improve-agents-md Skill

improve-agents-md

Compatibility Note

This pi wrapper preserves the original setup/update workflow while expressing any confirmation points as explicit manual checkpoints in the current chat instead of assuming a runtime interactive-question primitive.

Generate and manage AGENTS.md (rendered as CLAUDE.md) using a single base template with dynamic tooling values from tooling.yml.

Architecture

assets/
  AGENTS.md.template     # Single base template with {{VAR}} placeholders
  tooling.yml            # Per-type tooling values (package_manager, linter, etc.)
  extras/pep723.md       # Optional section injected for Python types
  templates/shared/      # Shared templates (.gitignore)
tools/
  bootstrap.py           # Entrypoint: setup, update, validate
  template_engine.py     # Reads tooling.yml, renders template, injects extras
  project_type.py        # Auto-detect project type from file indicators
  preserve_custom.py     # Backup + CLAUDE.md migration detection

Modes

setup

Generate AGENTS.md for a project:

Detect project type (or accept --type override)
Back up existing files if --force
Load tooling values from tooling.yml for detected type
Render AGENTS.md.template with tooling substitution
Inject extras (e.g., PEP 723 section for Python types)
Write rendered AGENTS.md to project root
Create CLAUDE.md and GEMINI.md as symlinks to AGENTS.md

update

Re-render AGENTS.md with latest template:

Re-detect project type (or accept --type override)
Preserve customizations if --force
Re-render AGENTS.md with current tooling values

validate

Check AGENTS.md content:

Verify AGENTS.md exists
Check for unresolved {{VAR}} placeholders
Verify required sections present
Check CLAUDE.md is a symlink to AGENTS.md
Warn if legacy PROJECT_AGENTS.md is present
Check template assets exist

Flags

| Flag | Default | Description | |------|---------|-------------| | --dry-run | false | Print actions without executing | | --force | false | Force overwrite (backup + re-render) | | --type | auto | Override project type detection |

Dynamic Rendering

Template variables resolved from tooling.yml:

| Variable | Description | |----------|-------------| | {{PACKAGE_MANAGER}} | Package manager command (uv, pnpm, cargo, etc.) | | {{TYPE_CHECKER}} | Type checking command | | {{LINTER}} | Linter command | | {{AFTER_EDIT}} | Post-edit validation command chain | | {{STYLE}} | Language-specific style conventions |

Null values in tooling.yml become (customizable) in output.

Extras Injection

Types with extras: [pep723] in tooling.yml get the PEP 723 section injected after ## Style & Conventions.

Questionnaire

Questions only asked when tooling.yml has null values for a type:

python-pip: Type checker (pyright/mypy/skip) + Linter (ruff/flake8/pylint)
generic: Project type selection (maps to known type or stays customizable)
Ambiguous detection: Primary project type confirmation

Supported Project Types

| Type ID | Detection | |---------|-----------| | typescript | package.json with typescript/@types | | ts-bun | bun.lockb present | | python-uv | uv.lock or [tool.uv] in pyproject.toml | | python-poetry | [tool.poetry] in pyproject.toml | | python-pip | requirements.txt or setup.py | | rust | Cargo.toml | | generic | fallback |

Customization Preservation

If CLAUDE.md exists as a regular file, its content is backed up before migration
AGENTS.md is the primary file; CLAUDE.md and GEMINI.md are symlinks
Timestamped backup created before any destructive operation
Legacy PROJECT_AGENTS.md is no longer created; validate warns if present

Agent Skills: improve-agents-md

Install this agent skill to your local

Skill Files