improve-code-coverage

name: improve-code-coverage description: >- Check whether branch changes meet Glific Codecov thresholds using only local git and ExCoveralls (no push, no CODECOV_TOKEN), add focused tests for gaps, and re-run until the local gate passes. Use when the user asks to improve coverage, fix Codecov failures, or satisfy codecov.yml on the current branch. disable-model-invocation: true

Improve Code Coverage

Iterative workflow: generate an ExCoveralls report, validate project and patch gates with a local script (mirrors codecov.yml), add tests for uncovered changed lines, and repeat until the gate is green.

No git push, no Codecov upload, and no network required.

Prerequisites

• On a feature branch (or master only if the user explicitly wants that).
• Postgres available locally (mix coveralls.json runs test_full and needs the test DB).
• Python 3 and jq (optional, for ad-hoc inspection) on the machine.
• Local git only: diff uses git merge-base HEAD <base> and git diff — no git fetch or origin push required.
• Committed branch changes are compared to local master (override with CODECOV_BASE_BRANCH). Uncommitted lib/*.ex changes are included only when using --include-worktree (see reference.md).

Read reference.md for commands and troubleshooting.

Project requirements (codecov.yml)

CI still uploads to Codecov after push; this skill approximates the same gates offline.

Hard rules

• Never weaken codecov.yml or disable checks to get green.
• Never add tests that only hit lines without asserting behavior.
• Never git push or call Codecov upload APIs as part of this skill unless the user explicitly asks for a remote check afterward.
• Stop and ask the user after 5 loops without green, or if fixes need large unrelated test work.
• Do not commit unless the user asked; run the loop on the current tree (commits ± optional worktree).

Progress checklist

- [ ] Loop N — mix coveralls.json
- [ ] Loop N — scripts/check_codecov_local.py (exit 0)
- [ ] Loop N — Add/fix tests for reported uncovered patch lines
- [ ] Final — local gate green (project + patch)

Improve loop (repeat until green)

Step 1 — Generate coverage artifact

mix coveralls.json

Step 2 — Local Codecov gate

From repo root:

python3 .claude/skills/improve-code-coverage/scripts/check_codecov_local.py

Include unstaged/staged lib/*.ex changes in the patch (optional):

python3 .claude/skills/improve-code-coverage/scripts/check_codecov_local.py --include-worktree

The script prints uncovered path:line for patch failures.

Step 3 — Add tests

1. Target lines/files from Step 2 patch FAIL output.
2. Add or extend tests under test/ mirroring lib/:
   • Glific.DataCase / ConnCase as appropriate
   • Organization fixtures / Repo.put_organization_id/1
   • Tesla.Mock for external HTTP
   • Real behavior assertions
3. Run targeted tests, then return to Step 1.

Step 4 — Loop exit

Repeat until Step 2 exits 0. Report branch, HEAD SHA, loops, tests added, and final percentages.

Optional: stricter project check vs local base

To mirror Codecov’s “no drop vs base branch” project rule:

1. On local master (or base branch): mix coveralls.json then
   python3 .../check_codecov_local.py --save-baseline (see reference).
2. On the feature branch: run the gate with --baseline-file .coverage-baseline.json.

Skip unless the user wants base-relative project comparison.

When to ask the user

• Five loops without green
• Patch failures only in files outside the intended PR scope
• Coverage gaps requiring broad unrelated suites
• Whether to use --include-worktree for uncommitted work

Related skills

• make-branch-ready-for-review — push + CI/Codecov after local gates pass
• fix-flaky-tests — failures during mix coveralls.json

Additional resources

• CLAUDE.md, codecov.yml, reference.md