docs-governance

Before you start

1. Check if docs/sessions/$(date +%Y-%m-%d).md exists
2. If not, create it with a session header: ## Session — HH:MM + **Objective:** one-line summary
3. Log your work continuously as you go — do not wait until the end

Docs Governance

Every piece of non-code documentation in this repo must live in a path that the Obsidian vault-sync agent mirrors. That way anyone browsing the vault, or the in-app "Vault Docs" tab, sees the full picture. Stray .md files are a bug.

The rule

Documentation lives in one of these locations (authoritative list — sync script mirrors each of them):

Anything else that is "documentation" (not code, not generated, not vendored) is stray and must be moved into one of those paths.

Not in scope

• src/** — inline code comments are not docs
• plans/**, templates/** — separate systems
• node_modules/**, .next/**, build artifacts
• Scoped AGENTS.md files (e.g. server/AGENTS.md) — legitimate, keep in place
• Tool-specific configs like .github/copilot-instructions.md
• tests/e2e-*.md test runbooks — judgment call: move to docs/testing/ unless the test runner reads them

Detection

Run the auditor:

scripts/docs/audit-stray-docs.sh

It globs every *.md in the repo, subtracts the in-scope and out-of-scope lists above, and prints what's left. Exit code 0 if clean, 1 if any stray docs remain.

Prompt-to-fix flow

For each stray file the auditor reports:

1. Read the first ~40 lines to understand what it is.
2. Propose a target under the allowlist using these filename heuristics (apply the first that matches):

3. Ask the user before moving — one confirmation per file, or a batched "move all to <target>" if several share the same target.
4. git mv the file (preserves history). Update any internal links that referenced the old path — grep the repo for the old basename first.
5. Wait for next vault-sync cycle (up to 5 min) — the old vault copy clears via rsync --delete.

Do not auto-move without confirmation. Stray files often encode history that's worth glancing at — and sometimes the right answer is "delete it," not "move it."

Skills inventory

The same principle applies to skills, commands, and agents. To see the current inventory classified as standard (present in templates/claude-project-template/.claude/) vs project-specific (only in this repo), regenerate the index:

scripts/docs/generate-skills-index.sh

This writes docs/meta/SKILLS.md. The file is committed and auto-syncs to the vault, so it's browsable from the Vault Docs tab. Regenerate it whenever you add, remove, or rename a skill / command / agent.

Classification rules the generator applies

• standard — filename exists in the template at the same relative path.
• project-specific — exists in this repo, not in the template.
• drifted — exists in both but content differs (informational only; drift is often intentional).
• missing — exists in template, not in this repo (flag for review).

When to run this skill

• After creating any .md outside the allowlist (the file watchers in editors don't enforce it).
• Before cutting a release, as part of the pre-commit walk.
• After adding / renaming / removing a skill, command, or agent.
• Periodically, as part of the /session-log housekeeping (~monthly).

Template propagation

This skill is part of the standard project template. When it changes, also update:

• templates/claude-project-template/.claude/skills/docs-governance/SKILL.md

…so /new-repo inherits the current version.