updating-coverage

name	updating-coverage
description	Refresh the README coverage badge by running coverage and rewriting the shields.io badge line.
user-invocable	true
allowed-tools	Read, Bash(pnpm run cover:*), Bash(pnpm run coverage:*), Bash(pnpm run test:cover:*), Bash(node:*), Bash(git:*)
model	claude-haiku-4-5
context	fork
metadata	{"internal":true}

updating-coverage

Runs the repo's coverage script and rewrites the README badge so the published number matches reality. Invoked directly via /update-coverage or as a phase of the updating umbrella.

When to use

• After landing a substantial change to test coverage (added a major feature with tests, removed a large untested module).
• Pre-release, to refresh the public badge.
• As part of updating umbrella flow when the repo declares a coverage script.

What it does NOT do

• Generate coverage from scratch. This skill consumes the output of the repo's existing coverage tooling (vitest / c8 / istanbul / node-test coverage). If no coverage script is declared in package.json, the skill reports that and exits.
• Compute coverage thresholds. The badge reflects what the tooling reports; tightening the threshold is a separate decision in the repo's vitest/c8 config.
• Modify nested READMEs. Only the repo-root README.md is rewritten. Nested READMEs under packages/* have their own badges and lifecycles.

Phases

#	Phase	Outcome
1	Discovery	Find the coverage script in package.json (cover / coverage / test:cover, in that preference).
2	Run	pnpm run <script>. Fail loudly if the run errors.
3	Rewrite	node scripts/fleet/gen/coverage-badge.mts — reads .cache/fleet/coverage/coverage-summary.json, rewrites the badge.
4	Commit	docs(readme): refresh coverage badge to N%. Direct-push per fleet norm.

The parse + rewrite math (read the summary, round the percent, pick the color bucket, edit the README) is owned by scripts/fleet/gen/coverage-badge.mts and its lib scripts/fleet/lib/coverage-badge.mts — the same owner the commit-time gate scripts/fleet/check/coverage-badge-is-current.mts reads. This skill never re-derives the number or the format in shell; if it did, the badge it wrote (e.g. two decimals, a hard-coded color) would be rejected by check --all. The skill is orchestration over those scripts; the judgment it keeps is surfacing a real coverage-run failure.

Phase 1: discovery

node -e "import('./scripts/fleet/lib/coverage-badge.mts').then(m => { const s = m.coverageScriptName(process.cwd()); if (!s) { process.exit(1) } console.log(s) })"

coverageScriptName returns the first of cover / coverage / test:cover declared in package.json, or exits non-zero when the repo tracks no coverage. That is not a failure mode — many fleet repos don't track coverage; the skill exits cleanly.

Phase 2: run

pnpm run <SCRIPT>

Use the standard pnpm runner so the repo's own env config (catalog versions, etc.) applies. A real coverage-run failure is surfaced, not swallowed — that's the judgment this skill keeps.

Phase 3: rewrite

node scripts/fleet/gen/coverage-badge.mts

This reads .cache/fleet/coverage/coverage-summary.json (the json-summary reporter's output) and rewrites the README badge in place: the percent is Math.round-ed to an integer and the color is the bucket badgeColor computes (red → brightgreen). Exit 0 = written or already current; exit 1 = no coverage data / no badge to fill. To see the before value first, run node scripts/fleet/gen/coverage-badge.mts --check — the dry-run the gate uses — and read its output.

The canonical badge line in README.md is the placeholder a seeded repo ships:

![Coverage](https://img.shields.io/badge/coverage-<PCT>%25-brightgreen)

The script fills <PCT> (and updates the color); the %25 is URL-encoded % and is left alone.

Phase 4: commit

git add README.md
git commit -m "docs(readme): refresh coverage badge to <N>%"
git push origin <default-branch>

Direct-push per the fleet's Commits & PRs → Push policy rule; fall back to PR if the remote rejects.

Output

When called via /update-coverage, emit a one-line summary of the integer percent before → after (read from the --check run). When no coverage script exists or the percentage is unchanged, exit silently.

Related

• .claude/skills/updating/SKILL.md: umbrella that calls this skill when applicable.
• .claude/skills/updating-security/SKILL.md: sibling under updating.
• template/README.md: canonical README skeleton ships the placeholder badge.