Repo Standardizer

GitHub Repo Standardizer

Detect a repository's current state, then polish its surface: issue forms, PR template, label taxonomy, CI, CODEOWNERS, rulesets, docs. Works on repo metadata and config files only — code logic is never touched. Idempotent — safe to re-run; fills gaps and reconciles drift without duplicating or clobbering.

When to use

• User says "standardize / tidy up / professionalize this repo"
• A new repo was just created and needs templates, labels, CI, and rules from day one
• A repo looks bare: no templates, no labels, no CI, no branch protection

Preflight (mandatory, in order)

Requires: gh CLI (authenticated), git, jq, python3 — verify they exist before starting (which gh git jq python3).

1. Authenticate — check gh login first

gh auth status 2>/dev/null || echo "NOT_LOGGED_IN"

• Logged in → continue; print gh api user -q .login so the user knows which account will act.
• Not logged in → STOP. Tell the user (do not guess):
  1. Run gh auth login (web/device flow), or
  2. Export a token: export GH_TOKEN=ghp_xxx — needs repo, workflow (for CI files) and, for org repos, admin:org (or admin:repo_hook / org membership admin) scopes.
  • Never paste tokens into chat, logs, or files. If the user pastes a token in chat, advise them to revoke it and re-issue.
  • If gh auth login is impossible in this environment (headless), suggest gh auth login --with-token reading from a file the user created.

Verify the acting account can write to the target:

# user repos: no extra check needed beyond token scopes
# org repos: must be a member/admin of the org
gh api "orgs/ORG/memberships/$(gh api user -q .login)" -q .role 2>/dev/null || echo "NO_ORG_ACCESS"

• admin/member → OK. 404 → stop and ask the user to add the account to the org first.

2. Detect repository type

gh repo view OWNER/REPO --json name,owner,visibility,defaultBranchRef,isArchived,isFork \
  -q '{name:.name, ownerType:.owner.type, visibility:.visibility, defaultBranch:.defaultBranchRef.name, archived:.isArchived, fork:.isFork}'

If the repo was not explicitly named by the user, confirm before touching an org or private repository.

3. Audit current state

gh label list --repo OWNER/REPO --limit 200
gh api repos/OWNER/REPO/contents/.github -q '.[].path' 2>/dev/null || echo "no .github dir"
gh api repos/OWNER/REPO/contents/.github/workflows -q '.[].name' 2>/dev/null || echo "no workflows"
gh api repos/OWNER/REPO/rulesets -q '.[] | {name:.name, enforcement:.enforcement}' 2>/dev/null || echo "no rulesets"
gh api repos/OWNER/REPO/branches -q '.[].name' 2>/dev/null
for f in README.md CONTRIBUTING.md SECURITY.md LICENSE .gitignore; do
  gh api "repos/OWNER/REPO/contents/$f" -q .name 2>/dev/null || echo "missing: $f"
done

4. Detect test framework (for CI module)

Check for these signals (first match wins):

gh api repos/OWNER/REPO/contents/package.json -q .name 2>/dev/null   # node → templates/ci-node.yml
gh api repos/OWNER/REPO/contents/pyproject.toml -q .name 2>/dev/null # python → templates/ci-python.yml
gh api repos/OWNER/REPO/contents/go.mod -q .name 2>/dev/null         # go → templates/ci-go.yml
gh api repos/OWNER/REPO/contents/Cargo.toml -q .name 2>/dev/null     # rust → templates/ci-rust.yml

No signal → propose the generic CI (or ask the user whether CI is wanted at all).

5. Ask about language (before planning)

• Ask the user which language CONTRIBUTING.md and the PR template should be written in (default: English, or the project's primary language for local projects). Translate the templates accordingly when generating — never assume a language.
• (README languages are asked separately in Module F.)

6. Ask about automation (decides the Governance labels)

Ask the user whether the repo has any automation bots or AI writers running on it — e.g. dependabot, a Stale bot, ClawSweeper, or an AI coding agent that auto-files PRs / auto-fixes issues.

• Why ask: r:* / clawsweeper:* / triage:* / close:* labels are signal labels, not categories — a bot reads them and acts (auto-close, auto-lock, auto-fix, auto-merge). Without running automation those labels are dead weight, so the answer decides whether the Governance dimension is added at all (Module A, Step 2).
• Yes → plan the Governance / auto-close rules dimension (add only the r:* rules matching the project's real reject criteria).
• No → skip that dimension entirely — never add r:* / bot labels to a repo with no automation.

Workflow

1. Preflight (above). If auth or access fails, stop with a clear message.
2. Dry-run plan — show the user a concise table of what will be created/updated/skipped. Get confirmation for: rulesets, branch deletion/protection changes, org-level changes, and anything destructive.
3. Apply modules (each idempotent; run in this order).
4. Verify — re-query and print an applied / skipped / failed checklist.

