-First Credential Broker", "category": "Security & Safety", "status": "emerging", "slug": "local-first-credential-broker", "tags": [ "credentials", "oauth", "api-keys", "proxy", "local-first", "agent-identity" ], "summary": "Keep raw secrets out of the agent process by injecting credentials at the network layer through a local broker, rather than handing the agent environment variables or config files." }, { "id": "markdown-polis-multi-vendor-agent-coordination-via-filesystem-constitution", "title": "Markdown Polis — Multi-Vendor Agent Coordination via Filesystem Constitution", "category": "Orchestration & Control", "status": "emerging", "slug": "markdown-polis-coordination", "tags": [ "multi-agent", "coordination", "markdown", "bandit-routing", "governance", "vendor-agnostic" ], "summary": "Coordinates agents from different vendors through versioned markdown files: capability cards, work contracts, and bandit routing that learns from settled work", "maturity": "maturing", "domains": [ "coding", "research", "ops", "documentation" ] }, { "id": "mcp-pattern-injection", "title": "MCP Pattern Injection", "category": "Tool Use & Environment", "status": "validated-in-production", "slug": "mcp-pattern-injection", "tags": [ "mcp", "code-patterns", "tool-injection", "context-enhancement", "langgraph" ], "summary": "Runs an MCP server that exposes framework best-practice patterns as tools, so the coding assistant fetches current patterns on demand" }, { "id": "memory-reinforcement-learning-memrl", "title": "Memory Reinforcement Learning (MemRL)", "category": "Learning & Adaptation", "status": "proposed", "slug": "memory-reinforcement-learning-memrl", "tags": [ "reinforcement-learning", "episodic-memory", "self-evolution", "value-aware-retrieval", "runtime-learning", "stability-plasticity" ], "summary": "Stores memories with learned utility scores, retrieves by similarity then re-ranks by utility, and updates scores from outcomes while the LLM stays frozen" }, { "id": "memory-synthesis-from-execution-logs", "title": "Memory Synthesis from Execution Logs", "category": "Context & Memory", "status": "emerging", "slug": "memory-synthesis-from-execution-logs", "tags": [ "memory", "logs", "diary", "synthesis", "pattern-detection", "knowledge-extraction", "learning" ], "summary": "Has the agent write a structured diary per task, then runs synthesis agents over many diaries to turn recurring patterns into rules, commands, and tests" }, { "id": "merged-code-language-skill-model", "title": "Merged Code + Language Skill Model", "category": "Reliability & Eval", "status": "emerging", "slug": "merged-code-language-skill-model", "tags": [ "model-merging", "transfer-learning", "coding-agent", "multilingual" ], "summary": "Fine-tunes separate language and code specialists from the same base model, then merges their weights into one model instead of one large joint training run" }, { "id": "milestone-escrow-for-agent-resource-funding", "title": "Milestone Escrow for Agent Resource Funding", "category": "UX & Collaboration", "status": "emerging", "slug": "agentfund-crowdfunding", "tags": [ "resource-funding", "escrow", "milestones", "agent-governance", "budget-controls" ], "summary": "Holds agent funding in escrow and releases each payment only after independent verification of a measurable milestone" }, { "id": "multi-model-orchestration-for-complex-edits", "title": "Multi-Model Orchestration for Complex Edits", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "multi-model-orchestration-for-complex-edits", "tags": [ "multi-model", "code-generation", "code-editing", "retrieval", "pipeline", "complex-tasks" ], "summary": "Splits complex code edits across specialized models: a retrieval model gathers context, a large model writes the changes, and smaller models apply them" }, { "id": "multi-platform-communication-aggregation", "title": "Multi-Platform Communication Aggregation", "category": "Tool Use & Environment", "status": "emerging", "slug": "multi-platform-communication-aggregation", "tags": [ "search", "aggregation", "parallel", "communication", "unified-interface" ], "summary": "Queries every communication platform in parallel through adapters that share one schema, then merges, deduplicates, and ranks the results" }, { "id": "multi-platform-webhook-triggers", "title": "Multi-Platform Webhook Triggers", "category": "Tool Use & Environment", "status": "emerging", "slug": "multi-platform-webhook-triggers", "tags": [ "webhooks", "triggers", "integrations", "slack", "notion", "jira", "scheduled-events", "event-driven" ], "summary": "Starts agent workflows from Slack, Notion, and Jira webhooks, emoji reactions, and schedules, with idempotency and signature checks on each event" }, { "id": "no-token-limit-magic", "title": "No-Token-Limit Magic", "category": "Reliability & Eval", "status": "experimental-but-awesome", "slug": "no-token-limit-magic", "tags": [ "performance", "cost", "experimentation" ], "summary": "Removes hard token limits during prototyping to learn what good behavior needs, then compresses context only after quality is stable and measured" }, { "id": "non-custodial-spending-controls", "title": "Non-Custodial Spending Controls", "category": "Security & Safety", "status": "emerging", "slug": "non-custodial-spending-controls", "tags": [ "wallet-controls", "spend-limits", "policy-enforcement", "non-custodial", "AI-agents", "safety" ], "summary": "Puts a policy layer between the agent and the transaction signer that checks each intent against allowlists, budgets, and rate limits, and fails closed" }, { "id": "non-generative-judgment-routing-with-typed-escalation", "title": "Non-Generative Judgment Routing with Typed Escalation", "category": "Orchestration & Control", "status": "emerging", "slug": "non-generative-judgment-routing", "tags": [ "routing", "latency", "structured-outputs", "classification", "escalation", "orchestration", "agent-loops" ], "summary": "Sends decision-only loop steps as batched typed questions to a non-generative judgment model and