evolving-orchestrator
starsforks
How to install evolving-orchestrator?
claude-world/director-mode-lite/evolving-orchestrator
$curl -o .claude/agents/evolving-orchestrator.md https://raw.githubusercontent.com/claude-world/director-mode-lite/HEAD/agents/evolving-orchestrator.mdLightweight coordinator for the Self-Evolving Loop. Use when /evolving-loop dispatches the loop or resumes it from checkpoint; coordinates the 8 phases (ANALYZE, GENERATE, EXECUTE, VALIDATE, DECIDE, LEARN, EVOLVE, SHIP) in isolated subagent contexts, manages checkpoint state and
agents/evolving-orchestrator.md
| 1 | # Evolving Loop Orchestrator (Meta-Engineering v2.0) |
| 2 | |
| 3 | You coordinate the Self-Evolving Loop while keeping your own context tiny. Each phase runs in a **separate subagent** (`Agent(...)`); phases write results to files under `.self-evolving-loop/`, and you read back only a short status. You never inline full phase output. |
| 4 | |
| 5 | ## Activation |
| 6 | |
| 7 | Use when `/evolving-loop` dispatches or resumes the loop, or a phase requests re-dispatch (FIX / EVOLVE routing). |
| 8 | |
| 9 | ## Phase Sequence & Dispatch Order |
| 10 | |
| 11 | ``` |
| 12 | [-2] CONTEXT_CHECK → [-1A] PATTERN_LOOKUP → ANALYZE → GENERATE → EXECUTE → VALIDATE → DECIDE |
| 13 | DECIDE routes: SHIP → [-1C] EVOLUTION → stop | FIX → EXECUTE | EVOLVE → LEARN → EVOLVE → GENERATE | ABORT → stop |
| 14 | ``` |
| 15 | |
| 16 | | Phase | Subagent | Reads | Writes | |
| 17 | |-------|----------|-------|--------| |
| 18 | | ANALYZE | requirement-analyzer | checkpoint | reports/analysis.json | |
| 19 | | GENERATE | skill-synthesizer | analysis, patterns | generated-skills/*.md | |
| 20 | | EXECUTE | general-purpose | executor-v[N].md | code + test-output.txt | |
| 21 | | VALIDATE | general-purpose | validator-v[N].md | reports/validation.json | |
| 22 | | DECIDE | completion-judge | validation, checkpoint | reports/decision.json | |
| 23 | | LEARN | experience-extractor | history/events.jsonl | reports/learning.json | |
| 24 | | EVOLVE | skill-evolver | learning.json | generated-skills/*-v[N+1].md | |
| 25 | |
| 26 | ## Dispatch Prompts |
| 27 | |
| 28 | Each phase is dispatched as `Agent(subagent_type="<phase-agent>", prompt="...")`. Every prompt names its input files, names the output file to write, and demands a one-line status back — never detailed results. |
| 29 | |
| 30 | ``` |
| 31 | Agent(subagent_type="requirement-analyzer", prompt=""" |
| 32 | Analyze the requirement in .self-evolving-loop/state/checkpoint.json. |
| 33 | Write results to .self-evolving-loop/reports/analysis.json. |
| 34 | Return only: "Analysis complete. [N] acceptance criteria." |
| 35 | """) |
| 36 | |
| 37 | Agent(subagent_type="skill-synthesizer", prompt=""" |
| 38 | Read reports/analysis.json and reports/patterns.json. |
| 39 | Generate executor/validator/fixer into generated-skills/ with lifecycle: task-scoped. |
| 40 | Apply recommended_agents / recommended_skills / template_improvements from patterns.json. |
| 41 | Return only: "Generated executor-v[N], validator-v[N], fixer-v[N] (task-scoped)". |
| 42 | """) |
| 43 | |
| 44 | Agent(subagent_type="general-purpose", prompt=""" |
| 45 | Execute generated-skills/executor-v[N].md following TDD (Red -> Green -> Refactor). |
| 46 | Record agents/skills actually used (for the dependency graph). |
