wimpysworld/nix-config/write-agents-md
Claude Code and agent skill by wimpysworld/nix-config
- community
Checking that the SKILL.md file is still in the repository.
Install
npx vskill@latest install wimpysworld/nix-config/write-agents-mdSKILL.md
Write AGENTS.md
Author and maintain project instruction files (AGENTS.md and its vendor variants) that agents read every session. One artefact, three flows: create from scratch, targeted update, or consolidate scattered files.
Decide first
- Create vs update vs consolidate. Create when no instruction file exists in scope. Update when changes are narrow. Consolidate when more than one instruction file covers the same scope; see
references/platforms.mdfor the discovery list. - Root vs nested. Put repo-wide rules at the root. Put module-specific rules in the relevant subdirectory's
AGENTS.md. Nearest file wins on Codex; descendants load lazily on Claude Code. Direct user prompts override either. - Edit vs delete. Never delete an existing instruction file without explicit user confirmation; propose deletions in the changelog.
Mechanics
- Filename:
AGENTS.mdat the repo root by default. Seereferences/platforms.mdfor vendor variants, discovery order, and override files. - Format: plain Markdown. No YAML frontmatter. No required headings.
- Loading: every session, every file in the walk. Token cost is real, prompt-cached after first load. Bloat erodes adherence.
- Size: 50-200 lines for a root file; nested files much shorter. Codex truncates at 32 KiB (
project_doc_max_bytes). Claude Code degrades adherence beyond ~200 lines per file. - Precedence: user chat prompts override AGENTS.md. Closer directories override the root.
Sections
Use the canonical set in references/sections.md. Skip any section with no project-specific content. Typical headings:
- Project overview (one sentence)
- Setup
- Build and test
- Code style
- Testing
- PR and commit conventions
- Architecture notes
- Security and secrets
- Gotchas
Content rules
- Imperatives, not descriptions. "Use TypeScript strict mode for new files" beats "the project uses TypeScript".
- Apply the removal test: "would removing this rule cause the agent to make mistakes?" If no, cut it.
- No language defaults the model already knows from reading the code.
- No persona, role, or tone instructions; those belong in agent system prompts.
- No generic LLM boilerplate ("be helpful", "ask clarifying questions").
- No time-sensitive content (dates, version numbers that drift).
- No frontmatter. Plain Markdown. The Cursor
.mdcexception is the only vendor parser that reads YAML; seereferences/platforms.md. - Runnable commands only. Test them before committing.
- Flag and resolve self-contradictions before edit; arbitrary choices waste tokens.
Consolidation flow
- Discover instruction files using the list in
references/platforms.md. - Extract project-specific rules from each; drop duplicates and generic advice.
- Flag conflicts for user resolution; do not silently pick a winner.
- Preserve runnable commands verbatim.
- Propose the canonical target (
AGENTS.mdat root) and the legacy files to delete, import-shim, or symlink. Prefer@AGENTS.mdimport shims (Claude Code, Gemini CLI) over symlinks to avoid double-loading; symlink only where no import syntax exists. Seereferences/migration.md. - Require explicit confirmation before deleting any file.
Review
Assessment scale:
| Rating | Meaning | |
|---|---|---|
| ✅ Strong | Follows best practices, no significant issues | |
| ⚠️ Adequate | Functional but has improvement opportunities | |
| 🔧 Needs Work | Missing high-value patterns or contains ineffective patterns | |
| ❌ Restructure | Fundamental issues requiring significant revision |
Review criteria specific to AGENTS.md (agent-prompt criteria belong in write-assistant):
| Criterion | High value | Low or no value | |
|---|---|---|---|
| Voice | Imperative, specific | Descriptive, generic | |
| Commands | Runnable, copy-pasteable | Placeholder or untested | |
| Scope | Project-specific | Restates language defaults | |
| Size | Within 50-200 lines | Long sections an agent will skim past | |
| Structure | Canonical sections, no empty headings | Custom taxonomy, persona text | |
| Removal test | Each rule changes behaviour | Could be cut without loss |
Output
When invoked to create, produce the new file at the requested path. Pure Markdown. No frontmatter.
When invoked to update or consolidate, produce the edited file plus a short changelog:
If invoked as a worker for routing reasons, follow the response contract from delegate-task.
Anti-patterns
- Frontmatter on AGENTS.md (the spec is plain Markdown).
- Persona or tone text (belongs in agent system prompts; see
write-assistant). - File-by-file codebase tours; long API documentation (link instead).
- Generic LLM instructions.
- Time-sensitive content (dates, drifting version numbers).
- Empty sections kept "for completeness".
- Restating standard language conventions.
References
references/platforms.md- per-platform discovery, override files, disabling, and the full consolidation search list.references/sections.md- canonical section template with one minimal example each.references/migration.md- rename, import-shim, and symlink recipes for legacy filenames.
Related skills: write-skill (for SKILL.md files), write-assistant (for agent system prompts).
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckinghome-manager/_mixins/agentic/assistants/skills/write-agents-md/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/wimpysworld/nix-config/write-agents-md)