Coding guidelines generator
Goal
Create nested AGENTS.md files (not repo root) that define:
- architecture preferences + boundaries (high level)
- formatting/lint/testing commands (runnable)
- a small verifiable config block the verifier skill can execute
Minimal questions to ask (don’t skip)
- Where are the module roots? (Confirm the folders that should each get an
AGENTS.md; avoid repo root unless explicitly requested.) - Architecture style: layered / hex / clean (pick one) and any hard boundaries (e.g.,
domain/must not importinfra/). - OK to add tooling when missing? (default: yes; Spotless for Java/Kotlin)
- Default commands: format / lint / test for each module (changed-files-first where practical).
Workflow (monorepo-aware)
- Scan for candidate modules and languages.
- Use scripts/scan_modules.py to produce a quick inventory.
- Propose the nested
AGENTS.mdplacement(s) and get confirmation.
- If the only detected module is repo root, suggest a subdir (e.g., src/, apps/<name>/, packages/<name>/) or ask the user where the code “starts”.
- For each confirmed module root:
- Create/update <module>/AGENTS.md using references/agents-template.md. - Fill the codex-guidelines JSON block (schema: references/verifiable-block.md) with runnable commands.
- Ensure formatting + linting exist (prefer existing; otherwise add best-practice defaults).
- JVM (Gradle/Maven): add/configure Spotless (see references/spotless.md). - Others: pick a minimal, common default and wire commands into AGENTS.md (see references/language-defaults.md).
Output
- One
AGENTS.mdper module root (nested scope). - Minimal tool config changes only where tooling is missing (format/lint first).