Module A — Labels (design first, then idempotent upsert)

Step 1 — Profile the project (adjust the taxonomy, never copy blindly):

• Rating labels → design a themed tier system for THIS project (never copy an existing repo's set verbatim):
  • Reference example — OpenClaw's official repo uses themed tiers with emoji icons and a low→high color gradient (EXAMPLE ONLY, do not copy): rating: 🧂 unranked krab → rating: 🦪 silver shellfish → rating: 🦐 gold shrimp → rating: 🦀 challenger crab → rating: 🐚 platinum hermit → rating: 🦞 diamond lobster
  • International project → universal grades also work (grade: S/A/B/C/D or ★–★★★★★)
  • Always redesign: pick a theme that fits the project (animals, gems, ranks, stars…) and the audience's language. Do not reuse any existing repo's rating labels as-is.
• Teams: if the repo has an explicit division of labor (CODEOWNERS, CONTRIBUTING, a team list in docs) → add one team: * label per group (e.g. team: frontend, team: algorithm). No team list → skip.
• Project type (library / app / coursework / org-infra) → decide which dimensions below are needed (dependencies, security, docs, …).
• Language: write every label name and description in the language chosen in Preflight step 5 (default English; local project → its primary language). Never assume — translate every label name and description into the chosen language (emoji glyphs stay as-is).
  • Exception — bot labels: r:* / clawsweeper:* / triage:* / close:* names are matched literally by automation code (almost always English). Keep those names in the bot's language — a translated name breaks the bot. Only their description may use the user's language.

Step 2 — Compose categories. Baseline lives in templates/labels.json (plain names, no emoji); extend or trim per the profile in Step 1.

Emoji policy — rating tiers are the ONLY mandatory-emoji labels. Every rating:* / issue-rating:* tier MUST carry an emoji with a clear low→high gradient (e.g. rating: 🦞 diamond lobster). All other labels: emoji is the agent's call — add icons where they aid scanning, omit them where they clutter. Consistency rule: within one dimension, either ALL labels carry an emoji or NONE do — never a mixed half-emoji dimension (e.g. don't ship 🐛 bug next to a plain enhancement).

Every dimension is opt-in except Type. Add a dimension only if the repo actually needs it; skip it otherwise. The menu below mirrors OpenClaw's official label taxonomy (the richest open reference) — cover every dimension that applies, but never force one the repo doesn't use.

Priority uses P0–P3 (OpenClaw's convention — P0 = emergency). Examples in the tables below show the optional emoji style — apply them all or none per dimension (baseline ships plain P0–P3):

Dimension menu (write each label in the user's chosen language):

Governance rules (r: / bot labels) — gated by Preflight step 6. Only consider this dimension if the user answered "yes" to automation bots / AI writers; skip it entirely otherwise. Auto-close and bot-state labels are signal labels, not categories — a bot (or Actions workflow) reads them and acts (auto-close, auto-lock, auto-fix, auto-merge). They are useless without the matching automation, so skip them unless a governance bot actually runs on the repo. If one does, add only the r:* rules matching the project's real off-topic / reject criteria (e.g. r: spam, r: support, r: no-ci-pr) plus the bot's own state labels (clawsweeper:*, triage:*, close:*). Never copy OpenClaw's set verbatim — its rules encode OpenClaw's specific product boundaries.

Impact dimension — the security "blast radius" judgement (generalized from OpenClaw; trim to the repo's actual failure modes):

Step 3 — Color rules (mandatory):

• Diverse palette: colors must be rich and varied — the whole label set should look like a palette, not a monochrome block. Even within one category, spread the hues (e.g. priority labels: red / orange / yellow / green, or four clearly different hues).
• Semantic hints (not hard mappings): ready to merge / merged / done → greens (never gray or red); wontfix → gray; in progress → blue. Everything else: pick colors that look good together and match the label's meaning loosely — but prefer variety over strict one-meaning-one-color.
• Emoji policy: rating tiers are emoji-mandatory — every rating:* / issue-rating:* label needs a clear low→high emoji gradient (e.g. rating: 🦞 diamond lobster). Everywhere else, emoji is the agent's call: use icons where they aid scanning, omit them where they clutter. Consistency: within one dimension, all labels carry an emoji or none do — never a mixed half-emoji dimension. If used, the emoji must match the label's meaning, never decorative-only.
• Rating labels need docs: when rating labels are added, also add LABELS.md (Step 4) describing each label's meaning and the explicit low→high order, so the ranking is unambiguous.
• Neighboring labels must be distinguishable. Forbidden: all-one-color, adjacent duplicates, or colors that contradict the label content.

Step 4 — Rating-label docs (only if rating labels exist). Generate LABELS.md from templates/LABELS.md (or extend an existing docs file): list every rating label with its meaning and the explicit low→high order, plus the rest of the taxonomy. Commit and push it together with the labels.

Step 5 — Idempotent upsert. GitHub has no PUT /labels/{name} endpoint. Upsert = check existence (GET /labels/{name}), then POST /labels (create) or PATCH /labels/{name} (update). Works for both map-form and array-form labels.json:

R="repos/OWNER/REPO"
jq -c 'if type == "array" then .[] else to_entries[] | {name: .key} + .value end' templates/labels.json | while read -r l; do
  name=$(echo "$l" | jq -r .name); color=$(echo "$l" | jq -r .color); desc=$(echo "$l" | jq -r .description)
  enc=$(python3 -c "import urllib.parse,sys;print(urllib.parse.quote(sys.argv[1]))" "$name")
  if gh api "$R/labels/$enc" >/dev/null 2>&1; then
    gh api -X PATCH "$R/labels/$enc" -f name="$name" -f color="$color" -f description="$desc" --silent && echo "label updated: $name"
  else
    gh api -X POST "$R/labels" -f name="$name" -f color="$color" -f description="$desc" --silent && echo "label created: $name"
  fi
done

• URL-encode label names (spaces, slashes).
• To reconcile drift (deleted manual labels), show the diff and ask before removing labels that are already in use.