impacted-specs

name	impacted-specs
description	Use when you want to find which RSpec specs to run for locally changed code, including uncommitted changes. Refreshes the coverage map produced by the test_impact gem's GitHub Actions collect workflow, then runs `test-impact plan` to narrow the run down to the impacted specs. Triggers on requests like "run only the tests related to my changes", "which specs does this diff affect", "figure out the impacted specs", or "what should I run for the current changes".
license	MIT

impacted-specs

This skill runs inside a project that consumes the test_impact gem. It determines which specs to run from local changes, including uncommitted ones.

Prerequisites:

• gh CLI is authenticated
• The consuming project has the test_impact gem in its Gemfile
• That project runs a collect workflow in GitHub Actions that uploads the coverage map as an artifact named test-impact-map

Steps

1. Refresh the base

Check base in .test_impact.yml (defaults to origin/main) and fetch the corresponding remote branch:

git fetch origin main

Why: if the local origin/main is stale, the map's commit_sha fails the history reachability check and the plan degrades to a full run (all).

2. Download the map (every time, auto-discovered)

Do not judge freshness even if a local map exists — always download the latest artifact and overwrite. Do not depend on a specific workflow name:

run_id=$(gh api 'repos/{owner}/{repo}/actions/artifacts?name=test-impact-map&per_page=1' \
  --jq '.artifacts[] | select(.expired == false) | .workflow_run.id' | head -1)
rm -f .test_impact/map.json.gz
gh run download "$run_id" -n test-impact-map -D .test_impact

Notes:

• Test expired with an explicit == false comparison (// treats false as falsy and falls back to the right-hand side, so an enabled // true style check misjudges it).
• gh run download fails when a file of the same name already exists in the destination, so remove it first with rm -f.
• If run_id comes back empty (no artifact found), report to the user that the collect workflow is not set up and stop here.

3. Run plan

bundle exec test-impact plan --format json --include-uncommitted > /tmp/plan.json

--format json always exits 0. stdout is a single line of JSON: {"mode":"all|partial|none","spec_files":[...],"reason":null|"..."}. Diagnostics (mode / reason / spec_files count, etc.) go to stderr.

4. Interpret the result

• mode: "partial" → run only the selected specs:

  bundle exec rspec $(jq -r '.spec_files[]' /tmp/plan.json)
  

• mode: "none" → report to the user that there are no specs to run.
• mode: "all" → do not run the full suite on your own. Report that a full run is required along with reason, and ask the user how to proceed. Common reason values:
  • uncovered file changed: <path> — a new or uncollected file was changed
  • no map available or backend invalid — the map is missing or unusable

Caveats

• If .test_impact/ is not in the consuming project's .gitignore, the map itself is treated as an untracked new file and triggers a full run.
• In a shallow clone, merge-base computation fails and the plan degrades to a full run.
• --include-uncommitted includes staged / unstaged / untracked changes in the diff, but never modifies repository state (read-only).