auto-delegation
Skill0 starsChanged 7 months agoMITView on GitHub
Use PROACTIVELY when user sends a free-text request (not a /command). Routes natural language to the right agent or plugin skill. NOT for Build-phase task assignment — see agent-detection for that.
In this repository17 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- agent-detectionSkill
- bookstackSkill
- delegatorSkill
- faceswapSkill
- git-standardsSkill
- headswapSkill
- health-dashboardSkill
- imageSkill
- motionSkill
- multi-terminalSkill
- project-docsSkill
- quality-gatesSkill
- safety-gateSkill
- upscaleSkill
- videoSkill
- web-researchSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 77
Before routing to a single agent, check if the request requires multiple agents:
- Very specific: line 101
When multi-step is detected, invoke the delegator skill for decomposition and wave-based dispatch.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Auto-Delegation
Route free-text user requests to agents and plugin skills.
Scope: Interactive user requests (natural language in chat).
Not for: Structured task assignment during Build phase (→ use agent-detection).
Decision Flow
User request
↓
Explicit /command? → Execute command
↓ no
Plugin skill matches? → Invoke skill
↓ no
CWE agent matches? → Delegate to agent
↓ no
Multi-step task? → Delegate to subagents (via delegator skill)
↓ no
Unclear? → Ask (max 2 questions)Override: Say "manual" or "no delegation" to disable.
Intent → Agent
| Intent | Agent | Keywords |
| Write/fix code | builder | implement, fix, build, create, code, feature, bug, refactor |
| Questions/discussion | ask | question, discuss, think about |
| Explain code/concepts | explainer | explain, how, what, why, understand |
| Testing/quality | quality | test, write tests, coverage, quality, validate, assert, metrics, flaky, gate |
| Security audit | security | security, audit, vulnerability, scan, gdpr, owasp, cve |
| Infrastructure | devops | deploy, docker, ci, cd, release, kubernetes, k8s, terraform |
| System design | architect | design, architecture, adr, api, schema |
| Research/docs | researcher | analyze, document, research, compare |
| Brainstorming | innovator | brainstorm, idea, ideas, what if, alternative, explore |
| Process improvement | guide | workflow, process, pattern, improve, optimize, optimization |
"test" Routing
"test" and related keywords always route to quality first. Quality decides next steps:
- Coverage analysis, metrics, review → quality handles directly
- Writing new tests (TDD, regression) → quality delegates to builder
- CI pipeline tests → quality delegates to devops
Intent → Plugin Skill
Plugin skills take priority over agent routing when matched.
| Keywords | Skill | Plugin |
| UI, frontend, component, page | frontend-design | frontend-design |
| simplify, cleanup, refactor | code-simplifier | code-simplifier |
| debug, investigate bug | systematic-debugging | superpowers |
| write plan, planning | writing-plans | superpowers |
| review code | requesting-code-review | superpowers |
| TDD, test first | test-driven-development | superpowers |
| update CLAUDE.md | claude-md-improver | claude-md-management |
| create plugin, hook, command | plugin-dev skills | plugin-dev |
| develop feature | /feature-dev | feature-dev |
| youtube/instagram/tiktok/media URL | transcript | cwe (auto via hook) |
Multi-step Detection
Before routing to a single agent, check if the request requires multiple agents:
Triggers (invoke delegator skill)
- Keywords match 2+ different agents (e.g., "build" + "test" + "document")
- Conjunctions across domains: "implement X and write tests and update docs"
- Completeness keywords: "full", "complete", "end-to-end", "komplett", "mit allem"
- Compound verbs: "implement, test, and deploy"
- Feature-level implicit: "Add user authentication" (code + tests + docs)
NOT Multi-step (stay with single agent)
- One verb, one domain: "Fix login bug" → builder
- Question across topics: "Explain auth and caching" → explainer
- Vague: "Look at the API" → ask (clarify first)
Examples
"Build auth with tests and docs" → delegator (builder + quality + researcher)
"Fix bug and add regression test" → delegator (builder + quality)
"Implement, test, and deploy" → delegator (builder + quality + devops)When multi-step is detected, invoke the delegator skill for decomposition and wave-based dispatch.
Context Injection (Automatic)
Standards are auto-loaded via .claude/rules/ with paths frontmatter.
The _index.yml maps standards to both file paths AND task keywords.
| Task Keywords | Rules Loaded |
| All files | .claude/rules/global-standards.md (always) |
| api, endpoint, REST | .claude/rules/api-standards.md |
| component, UI, frontend | .claude/rules/frontend-standards.md |
| migration, schema, database | .claude/rules/database-standards.md |
| docker, deploy, CI/CD | .claude/rules/devops-standards.md |
| test, coverage, TDD | .claude/rules/testing-standards.md |
| agent, skill, hook | .claude/rules/agent-standards.md |
Rules
- Never guess — ask when unclear (max 2 questions)
- Respect explicit commands — /cwe:builder, /feature-dev etc. bypass this
- Context isolation — agent result is summary, not full context
- Plugin skills first — check plugin skill match before agent match
Examples
"Fix login bug" → builder
"How does auth work?" → explainer
"Audit the API" → security
"Create release" → devops
"Write tests for auth" → quality (may delegate to builder)
"What if we used GraphQL" → innovator
"Look at this" → ASK: "Fix, explain, or analyze?"
"Build auth with tests and docs" → delegator (builder + quality + researcher)
"Fix bug and add regression test" → delegator (builder + quality)Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.