escalates what it cannot answer back to the LLM", "domains": [ "coding", "ops", "browser-automation" ] }, { "id": "one-os-user-per-agent", "title": "One OS User per Agent", "category": "Security & Safety", "status": "validated-in-production", "slug": "one-os-user-per-agent", "tags": [ "isolation", "multi-agent", "systemd", "process-identity", "least-privilege", "long-running-agents", "audit-trail" ], "summary": "Give each long-lived agent its own OS user account and a templated init-system unit, so identity, supervision and privilege come from the host instead of from a per-agent container." }, { "id": "opponent-processor-multi-agent-debate-pattern", "title": "Opponent Processor / Multi-Agent Debate Pattern", "category": "Orchestration & Control", "status": "emerging", "slug": "opponent-processor-multi-agent-debate", "tags": [ "multi-agent", "debate", "adversarial", "bias-reduction", "uncorrelated-context", "validation" ], "summary": "Spawns agents with opposing roles on the same context, lets them critique each other, then synthesizes their positions to expose bias and blind spots" }, { "id": "oracle-and-worker-multi-model-approach", "title": "Oracle and Worker Multi-Model Approach", "category": "Orchestration & Control", "status": "emerging", "slug": "oracle-and-worker-multi-model", "tags": [ "multi-model", "cost-optimization", "strategic-reasoning", "architecture" ], "summary": "Uses a fast, low-cost worker model for most tool use and code generation, and lets it consult an expensive oracle model when it is stuck" }, { "id": "orchestration-prompt-writing-benchmark", "title": "Orchestration Prompt-Writing Benchmark", "category": "Reliability & Eval", "status": "emerging", "slug": "orchestration-prompt-writing-benchmark", "tags": [ "multi-agent", "orchestration", "benchmark", "evaluation", "prompt-engineering", "role-assignment", "communication-topology", "sub-agent-prompting" ], "summary": "Scores an orchestrator on whether its sub-agent prompts give the right information fragments to the right roles, separate from end-task success", "maturity": "early", "domains": [ "multi-agent-systems", "evaluation", "orchestration" ] }, { "id": "out-of-process-provider-boundary-replay", "title": "Out-of-Process Provider-Boundary Replay", "category": "Reliability & Eval", "status": "emerging", "slug": "out-of-process-provider-boundary-replay", "tags": [ "record-replay", "determinism", "regression-testing", "http-boundary", "offline", "instrumentation-free" ], "summary": "Capture an agent run at the model-provider HTTP boundary from outside the process, then serve those bytes back so the agent re-executes its own logic with no provider call.", "maturity": "maturing", "domains": [ "coding", "ops" ] }, { "id": "output-verification-loop", "title": "Output Verification Loop", "category": "Reliability & Eval", "status": "emerging", "slug": "output-verification-loop", "tags": [ "verification", "hallucination-detection", "trust-scores", "fact-checking", "multi-agent" ], "summary": "Verify LLM outputs by extracting individual claims, checking each against evidence sources, and returning per-claim trust scores before acting on the result." }, { "id": "own-check-fault-injection", "title": "Own-Check Fault Injection", "category": "Reliability & Eval", "status": "emerging", "slug": "own-check-fault-injection", "tags": [ "fault-injection", "evals", "reliability", "self-verification", "guardrails", "chaos-engineering", "observability" ], "summary": "Plant a controlled fault inside a running agent pipeline and score whether the pipeline's own checks emit a detection act, keeping detected, reacted and recovered as separate verdicts.", "maturity": "early", "domains": [ "ops", "coding", "research" ] }, { "id": "parallel-tool-call-learning", "title": "Parallel Tool Call Learning", "category": "Orchestration & Control", "status": "emerging", "slug": "parallel-tool-call-learning", "tags": [ "parallelization", "latency-optimization", "tool-use", "reinforcement-learning", "performance" ], "summary": "Uses agent reinforcement fine-tuning to teach the model to issue independent tool calls in parallel, which cuts sequential rounds and latency" }, { "id": "patch-steering-via-prompted-tool-selection", "title": "Patch Steering via Prompted Tool Selection", "category": "Tool Use & Environment", "status": "best-practice", "slug": "patch-steering-via-prompted-tool-selection", "tags": [ "patching", "prompt-steering", "tool-selection", "coding-agent" ], "summary": "Tells the agent in the prompt which patch or refactoring tool to use, with usage examples, negative rules, and a fallback order" }, { "id": "persistent-test-memory-feedback-loop", "title": "Persistent Test Memory Feedback Loop", "category": "Learning & Adaptation", "status": "emerging", "slug": "persistent-test-memory-feedback-loop", "tags": [ "testing-agents", "persistent-memory", "feedback-loop", "self-healing", "regression-testing" ], "summary": "Preserve validated test lessons so an agent can reuse successful paths, recognize recurring failures, and retire stale knowledge.", "maturity": "early", "domains": [ "software-testing", "web", "mobile" ] }, { "id": "pii-tokenization", "title": "PII Tokenization", "category": "Security & Safety", "status": "established", "slug": "pii-tokenization", "tags": [ "privacy", "pii", "security", "mcp", "data-protection" ], "summary": "Replaces PII in tool results with placeholder tokens before the model sees them and restores the real values in outgoing tool calls" }, { "id": "plan-then-execute-pattern", "title": "Plan-Then-Execute Pattern", "category": "Orchestration & Control", "status": "established", "slug": "plan-then-execute-pattern", "tags": [ "planning", "control-flow-integrity", "prompt-injection" ], "summary": "Has the LLM fix the full sequence of tool calls before it reads untrusted data, then runs that sequence so tool outputs change only parameters" }, { "id": "planner-worker-separation-for-long-running-agents", "title": "Planner-Worker Separation for Long-Running