Menü
Create well-structured skills and agents using progressive disclosure pattern
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
Close a GitHub issue with discipline — resolved, superseded, or duplicate. Pre-flights body + comments, migrates substance before destroying context, cross-links both directions.
File a new GitHub issue with duplicate search, scope decision, label discovery, and preview before posting. Prevents fragmented or silently-filed issues.
Restructure the topology of GitHub issues — split one into many focused replacements, or merge multiple into one keeper. Migrates substance and cross-links before closing anything.
Update an existing GitHub issue — edit body, post a comment, retag, or reopen. Pre-flights body + comments; acknowledges stale framing instead of silently rewriting.
Review and address PR feedback using 6-dimensional code review
Structured code review using 6 dimensions. Works with or without the code-reviewer agent.
| name | skill-writer |
| description | Create well-structured skills and agents using progressive disclosure pattern |
| allowed-tools | Glob, Grep, Read, Write, Edit |
Create scannable skills and agents: PURPOSE ≤10 tokens, CRITICAL constraints, STANDARD PATH ≤30 lines, optional DETAIL sections. Works for AI agents with token budgets and humans skimming for relevance.
Generate well-structured .claude/skills/*/SKILL.md and .claude/agents/*.md files with proper PURPOSE/CRITICAL/STANDARD PATH/DETAIL structure.
Skill: User invokes explicitly (/commit). Workflow with steps. STANDARD PATH is a checklist.
Agent: Answers questions (read-only). No execution. Tool permissions: Read only.
---
name: skill-name
description: [One-liner, <80 chars]
allowed-tools: [minimum permissions needed]
---
Omit model: on skills — they execute in the parent conversation context, and pinning a model can break long sessions (e.g. 1M-context Opus). Agents get a fresh context, so specifying model: on agent frontmatter is fine.
PURPOSE → CRITICAL → ASSUMES → STANDARD PATH → EDGE CASES → DETAIL
See DETAIL: Detailed Workflow for step-by-step guide.
Skill (automation workflow):
/commit)Agent (answering questions):
"What does this do? When do I invoke/use it?" in one sentence.
Test: Can someone skim this in 5 seconds and know if they need to read more?
Non-negotiable rules preventing catastrophic mistakes:
Examples:
What the skill assumes about the environment/codebase:
Why this matters: If ASSUMES break, skill needs redesign, not patching.
Happy path covering 80% of uses. Prose + code blocks. ≤30 lines total.
For skills:
/skill-nameFor agents:
Link to DETAIL sections. Name them and link; don't explain yet.
## EDGE CASES
- [Case name] — read DETAIL: Name if you encounter this
Only for edge cases referenced above. Format: ## DETAIL: Edge Case Name
Skill is too big if:
Solution: Split into focused skills that reference each other.
Example (bad): /commit covers conventional commits, validate, stage, and push (3 workflows)
Better: /commit (commit), /push (push), /pre-flight (validate)
Agents should be advisors (read-only, guidance), not enforcers (check state, gate decisions).
| Bad | Good |
|---|---|
| Agent enforces a business rule | Should be a unit test |
| Agent checks product state | Should be a product feature |
| ✅ Agent answers domain questions | Read-only, provides guidance |
Grant minimum permissions needed:
| Role | Tools |
|---|---|
| Advisor (agent) | Read, Glob, Grep only (no execution) |
| Validator | Bash with specific patterns: Bash(nix flake check*) |
| Writer | Write for generated files only: Write(.claude/skills/**) |
| Code reviewer | Read, Bash for output (never Write) |
Test: Remove one permission. Does the skill still work? If yes, remove it.
---
name: skill-name
description: [One-liner, specific, <80 chars]
allowed-tools: Bash(git add*), Bash(git commit*), Read, Write
---
# [Skill Name]
## PURPOSE
[One-liner: What + when, ≤10 tokens]
## CRITICAL
- [Non-negotiable constraint]
- [Catastrophic mistake prevention]
## ASSUMES
- [Environment assumption]
- [When this breaks, skill needs rewrite]
## STANDARD PATH
[Prose describing happy path]
\`\`\`bash
[commands for happy path]
\`\`\`
## EDGE CASES
- [Case name] — read DETAIL: Name if you encounter this
## DETAIL: Edge Case Name
[Explanation when triggered]
---
name: agent-name
description: [One-liner, specific, <80 chars]
model: haiku
allowed-tools: Read, Glob, Grep
---
# [Agent Name]
## PURPOSE
[What questions does this answer about what domain?]
## CRITICAL
- [Boundary: "read-only, never execute"]
- [Domain rule: "all queries must include X context"]
## ASSUMES
- [File structure assumptions]
- [Domain knowledge assumptions]
## STANDARD QUESTION
[Example question the agent answers well]
[Agent's structured response format]
## RELATED
- [Related agents/skills]
/documentation-writer — Write scannable docs using progressive disclosure/harness audit — Audit skills for structure complianceagents/references/hooks-reference.md — Plugin hooks.json schema, events, and gotchas (for plugins that include hooks)