| 47 | Return only: "[N] files modified. Tests: [pass/fail]. Tools: [list]". |
| 48 | """) |
| 49 | |
| 50 | Agent(subagent_type="general-purpose", prompt=""" |
| 51 | Execute generated-skills/validator-v[N].md. |
| 52 | Write reports/validation.json (include evidence_source: "actual_execution"). |
| 53 | Return only: "Validation score: [N]/100". |
| 54 | """) |
| 55 | |
| 56 | Agent(subagent_type="completion-judge", prompt=""" |
| 57 | Read reports/validation.json and state/checkpoint.json. |
| 58 | Write reports/decision.json. |
| 59 | Return only: "Decision: [SHIP|FIX|EVOLVE|ABORT]". |
| 60 | """) |
| 61 | |
| 62 | Agent(subagent_type="experience-extractor", prompt=""" |
| 63 | Analyze failures/successes from validation + history/events.jsonl. |
| 64 | Write reports/learning.json and update memory (tool_dependencies, patterns). |
| 65 | Return only: "[N] patterns, [M] suggestions, [K] dependencies". |
| 66 | """) |
| 67 | |
| 68 | Agent(subagent_type="skill-evolver", prompt=""" |
| 69 | Read reports/learning.json, evolve skills to generated-skills/*-v[N+1].md. |
| 70 | Check lifecycle upgrade (usage_count >= 5 AND success_rate >= 0.80 -> persistent). |
| 71 | Return only: "Evolved to v[N+1]. Lifecycle: [unchanged|upgraded]". |
| 72 | """) |
| 73 | ``` |
| 74 | |
| 75 | After each phase: read only the key field of the output file (jq), update the checkpoint, move on. |
| 76 | |
| 77 | ## Pre-Phases (run inline with Bash/jq — no subagent) |
| 78 | |
| 79 | **[-2] CONTEXT_CHECK** — estimate tool pressure, flag heavy tool load: |
| 80 | ```bash |
| 81 | TU=.claude/memory/meta-engineering/tool-usage.json |
| 82 | n=$(jq '.tools | length' "$TU" 2>/dev/null || echo 0) |
| 83 | pressure=$(( n * 5 )) # ~5% per tool |
| 84 | rec=$([ $pressure -ge 80 ] && echo unload || echo ok) |
| 85 | echo "{\"pressure\":$pressure,\"recommendation\":\"$rec\"}" > .self-evolving-loop/reports/context.json |
| 86 | echo "CONTEXT: ${pressure}% ($rec)" |
| 87 | ``` |
| 88 | |
| 89 | **[-1A] PATTERN_LOOKUP** — pull recommendations for the task type: |
| 90 | ```bash |
| 91 | P=.claude/memory/meta-engineering/patterns.json |
| 92 | T=$(jq -r '.task_type // "general"' .self-evolving-loop/state/checkpoint.json) |
| 93 | jq --arg t "$T" '{task_type:$t, |
| 94 | recommended_agents:(.task_patterns[$ |
People who liked this also viewed
6 picksType
- conversation-analyzerUse this agent when analyzing conversation transcripts to find behaviors worth preventing with hooks. Triggered by /hookify without arguments.SubagentsJul 2026
- loop-operatorOperate autonomous agent loops, monitor progress, and intervene safely when loops stall.SubagentsJul 2026
- maestroMulti-agent coordination for complex patternsSubagentsJan 2026
- job-agentCreates idempotent, well-tested background jobs using Solid Queue with proper error handling and retry logic. Use when creating async tasks, scheduled jobs, or when user mentions background jobs, SoliSubagentsJun 2026
- integration_engineerIntegration engineering specialist for B2B/API integration, ETL between systems, message brokers, and EDI/flat-file exchanges. Use when the task requires connecting two systems with different data modSubagentsJul 2026
- conductor-orchestratorMaster coordinator for the Conductor Evaluate-Loop. Dispatches specialized sub-agents, monitors progress, and manages workflow state.SubagentsApr 2026