Agents", "category": "Orchestration & Control", "status": "emerging", "slug": "planner-worker-separation-for-long-running-agents", "tags": [ "multi-agent", "coordination", "long-running", "hierarchical", "parallelism" ], "summary": "Splits agents into planners that create tasks, workers that complete them in isolation, and a judge that decides each cycle whether to continue" }, { "id": "policy-gated-tool-proxy", "title": "Policy-Gated Tool Proxy", "category": "Security & Safety", "status": "emerging", "slug": "policy-gated-tool-proxy", "tags": [ "governance", "policy-engine", "mcp", "audit-trail", "tool-proxy", "access-control", "compliance" ], "summary": "Insert a transparent proxy between agents and tool servers that evaluates every tool call against a policy engine before forwarding, producing an immutable audit trail of all decisions." }, { "id": "precomputed-code-graph-lookup", "title": "Precomputed Code Graph Lookup", "category": "Tool Use & Environment", "status": "emerging", "slug": "precomputed-code-graph-lookup", "tags": [ "code-graph", "static-analysis", "precomputed-index", "impact-analysis", "deterministic", "tree-sitter", "tool-design" ], "summary": "Answer an agent's structural code questions from a precomputed, deterministic graph queried by intent-shaped verbs, instead of a retrieval-time search-and-read loop.", "maturity": "early" }, { "id": "proactive-agent-state-externalization", "title": "Proactive Agent State Externalization", "category": "Context & Memory", "status": "emerging", "slug": "proactive-agent-state-externalization", "tags": [ "state-externalization", "memory-management", "self-documentation", "note-taking" ], "summary": "Gives agents note templates, completeness checks, and an external memory fallback so self-written notes keep objectives, decisions, and knowledge gaps" }, { "id": "proactive-trigger-vocabulary", "title": "Proactive Trigger Vocabulary", "category": "UX & Collaboration", "status": "emerging", "slug": "proactive-trigger-vocabulary", "tags": [ "ux", "triggers", "intent-detection", "skill-routing", "natural-language" ], "summary": "Gives each skill an explicit, documented list of trigger phrases and patterns so input routes to skills predictably, with optional proactive activation" }, { "id": "progressive-autonomy-with-model-evolution", "title": "Progressive Autonomy with Model Evolution", "category": "Orchestration & Control", "status": "best-practice", "slug": "progressive-autonomy-with-model-evolution", "tags": [ "model-evolution", "scaffolding", "autonomy", "system-prompts", "capabilities", "model-intelligence" ], "summary": "Audits prompts and orchestration after each model upgrade and removes the scaffolding that evals show the new model no longer needs" }, { "id": "progressive-complexity-escalation", "title": "Progressive Complexity Escalation", "category": "Orchestration & Control", "status": "emerging", "slug": "progressive-complexity-escalation", "tags": [ "capabilities", "gradual-rollout", "risk-management", "evolution", "adaptive-systems", "complexity-management" ], "summary": "Deploys agents on low-complexity, high-reliability tasks first and unlocks higher capability tiers when metrics and human review gates show proven reliability" }, { "id": "progressive-disclosure-for-large-files", "title": "Progressive Disclosure for Large Files", "category": "Context & Memory", "status": "emerging", "slug": "progressive-disclosure-large-files", "tags": [ "progressive-disclosure", "large-files", "context-optimization", "file-management", "lazy-loading", "file-handling", "metadata" ], "summary": "Puts only file metadata in the prompt and gives the agent load, peek, and extract tools to pull file content into context on demand" }, { "id": "progressive-tool-discovery", "title": "Progressive Tool Discovery", "category": "Tool Use & Environment", "status": "established", "slug": "progressive-tool-discovery", "tags": [ "mcp", "tool-discovery", "context-optimization", "lazy-loading" ], "summary": "Organizes tools in a browsable hierarchy and lets the agent load names, descriptions, or full schemas only for the tools it needs" }, { "id": "prompt-caching-via-exact-prefix-preservation", "title": "Prompt Caching via Exact Prefix Preservation", "category": "Context & Memory", "status": "emerging", "slug": "prompt-caching-via-exact-prefix-preservation", "tags": [ "prompt-caching", "exact-prefix", "performance", "stateless", "zero-data-retention", "message-ordering", "optimization" ], "summary": "Keeps static prompt content first in a fixed order and only appends new messages, including config changes, so each request reuses the cached prefix" }, { "id": "reasoning-token-firewall", "title": "Reasoning-Token Firewall", "category": "Reliability & Eval", "status": "validated-in-production", "slug": "reasoning-token-firewall", "tags": [ "reasoning", "chain-of-thought", "streaming", "output-hygiene", "safety", "harness" ], "summary": "Assemble an agent's result only from answer-typed stream events, never by string-stripping interleaved reasoning tokens." }, { "id": "recursive-best-of-n-delegation", "title": "Recursive Best-of-N Delegation", "category": "Orchestration & Control", "status": "emerging", "slug": "recursive-best-of-n-delegation", "tags": [ "recursion", "best-of-n", "parallel-sandboxes", "judge", "delegation", "rlms", "selection", "sub-agents" ], "summary": "Runs several parallel candidate workers per subtask in a recursive agent tree, scores them with tests and a judge, and promotes the best result upward" }, { "id": "reflection-loop", "title": "Reflection Loop", "category": "Feedback Loops", "status": "established", "slug": "reflection", "tags": [ "self-feedback", "iterative-improvement", "evaluation" ], "summary": "Scores each draft against a fixed rubric, feeds the critique into a revision, and repeats until the draft passes a threshold or the retry budget ends" }, { "id": "reliability-problem-map-checklist-for-rag-and-agents", "title": "Reliability Problem Map Checklist for RAG and Agents", "category": "Reliability & Eval", "status": "proposed", "slug": "wfgy-reliability-problem-map", "tags": [ "reliability", "evaluation", "rag", "agents", "debugging", "failure-modes", "checklist" ], "summary": "Runs a fixed 16-question failure checklist on each RAG or agent incident, maps the result to repair actions, and re-tests the same failing case" }, { "id": "rendered-ui-finish-gate", "title": "Rendered UI Finish Gate", "category": "Feedback Loops", "status": "emerging", "slug": "rendered-ui-finish-gate", "tags": [ "ui-quality", "frontend", "coding-agents", "verification", "design-systems", "feedback-loops" ], "summary": "Verify an agent-built interface against its intended design contract, required states, interaction semantics, and rendered output before merging.", "domains": [ "coding", "design", "frontend" ] }, { "id": "rich-feedback-loops-perfect-prompts", "title": "Rich Feedback Loops > Perfect Prompts", "category": "Feedback Loops", "status": "validated-in-production", "slug": "rich-feedback-loops", "tags": [ "feedback", "testing", "reliability", "user-feedback", "positive-reinforcement", "corrections" ], "summary": "Returns compiler errors, test failures, lint output, and human feedback to the agent after each tool call so it can plan fixes and self-correct" }, { "id": "rlaif-reinforcement-learning-from-ai-feedback", "title": "RLAIF (Reinforcement Learning from AI Feedback)", "category": "Reliability & Eval", "status": "emerging", "slug": "rlaif-reinforcement-learning-from-ai-feedback", "tags": [ "rlhf", "rlaif", "constitutional-ai", "synthetic-data", "feedback", "alignment", "evaluation" ], "summary": "Uses an AI model guided by written principles to critique outputs and label preferences, which train a reward model that optimizes the policy" }, { "id": "sandboxed-tool-authorization", "title": "Sandboxed Tool Authorization", "category": "Security & Safety", "status": "validated-in-production", "slug": "sandboxed-tool-authorization", "tags": [ "authorization", "policy", "allowlist", "deny-by-default", "pattern-matching", "subagent-security" ], "summary": "Filters an agent's tools through deny-first allow and deny patterns, profile presets, and subagent policies that inherit parent restrictions" }, { "id": "schema-validation-retry-with-cross-step-learning", "title": "Schema Validation Retry with Cross-Step Learning", "category": "Reliability & Eval", "status": "emerging", "slug": "schema-validation-retry-cross-step-learning", "tags": [ "retry", "validation", "cross-step-learning", "structured-output", "zod", "error-accumulation" ], "summary": "Retries failed structured outputs with the validation errors as feedback and adds recent errors from earlier steps to later prompts so mistakes do not repeat" }, { "id": "schema-guided-graph-retrieval-for-multi-hop-reasoning", "title": "Schema-Guided Graph Retrieval for Multi-Hop Reasoning", "category": "Context & Memory", "status": "emerging", "slug": "schema-guided-graph-retrieval", "tags": [ "graphrag", "schema-guided", "multi-hop-reasoning", "query-decomposition", "type-filtered-retrieval", "knowledge-graph", "schema-evolution", "community-detection" ], "summary": "Use one shared domain schema to align graph construction, schema evolution, query decomposition, and typed retrieval so multi-hop reasoning over private knowledge stays precise as domains change." }, { "id": "seamless-background-to-foreground-handoff", "title": "Seamless Background-to-Foreground Handoff", "category": "UX & Collaboration", "status": "emerging", "slug": "seamless-background-to-foreground-handoff", "tags": [ "background-agent", "human-in-the-loop", "task-handoff", "interactive-refinement", "agent-collaboration", "developer-workflow" ], "summary": "Lets a user take over a background agent's unfinished work in the foreground, with the agent's branch, PR, and summaries carried over as context" }, { "id": "self-critique-evaluator-loop", "title": "Self-Critique Evaluator Loop", "category": "Feedback Loops", "status": "established", "slug": "self-critique-evaluator-loop", "tags": [ "self-critique", "evaluator", "reward-model", "synthetic-data", "reflexion", "rlaif" ], "summary": "Trains a judge model on its own synthetic comparisons of candidate outputs and uses it as a reward model or quality gate for the main agent" }, { "id": "self-discover-llm-self-composed-reasoning-structures", "title": "Self-Discover: LLM Self-Composed Reasoning Structures", "category": "Feedback Loops", "status": "emerging", "slug": "self-discover-reasoning-structures", "tags": [ "reasoning", "self-improvement", "meta-learning", "problem-solving", "task-specific", "optimization" ], "summary": "Has the LLM select, adapt, and compose reasoning modules into a task-specific reasoning structure, then solve the task by following that structure" }, { "id": "self-identity-accumulation", "title": "Self-Identity Accumulation", "category": "Context & Memory", "status": "emerging", "slug": "self-identity-accumulation", "tags": [ "self-identity", "persona", "session-hooks", "familiarity", "cross-session", "profile", "soul-document", "agent-personality" ], "summary": "Injects a persistent identity document at session start and updates it with new user insights at session end through lifecycle hooks" }, { "id": "self-rewriting-meta-prompt-loop", "title": "Self-Rewriting Meta-Prompt Loop", "category": "Orchestration & Control", "status": "emerging", "slug": "self-rewriting-meta-prompt-loop", "tags": [ "meta-prompting", "self-improvement", "system-prompt", "reflection" ], "summary": "Has the agent reflect after each episode, draft edits to its own system prompt, validate them through guardrails, and save the new version" }, { "id": "semantic-context-filtering-pattern", "title": "Semantic Context Filtering Pattern", "category": "Context & Memory", "status": "emerging", "slug": "semantic-context-filtering", "tags": [ "context-filtering", "token-optimization", "semantic-extraction", "noise-reduction" ], "summary": "Extracts only the semantic or interactive parts of raw data, such as