一键导入
skill-creator
Use when asked to create or update skills or extend agent capabilities in this configuration.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when asked to create or update skills or extend agent capabilities in this configuration.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Systematic analysis and reasoning workflows. Use when performing audits, investigation, requirements analysis, risk analysis, scenario/edge-case enumeration, or design analysis that demands evidence-based, comprehensive coverage.
Code quality and safety standards. Activates when editing or reviewing code. Use for quality priorities, safety checks, and code correctness verification.
Use when performing coding tasks (implement features, fix bugs, refactor code). Orchestrates language detection, mode switching (normal/autopilot/full-auto), quality enforcement, and verification.
Break a problem into atomic, testable subproblems. Use when facing complex tasks, unclear requirements, or planning implementation strategy.
Start structured feature implementation workflow. Use when implementing new features, adding functionality, or building complete user-facing capabilities.
Use when starting work that needs isolated branches or parallel feature development using git worktrees.
| name | skill-creator |
| description | Use when asked to create or update skills or extend agent capabilities in this configuration. |
| allowed-tools | ["write_file","read_file"] |
| metadata | {"short-description":"Create new Codex skills systematically"} |
Guide for creating new Codex skills with the same rigor as existing ones.
The context window is a public good. Only include what the model does not already know. Prefer concise guidance and move heavy references into separate files.
Match instruction specificity to task fragility:
Use a three-layer structure:
Before creating or editing a skill:
Never summarize guidance documents. Always instruct to read the full documents, allowing all guidance to apply.
Bad (summarizing):
Per quality.md:
- Correctness first
- Then safety
Good (explicit read):
**Read these references:**
- `~/.config/agent/domain/coding/quality.md` - Complete quality priorities
All files in chezmoi source: ~/.local/share/chezmoi/home/
Skill Files (Parity Required):
dot_claude/skills/<name>/SKILL.mddot_codex/skills/<name>/SKILL.mdApply with chezmoi:
chezmoi apply ~/.codex
Template:
---
name: skill-name
description: Use when <trigger conditions only; no workflow summary>
allowed-tools: ["shell", "apply_patch", "read_file", "write_file"]
metadata:
short-description: Brief one-liner
---
# Skill Title
**Read these references:**
- `references/file1.md` - Purpose/summary
- `~/.config/agent/path/to/file2.md` - Purpose/summary
## Instructions
<Skill-specific guidance>
Apply all guidance from references listed above.
## Arguments
Target: ${ARGUMENTS}
Key points:
description field triggers auto-activation - be specific about triggersallowed-tools is JSON array format: ["shell", "read_file"]metadata.short-description for UI display${ARGUMENTS} for argument substitution (NOT $ARGUMENTS)@ includes - use explicit "Read these references" insteadReference Formats:
references/file.md~/.config/agent/path/to/file.mdFor skills needing domain-specific guidance:
# In deployed directory (NOT chezmoi source!)
cd ~/.codex/skills/skill-name
ln -s ~/.config/agent/domain/path references
# If also using Claude Code, mirror the same symlink:
cd ~/.claude/skills/skill-name
ln -s ~/.config/agent/domain/path references
Example: Go skills symlink to ~/.config/agent/domain/coding/go
Which modules from ~/.config/agent/ apply?
core/ - Universal behaviors (methodology, task-management, etc.)domain/ - Domain-specific (coding, testing, review, architecture)workflows/ - Task workflows (feature implementation, etc.)In chezmoi source: ~/.local/share/chezmoi/home/dot_codex/skills/<name>/SKILL.md
chezmoi apply ~/.codex
In deployed directory: ~/.codex/skills/<name>/
ln -s ~/.config/agent/domain/path references
$skill-name <arguments>All paths relative to ~/.config/agent/:
Core:
core/behavior.md - Critical thinking, evidence-based reasoningcore/communication.md - Tone, style, concisenesscore/methodology.md - 5-phase problem-solvingcore/efficiency.md - Parallelization, throughputcore/task-management.md - TODO disciplinecore/principles.md - Engineering principlesArchitecture:
domain/architecture/decomposition.md - Problem breakdowndomain/architecture/parallelization.md - Concurrent executionCoding (General):
domain/coding/workflow.md - Universal coding workflowdomain/coding/quality.md - Quality prioritiesdomain/coding/safety.md - Runtime + security safetyCoding (Go):
domain/coding/go/style.md - Go code styledomain/coding/go/idioms.md - Go conventionsdomain/coding/go/concurrency.md - Go thread safetydomain/coding/go/testing.md - Go test practicesdomain/coding/go/tooling.md - Go commandsdomain/coding/go/organization.md - Go code structuredomain/coding/go/performance.md - Go optimizationTesting:
domain/testing/unit.md - Unit test strategiesdomain/testing/coverage.md - Coverage requirementsReview:
domain/review/process.md - Review workflowdomain/review/priorities.md - Review focus areasWorkflows:
workflows/feature.md - Feature implementation❌ Summarizing guidance:
From quality.md, the priorities are: correctness, safety...
❌ Selective quoting:
Key rules: Rule A, Rule B
❌ Using @ includes (doesn't work in Codex):
Apply all guidance from:
- @~/.config/agent/file.md
✅ Explicit read instructions:
**Read these references:**
- `references/file.md` - Complete guidance on topic
- `~/.config/agent/path/to/file.md` - Full details
See existing Codex skills:
~/.codex/skills/go-development/SKILL.md - With references symlink~/.codex/skills/go-test/SKILL.md - Command-style skill~/.codex/skills/problem-solving/SKILL.md - With multiple references~/.codex/skills/feature/SKILL.md - Workflow orchestration