القائمة
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.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Change Model skill template. Transforms code changes into structured change reports (WHY/WHAT/HOW/VALIDATION four-layer architecture), with call-chain checking, archival storage, and automatic Git analysis. Triggered when generating change reports, DiffLogs, Release Notes, or archiving change records.
Decomposition-driven skill. Defines the principles and methods for task decomposition, model tier routing, and sub-task dispatch. Trigger when decomposing complex tasks, determining model tiers, or dispatching sub-agents.
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.
AI Agent Skills creation methodology and execution pipeline. Triggered when creating skill systems for projects, generating project-specific skills, understanding skill templates, model delegation, or validating skill accuracy. This is a META skill — it creates other skills as its output.
| 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 |
| composes | [] |
| composed_by | [{"meta":"skill-builder-guide"}] |
| context_budget | {"l1_metadata":100,"l2_body":1500,"l3_references":3000} |
| version | 1.1.0 |
| status | active |
| review_by | "2026-07-28T00:00:00.000Z" |
| trust_level | internal |
| requires_network | false |
| requires_file_write | false |
| compatibility | universal |
| allowed_tools | Agent Read Bash |
| evolution | {"usage_count":0,"last_corrections":[],"stale_markers":[]} |
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.
Deep reference: See references/upgrade-strategy.md for upgrade strategy details.
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.
See references/upgrade-strategy.md.
L1 — Sonnet / atomic tier: Rule interpretation and orchestration, requires reasoning. This skill is a template reference. See delegation for actual execution.