accessibility trees or relevant fields, before sending it to the LLM" }, { "id": "session-scoped-context-runtime-for-agent-tools", "title": "Session-Scoped Context Runtime for Agent Tools", "category": "Context & Memory", "status": "emerging", "slug": "session-scoped-context-runtime-for-agent-tools", "tags": [ "mcp", "context-compression", "session-cache", "agent-tools", "coding-assistants" ], "summary": "Interpose a context runtime that caches structured reads and normalizes tool output so sessions reuse compact representations instead of repeating raw tokens.", "domains": [ "coding" ] }, { "id": "shell-command-contextualization", "title": "Shell Command Contextualization", "category": "Tool Use & Environment", "status": "established", "slug": "shell-command-contextualization", "tags": [ "shell integration", "context management", "local execution", "bash", "cli", "interactive tools" ], "summary": "Lets the user run a shell command with a prefix such as ! and injects the command and its full output into the agent's context" }, { "id": "shipping-as-research", "title": "Shipping as Research", "category": "Learning & Adaptation", "status": "emerging", "slug": "shipping-as-research", "tags": [ "research", "experimentation", "rapid-iteration", "learning", "dogfooding", "shipping", "uncertainty" ], "summary": "Releases reversible, instrumented features to learn whether they work, then doubles down or removes them based on usage data and feedback" }, { "id": "signal-driven-agent-activation", "title": "Signal-Driven Agent Activation", "category": "Orchestration & Control", "status": "emerging", "slug": "signal-driven-agent-activation", "tags": [ "signals", "event-driven", "automation", "orchestration", "reactive" ], "summary": "Watches external sources for structured signals and starts predefined agent workflows when declarative rules with thresholds and cooldowns match" }, { "id": "skill-activation-as-a-precisionrecall-measurement", "title": "Skill Activation as a Precision/Recall Measurement", "category": "Reliability & Eval", "status": "emerging", "slug": "skill-activation-as-precision-recall", "tags": [ "evals", "skills", "routing", "precision-recall", "classification", "ci-cd", "regression-testing" ], "summary": "Measures skill routing as classification with a labelled prompt dataset, per-skill precision and recall, and CI thresholds that fail the build", "domains": [ "coding", "ops" ] }, { "id": "skill-library-evolution", "title": "Skill Library Evolution", "category": "Learning & Adaptation", "status": "established", "slug": "skill-library-evolution", "tags": [ "code-reuse", "skills", "learning", "capabilities", "evolution", "progressive-disclosure", "on-demand-loading", "mcp", "lazy-loading" ], "summary": "Saves working agent code as reusable skills in a skills directory, documents and tests them over time, and loads skill details only on demand" }, { "id": "soulbound-identity-verification", "title": "Soulbound Identity Verification", "category": "Security & Safety", "status": "emerging", "slug": "soulbound-identity-verification", "tags": [ "identity", "verification", "trust", "soulbound-token", "blockchain", "agent-identity" ], "summary": "Binds agent identity to a non-transferable credential with a committed state hash and logs signed state changes so verifiers can check continuity" }, { "id": "spec-as-test-feedback-loop", "title": "Spec-As-Test Feedback Loop", "category": "Feedback Loops", "status": "emerging", "slug": "spec-as-test-feedback-loop", "tags": [ "validation", "drift-detection", "continuous-testing" ], "summary": "Generates executable tests from the spec on every spec or code commit and opens agent PRs that fix code or flag unclear spec parts" }, { "id": "specification-driven-agent-development", "title": "Specification-Driven Agent Development", "category": "Orchestration & Control", "status": "proposed", "slug": "specification-driven-agent-development", "tags": [ "spec-first", "scaffolding", "contract", "requirements" ], "summary": "Makes a version-controlled spec file the agent's main input, scaffolds code from it, and links each artifact back to a spec clause" }, { "id": "spectrum-of-control-blended-initiative", "title": "Spectrum of Control / Blended Initiative", "category": "UX & Collaboration", "status": "validated-in-production", "slug": "spectrum-of-control-blended-initiative", "tags": [ "human-agent-collaboration", "autonomy-spectrum", "interactive-control", "task-delegation", "code-editing", "ide-integration" ], "summary": "Gives users several autonomy modes, from inline completion to background agents, and lets them switch modes per task" }, { "id": "static-service-manifest-for-agents", "title": "Static Service Manifest for Agents", "category": "Tool Use & Environment", "status": "emerging", "slug": "static-service-manifest-for-agents", "tags": [ "service-discovery", "agent-infrastructure", "llms-txt", "machine-readable", "api-design", "well-known", "tool-discovery" ], "summary": "Serves a static llms.txt or agent.json file at a well-known URL that lists services, auth, and limits so agents can plan before they call" }, { "id": "stop-hook-auto-continue-pattern", "title": "Stop Hook Auto-Continue Pattern", "category": "Orchestration & Control", "status": "emerging", "slug": "stop-hook-auto-continue-pattern", "tags": [ "hooks", "automation", "testing", "determinism", "success-criteria", "continuous-execution" ], "summary": "Runs a stop hook after each agent turn that checks success criteria and makes the agent continue until tests or checks pass" }, { "id": "structured-output-specification", "title": "Structured Output Specification", "category": "Reliability & Eval", "status": "established", "slug": "structured-output-specification", "tags": [ "structured-output", "schema", "validation", "reliability", "type-safety", "integration" ], "summary": "Constrain agent outputs using deterministic schemas that enforce structured, machine-readable results, enabling reliable validation, parsing, and integration with downstream