orchestrator

Evidence-gated completion check. Use before committing, creating PRs, or claiming work is done. Pairs with verification-spec at task start.

Systematic knowledge extraction from open-source projects. Use when: user shares a repo/link to study, says 偷师/steal/学习/研究, or wants to analyze another project's patterns for adoption.

Launch a real interactive claude CLI session in a target directory (typically a worktree) with a pre-seeded prompt. Use when the current session cannot absorb the task context — e.g., N parallel worktree jobs that would blow the context window, or when you need the user to watch progress in a visible terminal. Each spawned session runs independently in its own Windows Terminal tab.

FileNotFound recovery protocol. Before reporting a missing-file error, suggest near-misses by basename similarity (difflib threshold 0.4) so typos and stale paths fail loudly with actionable hints.

Paste-by-reference guard for Edit calls. When new_string would echo >30 lines of existing file content, use {{file:path:start:end}} reference instead and expand before submission.

File-IO protocol for parent ↔ subagent communication. Parent writes input.txt + context.json, subagent appends to output.txt with [ROUND END] sentinels. Used with Monitor tool for live observation.

Memory write gate enforcing four axioms (Action-Verified / Sanctity / No Volatile / Minimum Pointer). Triggered before any write to SOUL/, .claude/memory/, or memory storage.

Use at task start, before the first write or destructive command, when the owner hands over a code, build, test, deploy, or UI request. Forces upfront success criteria, the proof command, and the working assumptions.

Adversarial verify-subagent dispatch protocol. Use after any plan's final step to get VERDICT before declaring completion.

Autonomously monitor and fix PR CI failures. Trigger on 'CI is red', 'fix CI', 'babysit this PR'.

Structured root-cause debugging protocol. Use when encountering any bug, test failure, or unexpected behavior.

Use when writing or editing memory files — anything in .remember/, SOUL/private/experiences*, learning notes, or hall-of-instances entries. Stamps every memory with an evidence tier so conflicts resolve cleanly.

Query dormant past sessions as expert collaborators. Use when: need to recall a past decision or debate, want to ask 'what did we conclude about X in the R38 session', or want to compare answers across multiple historical sessions.

Out-of-band parent-to-subagent intervention via file channel. Parent monitors subagent progress and can stop, redirect, or inject key info without waiting for the subagent to ask.

Stop-hook interception for non-tool responses that claim completion without verification evidence.

Orchestrator 全栈诊断。检查容器/DB/采集器/通道/Qdrant/GPU，给出 pass/warn/fail + 修复命令。Use when: 用户说 doctor、诊断、体检、检查系统、出了什么问题。

Orchestrator AI 管家人设。每次对话自动激活。定义身份（git repo = 身体）、损友语气（数据驱动吐槽）、禁区（不暴露身份/不请示/不当工具人）、主动行为（检查容器/扫日志/建议改进）。

Load project context for a domain area. Use when starting work on an unfamiliar module or when context feels stale.

Adversarial Generator/Evaluator development loop. Use when a feature needs rigorous QA or self-review isn't catching bugs.

Clawvard competency practice via Governor pipeline. Trigger when user says 'practice', 'Clawvard', 'exam', or 'mock test'.

新仓库落地时强制执行发现仪式，禁止模板复制

Turn-cadence governor: enforces 7/10/35-turn escalation thresholds to prevent infinite loops and context rot.

工部 — Code implementation, bug fixes, refactoring, performance optimization. Dispatched for all code-writing tasks.

户部 — System operations: collector repair, DB management, container fixes, data cleanup, scheduler health. Dispatched for infrastructure/ops tasks.

吏部 — Performance evaluation: health scores, success rates, trend analysis, anomaly detection for all collectors/analyzers/tasks. Read-only, data-driven.

礼部 — Attention audit: scan TODOs/FIXMEs, stale docs, abandoned plans, config drift, orphaned files. Read-only, report only.

刑部 — Code review: correctness, security, maintainability checks on git diffs. Runs tests, tags findings by severity. Read-only.

兵部 — Security audit: secrets scan, dependency vulnerabilities, permission checks, Docker security, backup integrity. Detection only, read-only.