agentleFS
fjrevoredofjrevoredo/mini-diarium/.agents/skills/ci-gate-diagnosis/SKILL.md

ci-gate-diagnosis

Skill321 starsChanged 7 days agoMITView on GitHub

Diagnose a failing SonarCloud, Codecov, or Rust CRAP gate on a Mini Diarium PR, and mirror the Codecov patch gate locally before pushing. Use when CI reports coverage, duplication, or CRAP failures, including missing inputs or tool errors. Triggers: SonarCloud, quality gate, new_duplicated_lines_density, new_coverage, Codecov, patch coverage, coverage:diff, lcov, cargo-llvm-cov, CRAP, cargo-crap, crap-delta, crap-baseline.

  • Installs packages

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the failing check's evidence before changing code or thresholds.

    line 8
  2. Distinctive:

    When the "SonarCloud Code Analysis" check fails on a PR, the PR comment gives only a summary.

    line 12
  3. Distinctive:

    Run the local mirror:

    line 43
  4. Very specific:

    Read the CRAP policy gate (whole-function, repository-wide) step.

    line 70

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

CI gate diagnosis (SonarCloud + Codecov + CRAP)

Read the failing check's evidence before changing code or thresholds. SonarCloud has a public API; GitHub artifacts and logs require repository access.

SonarCloud quality gate — read the API, don't guess

When the "SonarCloud Code Analysis" check fails on a PR, the PR comment gives only a summary. To find which files are responsible, use the public API directly — no login required:

# Which condition failed and by how much
curl -s "https://sonarcloud.io/api/qualitygates/project_status?projectKey=fjrevoredo_mini-diarium&pullRequest=<PR>" | jq .

# Per-file breakdown (replace metric key as needed: new_duplicated_lines_density, new_coverage, etc.)
curl -s "https://sonarcloud.io/api/measures/component_tree?component=fjrevoredo_mini-diarium&pullRequest=<PR>&metricKeys=new_duplicated_lines_density,new_duplicated_lines&strategy=leaves&ps=50" | jq '.components[] | select(.measures[].value != "0.0") | {name: .name, measures: .measures}'

Common failures and their usual causes:

  • new_duplicated_lines_density > 3%: copy-pasted test helpers or fixture objects — extract to a shared constant/function in the same file.
  • new_coverage < threshold: new logic in a file that generatePdfFromElement-style functions (html2canvas/jsPDF) can't be tested in JSDOM — mock the module boundary instead.
  • new_security_rating worse than A: query the issue API for the actual rule and file; a project-status summary does not identify the finding:

    curl -s "https://sonarcloud.io/api/issues/search?componentKeys=fjrevoredo_mini-diarium&pullRequest=<PR>&types=VULNERABILITY&resolved=false&ps=100"

    Check git log origin/master..HEAD and the affected file's history before changing it. A PR can inherit earlier unmerged work; agree on its scope before editing that work or dismissing a finding. For javascript:S4036, do not hide the warning: select a trusted absolute executable instead of inherited PATH, test failure without fallback, and document any explicit nonstandard-installation choice. This protects executable selection, not a compromised host installation.

Note: website/**/*.html is excluded from duplication detection — every new blog post would otherwise fail the gate.

Codecov patch check — mirror it locally before pushing

CI uploads coverage/lcov.info (frontend) and src-tauri/lcov.info (backend) to Codecov, which enforces patch ≥ 80% (new/changed lines) and project: auto (no total regression) per codecov.yml.

The Vitest thresholds in vitest.config.ts are a coarse frontend-only global floor and do NOT catch patch/project failures — you can pass locally and still fail Codecov.

Run the local mirror:

cmd.exe /c bun run coverage:diff       # gate existing lcov files
cmd.exe /c bun run coverage:check      # generate coverage (--generate), then gate
cmd.exe /c bun run coverage:self-test  # parser self-test

scripts/check-diff-coverage.mjs consumes the same lcov files plus git diff origin/master, fails below 80%, and lists every uncovered new line as file:line. This mirrors the patch check (the most common CI failure); the project total-regression check needs a base-branch coverage baseline and is not replicated locally.

The gate also runs as step 9 of bun run pre-commit (via --working-tree, so it checks not-yet-committed changes against origin/master); that run generates both lcov files by running the frontend/backend tests with coverage.

Generating the lcov files by hand:

  • Frontend: bun run test:coverage
  • Backend: requires cargo-llvm-cov + cargo-nextest (cargo install cargo-llvm-cov cargo-nextest --locked), then from src-tauri/: cargo llvm-cov nextest --workspace --lcov --output-path lcov.info (--workspace so the mini-diarium-core and mini-diarium-crypto crates are covered; lcov still lands at src-tauri/lcov.info.)

Flags for coverage:diff: --generate (run both), --base <ref>, --fail-under <pct>, --no-fail, --frontend/--backend.

See CI Best Practices → Coverage Gating.

Rust CRAP gate — use the delta, not a second analyzer pass

The blocking comparison runs only in Linux CI. It checks whole functions across the repository, not only changed lines: New functions must score ≤30; existing functions must not increase by more than epsilon 0.01. Unchanged legacy debt passes. Local comparisons in bun run pre-commit are advisory on every OS, and cannot replace a Linux verdict.

  1. Read the CRAP policy gate (whole-function, repository-wide) step. Exit 1 means policy violations; stdout and annotations identify the same offenders. Download the crap-delta artifact, then run node scripts/check-crap-gate.mjs --report <downloaded-crap-delta.json> to interpret that exact input. Do not compute alternative metrics or add another --format github pass.
  2. For exit 2, diagnose inputs/tools first: successful same-run backend coverage, non-empty src-tauri/lcov.info, exact cargo-crap 0.6.1, a valid committed baseline, successful --for-comparison normalization, and a valid pinned delta-v2 report. A missing artifact after analyzer failure is expected; an absent report is never a pass.
  3. Download crap-report for producer provenance and scope diagnostics. Both Codecov upload attempts occur before CRAP; report and delta uploads run even after gate failure. Check their actual step results rather than assuming uploads succeeded.
gh run download <run-id> --name crap-delta --dir <evidence-directory>
gh run download <run-id> --name crap-report --dir <evidence-directory>
node scripts/check-crap-gate.mjs --report <evidence-directory>/crap-delta.json
node scripts/check-crap-gate.mjs --self-test
node scripts/normalize-crap-baseline.mjs --self-test

An all-0% table can mean path mismatch. Compare LCOV SF: paths, the recorded producer root, matched-file diagnostics, and analyzer scope. The committed baseline is relative; comparisons need an ignored root-aligned crap-baseline-input.json made with --for-comparison. Direct relative-versus-absolute comparison can mark repeated names as New. Preserve --missing pessimistic, --exclude "**/tests.rs" --exclude "**/test_support.rs", and the baseline's pinned configuration. skip is for diagnosis only; do not use it to pass CI.

Do not raise the baseline in a feature change. Baseline updates require a separate reviewed commit from probe-free master Linux CI, with exact production provenance. Reviewed improvements and identity updates are permitted; unexplained increases are not. See CI Best Practices → Rust CRAP Gate for the full policy and commands.

Limits apply even when the gate passes: closure branches, including with_unlocked_db(|db| ...), do not increase scored complexity; macros are not expanded; cfg-dead spans in matched files can show 100% coverage. Do not treat a passing score as proof of command-layer test coverage.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.