systems." }, { "id": "sub-agent-spawning", "title": "Sub-Agent Spawning", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "sub-agent-spawning", "tags": [ "orchestration", "context", "scalability", "subagents", "yaml-configuration", "virtual-files", "subject-hygiene", "parallel-delegation" ], "summary": "Lets the main agent spawn sub-agents with fresh context and scoped tools to work on subtasks in parallel, then merges their results" }, { "id": "subagent-compilation-checker", "title": "Subagent Compilation Checker", "category": "Reliability & Eval", "status": "emerging", "slug": "subagent-compilation-checker", "tags": [ "subagent", "compilation", "modularity", "error-isolation" ], "summary": "Spawns one subagent per module to build and check it, and returns only a short structured error list or artifact reference to the main agent" }, { "id": "subject-hygiene-for-task-delegation", "title": "Subject Hygiene for Task Delegation", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "subject-hygiene", "tags": [ "subagents", "delegation", "traceability", "naming", "clarity", "anti-pattern" ], "summary": "Requires a specific action-plus-target subject on every subagent task so each delegated job stays traceable and easy to reference" }, { "id": "swarm-migration-pattern", "title": "Swarm Migration Pattern", "category": "Orchestration & Control", "status": "validated-in-production", "slug": "swarm-migration-pattern", "tags": [ "swarm", "map-reduce", "migration", "parallelization", "sub-agents", "scalability", "framework-migration" ], "summary": "Main agent orchestrates 10+ parallel subagents working simultaneously on independent migration chunks, achieving 10x+ speedup for large-scale framework upgrades, lint rule rollouts, and API migrations." }, { "id": "team-shared-agent-configuration-as-code", "title": "Team-Shared Agent Configuration as Code", "category": "UX & Collaboration", "status": "best-practice", "slug": "team-shared-agent-configuration", "tags": [ "configuration", "version-control", "team-collaboration", "permissions", "consistency", "onboarding" ], "summary": "Check agent configuration into version control as code, enabling consistent behavior across teams, faster onboarding, and collaborative improvement through PRs and code review." }, { "id": "three-stage-perception-architecture", "title": "Three-Stage Perception Architecture", "category": "Orchestration & Control", "status": "established", "slug": "three-stage-perception-architecture", "tags": [ "architecture", "perception", "processing", "action", "pipeline", "modular-design" ], "summary": "Splits the agent into separate perception, processing, and action stages so each stage can be built, tested, and scaled on its own" }, { "id": "tier-auto-apply-by-mechanical-impact", "title": "Tier Auto-Apply by Mechanical Impact", "category": "Reliability & Eval", "status": "emerging", "slug": "tier-auto-apply-by-mechanical-impact", "tags": [ "auto-apply", "tiered-autonomy", "self-improving-agent", "findings-triage", "reversibility" ], "summary": "Decides which self-proposed changes auto-apply by matching the real diff's file paths and change kinds against a trusted tier table, not by finding text" }, { "id": "tool-capability-compartmentalization", "title": "Tool Capability Compartmentalization", "category": "Orchestration & Control", "status": "emerging", "slug": "tool-capability-compartmentalization", "tags": [ "capability-segregation", "least-privilege", "tool-permissions" ], "summary": "Splits tools into reader, processor, and writer classes with scoped permissions and blocks tool chains that combine private data, untrusted input, and external writes" }, { "id": "tool-search-lazy-loading", "title": "Tool Search Lazy Loading", "category": "Context & Memory", "status": "emerging", "slug": "tool-search-lazy-loading", "tags": [ "mcp", "tool-discovery", "context-optimization", "lazy-loading", "search", "dynamic-loading" ], "summary": "Dynamically load tools via search instead of preloading all available tools to reduce context usage", "maturity": "maturing", "domains": [ "tool-use", "context-optimization" ] }, { "id": "tool-selection-guide", "title": "Tool Selection Guide", "category": "Orchestration & Control", "status": "emerging", "slug": "tool-selection-guide", "tags": [ "tools", "workflow", "best-practices", "efficiency", "patterns", "exploration", "modification" ], "summary": "Maps each task type to a preferred tool: Glob, Grep, and Read to explore, Edit to modify, Bash to verify, and Task with a clear subject to delegate" }, { "id": "tool-use-incentivization-via-reward-shaping", "title": "Tool Use Incentivization via Reward Shaping", "category": "Feedback Loops", "status": "emerging", "slug": "tool-use-incentivization-via-reward-shaping", "tags": [ "tool-use", "reward-shaping", "coding-agent", "RL" ], "summary": "Gives dense RL rewards for useful intermediate tool calls such as compile, lint, and test so the agent learns to use tools instead of only thinking" }, { "id": "tool-use-steering-via-prompting", "title": "Tool Use Steering via Prompting", "category": "Tool Use & Environment", "status": "best-practice", "slug": "tool-use-steering-via-prompting", "tags": [ "tool use", "prompting", "agent guidance", "custom tools", "cli", "natural language control" ], "summary": "Tells the agent in the prompt which tool to use, how to learn a custom tool, and which shorthands map to tool sequences" }, { "id": "tracker-as-desired-state-reconciliation", "title": "Tracker-as-Desired-State Reconciliation", "category": "Orchestration & Control", "status": "emerging", "slug": "tracker-as-desired-state-reconciliation", "tags": [ "orchestration", "reconciliation", "level-triggered", "issue-tracker", "dispatch", "crash-recovery", "idempotency", "human-in-the-loop" ], "summary": "Derive agent dispatch by re-reading workflow states in the team's issue tracker every tick, keeping only a short-lived claim locally, so missed events, restarts, and human edits all self-heal.", "maturity": "maturing", "domains": [ "coding", "ops" ] }, { "id": "transitive-vouch-chain-trust", "title": "Transitive Vouch-Chain Trust", "category": "Security & Safety", "status": "emerging", "slug": "transitive-vouch-chain-trust", "tags": [ "trust", "identity", "vouch", "trust-chain", "agent-identity", "decentralized-trust", "ed25519", "reputation" ], "summary": "Builds a graph of signed vouches between agents and derives trust in an unknown agent from the chain back to a trusted one, with decay at each hop" }, { "id": "tree-of-thought-reasoning", "title": "Tree-of-Thought Reasoning", "category": "Orchestration & Control", "status": "established", "slug": "tree-of-thought-reasoning", "tags": [ "branching", "deliberate-reasoning", "search" ], "summary": "Expands a tree of candidate reasoning steps, scores partial states, prunes weak branches, and picks the best path instead of one linear chain" }, { "id": "unified-tool-gateway", "title": "Unified Tool Gateway", "category": "Tool Use & Environment", "status": "emerging", "slug": "unified-tool-gateway", "tags": [ "tool-gateway", "mcp", "tool-aggregation", "agent-tooling", "api-gateway", "tool-routing" ], "summary": "Route all agent tool calls through a single gateway that handles discovery, authentication, billing, and execution across many heterogeneous providers.", "maturity": "maturing", "domains": [ "tool-use", "infrastructure", "ops" ] }, { "id": "variance-based-rl-sample-selection", "title": "Variance-Based RL Sample Selection", "category": "Learning & Adaptation", "status": "validated-in-production", "slug": "variance-based-rl-sample-selection", "tags": [ "reinforcement-learning", "sample-efficiency", "variance", "data-quality", "agent-rft" ], "summary": "Runs the base model several times per sample and trains RL only on samples with score variance, skipping ones that are always right or always wrong" }, { "id": "verbose-reasoning-transparency", "title": "Verbose Reasoning Transparency", "category": "UX & Collaboration", "status": "best-practice", "slug": "verbose-reasoning-transparency", "tags": [ "explainability", "debugging", "transparency", "agent reasoning", "verbose mode", "introspection" ], "summary": "Lets users open a verbose view on demand that shows the agent's interpretation, tool choices, intermediate steps, and raw tool outputs" }, { "id": "versioned-constitution-governance", "title": "Versioned Constitution Governance", "category": "Reliability & Eval", "status": "emerging", "slug": "versioned-constitution-governance", "tags": [ "constitution", "alignment", "governance", "signed-commits", "policy", "rlaif", "critique-revise" ], "summary": "Stores the agent constitution in a signed Git repository where the agent can only propose changes and reviewers or CI gates merge them" }, { "id": "virtual-machine-operator-agent", "title": "Virtual Machine Operator Agent", "category": "Tool Use & Environment", "status": "established", "slug": "virtual-machine-operator-agent", "tags": [ "computer operation", "virtual machine", "execution environment", "agent capability" ], "summary": "Gives the agent a dedicated virtual machine where it can run code, install packages, use the file system, and operate CLI tools" }, { "id": "visual-ai-multimodal-integration", "title": "Visual AI Multimodal Integration", "category": "Tool Use & Environment", "status": "emerging", "slug": "visual-ai-multimodal-integration", "tags": [ "multimodal", "vision", "video", "image-processing", "visual-understanding", "agent-capabilities" ], "summary": "Adds multimodal models to the agent so it can analyze images, video, and screenshots and combine them with text to reason and act" }, { "id": "workflow-evals-with-mocked-tools", "title": "Workflow Evals with Mocked Tools", "category": "Reliability & Eval", "status": "emerging", "slug": "workflow-evals-with-mocked-tools", "tags": [ "evals", "testing", "ci-cd", "mocked-tools", "simulations", "workflow-validation", "end-to-end-testing" ], "summary": "Runs complete agent workflows against mocked tools in CI and checks which tools were called plus agent-as-judge quality criteria" }, { "id": "working-memory-via-todowrite", "title": "Working Memory via TodoWrite", "category": "Context & Memory", "status": "emerging", "slug": "working-memory-via-todos", "tags": [ "context", "memory", "working-memory", "state", "todo-tracking", "dependencies", "session-management" ], "summary": "Keeps an explicit todo list with status, blockers, and next steps during the session so agent and user can track progress" }, { "id": "workspace-native-multi-agent-orchestration", "title": "Workspace-Native Multi-Agent Orchestration", "category": "Orchestration & Control", "status": "emerging", "slug": "workspace-native-multi-agent-orchestration", "tags": [ "multi-agent", "orchestration", "workflow-automation", "workspace", "mcp", "knowledge-base", "persistent-memory", "integrations", "collaboration" ], "summary": "Runs agents inside the team's workspace platform so they share its memory, knowledge sources, event triggers, and integrations with humans" }, { "id": "zero-knowledge-verified-agent-egress", "title": "Zero-Knowledge Verified Agent Egress", "category": "Security & Safety", "status": "emerging", "slug": "zero-knowledge-verified-agent-egress", "tags": [ "egress", "zero-knowledge-proof", "source-of-truth", "allow-list", "mcp", "outbound-verification", "runtime-protection" ], "summary": "Intercepts each outbound HTTP or MCP call, proves its claims against a private source of truth with a zero-knowledge proof, and blocks calls that fail" }, { "id": "zero-trust-agent-mesh", "title": "Zero-Trust Agent Mesh", "category": "Security & Safety", "status": "established", "slug": "zero-trust-agent-mesh", "tags": [ "zero-trust", "identity", "delegation", "multi-agent", "cryptography", "ed25519", "governance" ], "summary": "Gives each agent a cryptographic identity and verifies identity, signed delegation tokens, and chain depth on every inter-agent request" } ], "edges": [ { "source": "agent-circuit-breaker", "target": "failover-aware-model-fallback", "type": "related" }, { "source": "authenticated-authority-channel", "target": "action-selector-pattern", "type": "related" }, { "source": "authenticated-authority-channel", "target": "policy-gated-tool-proxy", "type": "related" }, { "source": "authenticated-authority-channel", "target": "context-minimization-pattern", "type": "related" }, { "source": "authenticated-authority-channel", "target": "sandboxed-tool-authorization", "type": "related" }, { "source": "authenticated-authority-channel", "target": "lethal-trifecta-threat-model", "type": "related" }, { "source": "board-mediated-async-inter-agent-coordination", "target": "cross-cycle-consensus-relay", "type": "related" }, { "source": "board-mediated-async-inter-agent-coordination", "target": "memory-synthesis-from-execution-logs", "type": "related" }, { "source": "classify-then-act-for-background-agents", "target": "custom-sandboxed-background-agent", "type": "related" }, { "source": "commitment-ledger-with-reality-gated-credit", "target": "episodic-memory-retrieval-injection", "type": "related" }, { "source": "commitment-ledger-with-reality-gated-credit", "target": "memory-reinforcement-learning-memrl", "type": "related" }, { "source": "commitment-ledger-with-reality-gated-credit", "target": "incident-to-eval-synthesis", "type": "related" }, { "source": "consequence-family-coverage-audit", "target": "policy-gated-tool-proxy", "type": "related" }, { "source": "consequence-family-coverage-audit", "target": "sandboxed-tool-authorization", "type": "related" }, { "source": "denial-tracking-permission-escalation", "target": "sandboxed-tool-authorization", "type": "related" }, { "source": "design-time-file-partition-as-concurrency-control", "target": "lane-based-execution-queueing", "type": "related" }, { "source": "design-time-file-partition-as-concurrency-control", "target": "workspace-native-multi-agent-orchestration", "type": "related" }, { "source": "design-time-file-partition-as-concurrency-control", "target": "deterministic-zero-llm-orchestration", "type": "related" }, { "source": "evidence-questions-not-verdict-questions", "target": "policy-gated-tool-proxy", "type": "related" }, { "source": "evidence-questions-not-verdict-questions", "target": "consequence-family-coverage-audit", "type": "related" }, { "source": "evidence-questions-not-verdict-questions", "target": "deterministic-grader-in-the-loop", "type": "related" }, { "source": "exact-action-authorization-binding", "target": "authenticated-authority-channel", "type": "related" }, { "source": "exact-action-authorization-binding", "target": "policy-gated-tool-proxy", "type": "related" }, { "source": "exact-action-authorization-binding", "target": "cryptographic-governance-audit-trail", "type": "related" }, { "source": "non-generative-judgment-routing-with-typed-escalation", "target": "budget-aware-model-routing-with-hard-cost-caps", "type": "related" }, { "source": "one-os-user-per-agent", "target": "local-first-credential-broker", "type": "related" }, { "source": "one-os-user-per-agent", "target": "custom-sandboxed-background-agent", "type": "related" }, { "source": "one-os-user-per-agent", "target": "isolated-vm-per-rl-rollout", "type": "related" }, { "source": "orchestration-prompt-writing-benchmark", "target": "declarative-multi-agent-topology-definition", "type": "related" }, { "source": "orchestration-prompt-writing-benchmark", "target": "workflow-evals-with-mocked-tools", "type": "related" }, { "source": "orchestration-prompt-writing-benchmark", "target": "sub-agent-spawning", "type": "related" }, { "source": "out-of-process-provider-boundary-replay", "target": "workflow-evals-with-mocked-tools", "type": "related" }, { "source": "output-verification-loop", "target": "reflection-loop", "type": "related" }, { "source": "output-verification-loop", "target": "self-critique-evaluator-loop", "type": "related" }, { "source": "persistent-test-memory-feedback-loop", "target": "memory-synthesis-from-execution-logs", "type": "related" }, { "source": "persistent-test-memory-feedback-loop", "target": "workflow-evals-with-mocked-tools", "type": "related" }, { "source": "precomputed-code-graph-lookup", "target": "agentic-search-over-vector-embeddings", "type": "related" }, { "source": "precomputed-code-graph-lookup", "target": "curated-code-context-window", "type": "related" }, { "source": "precomputed-code-graph-lookup", "target": "agent-powered-codebase-qa-onboarding", "type": "related" }, { "source": "reasoning-token-firewall", "target": "chain-of-thought-monitoring-interruption", "type": "related" }, { "source": "reasoning-token-firewall", "target": "verbose-reasoning-transparency", "type": "related" }, { "source": "reasoning-token-firewall", "target": "structured-output-specification", "type": "related" }, { "source": "tier-auto-apply-by-mechanical-impact", "target": "canary-rollout-and-automatic-rollback-for-agent-policy-changes", "type": "related" }, { "source": "tool-search-lazy-loading", "target": "context-minimization-pattern", "type": "related" }, { "source": "tool-search-lazy-loading", "target": "progressive-tool-discovery", "type": "related" }, { "source": "tool-search-lazy-loading", "target": "dynamic-context-injection", "type": "related" }, { "source": "tracker-as-desired-state-reconciliation", "target": "board-mediated-async-inter-agent-coordination", "type": "related" }, { "source": "tracker-as-desired-state-reconciliation", "target": "signal-driven-agent-activation", "type": "related" }, { "source": "tracker-as-desired-state-reconciliation", "target": "multi-platform-webhook-triggers", "type": "related" }, { "source": "tracker-as-desired-state-reconciliation", "target": "filesystem-based-agent-state", "type": "related" }, { "source": "unified-tool-gateway", "target": "progressive-tool-discovery", "type": "related" }, { "source": "unified-tool-gateway", "target": "tool-search-lazy-loading", "type": "related" }, { "source": "unified-tool-gateway", "target": "tool-capability-compartmentalization", "type": "related" } ] }