Menü
Delegation rules skill template. Used to guide creation of project-specific delegation rules. Defines which tasks must be delegated to Haiku, delegation format, and sub-agent output standards.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
AI Agent Skills creation methodology. Triggered when creating skill systems for projects, generating project-specific skills, understanding skill templates, model tier delegation, or validating skill accuracy. META skill — creates other skills as output. Deploy: copy this directory to .claude/skills/skill-builder/.
Call-chain tracing skill template. Guides creation of project-specific call-chain skills for tracing API data flow, verifying type matching, and checking endpoint correctness.
Code map skill template. Used to guide creation of project-specific code map skills. Triggered when asking about file locations, directory structure, component navigation, or route mapping. L0 — Must be executed by Haiku, pure file path lookup.
Development Standards skill template. Guides creation of project-specific development standards skills. Triggers on questions about tech stack, API conventions, code standards, coding style, or layered architecture.
Development workflow skill template. Guides creation of project-specific workflow skills. Triggered when asking about development steps, phase order, or checklists for common tasks.
Little Notebook — records user dissatisfaction and satisfaction moments during agent interaction. Use when user expresses anger, frustration, abusive language, or strong satisfaction. L0 — delegate to Haiku for recording and cleanup.
| name | example-delegation |
| description | Delegation rules skill template. Used to guide creation of project-specific delegation rules. Defines which tasks must be delegated to Haiku, delegation format, and sub-agent output standards. |
| model_tier | L1 |
| skill_tier | atomic |
| version | 1.2.0 |
| status | active |
知人者智，自知者明。 ——《道德经》 把对的任务交给对的模型，不是省 token，是尊重每一层的能力边界。
Position: Atomic tier / L1 execution tier — a reference template for other projects to create project-level delegation rules. Composed by skill-builder-guide (meta).
Full implementation: See delegation (planning tier) for this project's actual delegation skill.
Main model must not execute L0 tasks. All L0 tasks must be delegated via Agent(model: "haiku").
| Category | Example Tasks | Target Skill |
|---|---|---|
| File lookup | Find file location, directory structure | code-map |
| Info lookup | Check version number, API routes | dev |
| Command execution | Deploy, package, service start/stop | scripts |
| Static tracing | Draw dataflow diagram, verify API list | code-map / dev |
| Mechanical edit | Rename variable, update version number | (direct) |
Agent(
description: "3-5 word task description",
model: "haiku",
prompt: """
【Task】What specifically to do
【Files】List of paths to read
【Output】Conclusion/Basis/Uncertainty format
"""
)
Conclusion: One sentence answering the assigned goal
Basis: Specific evidence, observations, reasoning path
Uncertainty: Risks, missing information, failure modes (write "None" if none)
Main model does only three things after receiving: extract conclusions → identify conflicts → decide.
| Trigger | Threshold | Action |
|---|---|---|
| User repeatedly unsatisfied | Same task ≥2 correction rounds failed | Package context, submit to top-tier model |
| Work rework | Same code repeatedly modified, problem not converging | Stop modifying, re-analyze root cause |
Upgrade info package: Original requirements + attempted solutions and failure reasons + current blockers + excluded assumptions.
After completing this skill, recommend the next skill based on output characteristics:
| Condition | Recommend |
|---|---|
| Delegation revealed L0 subtasks | Any project skill as needed |
L1 — Sonnet / atomic tier: Rule interpretation and orchestration, requires reasoning. This skill is a template reference. See delegation for actual execution.