Run Verification

Select and execute this repository's verification matrix based on change scope, and output a structured verification report (executed / not executed / pre-existing failures / new failures). This is the executable form of AGENTS.md §6.

Source of truth: repository root AGENTS.md. If the matrix here drifts, AGENTS.md and the actual CI workflows win. Recipes: .claude/skills/references/test-command-recipes.md. Config guard: .claude/skills/references/hard-rules.md §2.

Usage

/run-verification [--scope backend|web|desktop|docs|workflow|ai-assets|auto]

--scope auto (default): derive the change scope from git diff --name-only <baseline>...; multiple scopes can apply at once.

Instructions

Step 1: Derive change scope

git diff --name-only origin/main...HEAD

Path hit	Scope
main.py, src/**, data_provider/**, api/**, bot/**, tests/**	backend
apps/dsa-web/**	web
apps/dsa-desktop/**, scripts/*desktop*	desktop
README.md, docs/**	docs
.github/**, scripts/**, docker/**	workflow
AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, .github/instructions/**, .claude/skills/**	ai-assets

Step 2: Execute per scope

backend (prefer the full gate; fall back to itemized commands):

./scripts/ci_gate.sh
# Minimum requirement:
python -m py_compile <changed_python_files>
python -m pytest -m "not network" <relevant test paths>

When the diff touches .env.example, src/core/config_registry*, Settings help, or documented env keys, also run:

python scripts/check_config_doc_consistency.py
python -m pytest tests/core/test_env_example_config_registry_guard.py -q

Never expand the unregistered-debt baseline / ceiling to green CI — register the key instead (hard-rules §2).

web:

cd apps/dsa-web
npm ci
npm run lint
npm run test
npm run build

If i18n is touched (src/i18n/**, src/locales/**, or any user-visible copy), additionally:

npm run test:i18n

web e2e (only when the change targets e2e specs or components covered by them — e.g. overlays, auth flows; not part of the default web gate):

cd apps/dsa-web
npm run test:e2e-security-preflight
npx playwright install --with-deps chromium
npm run test:smoke                 # playwright.config.ts self-hosts backend + web via webServer
# Single spec: node e2e/run-playwright-tests.mjs <spec-file>

The backend webServer requires Python deps installed (python -m pip install --build-constraint build-constraints.txt -r .github/requirements-ci.txt). e2e failures follow the same Step 3 baseline attribution as unit tests — many e2e specs are already red on main.

desktop (web first, then desktop):

cd apps/dsa-web && npm run build
cd ../dsa-desktop && npm install && npm run build

When platform limits prevent full verification, list verified vs unverified platforms explicitly in the report.

docs: tests are not enforced, but every command, config entry, filename, and workflow name mentioned in the docs must be checked against the actual repository, and the report must state the check results.

workflow: run the local validation closest to the change (e.g. bash -n, a docker build smoke, python scripts/<changed>.py --help). Anything not locally verifiable (e.g. Actions trigger conditions) is listed as unverified with the risk stated.

ai-assets:

python scripts/check_ai_assets.py

Step 3: Red-test attribution (the key discipline)

On any failure, compare against the main baseline before drawing conclusions:

git stash            # or use a clean worktree: git worktree add /tmp/wt-main origin/main
# Run the same command on the baseline and record its failure set
git stash pop

Worktree tip: symlink node_modules into the worktree (ln -s <repo>/apps/dsa-web/node_modules /tmp/wt-main/apps/dsa-web/node_modules) instead of reinstalling; remove the worktree afterwards (git worktree remove --force /tmp/wt-main).

• Red on baseline and untouched by the change → "pre-existing failure": listed in the report, does not block this task.
• Green on baseline, red after the change → "new failure": must be fixed; never bypassed.
• When attribution is unclear, treat it as a new failure.

Step 4: Output the report

### Verification Report

- Scope: <derived scopes>
- Executed: <command> → <result> (one line each)
- Not executed: <item> — <reason>
- Pre-existing failures (also red on origin/main): <list or none>
- New failures: <must be empty to pass>

Division of labor vs existing skills

• This skill owns only "how to run verification and attribute results".
• test-change is the issue #890 entry name and pre-PR evidence checklist; it delegates execution here.
• Development flow: develop-feature / fix-issue. Review: review-pr / analyze-pr.

Allowed Auto-Actions (No Confirmation Needed)

• All verification commands above (all non-destructive local operations)
• Creating/cleaning temporary worktrees for baseline comparison

Actions Requiring Confirmation

• None (this skill contains no actions that change remote or commit state)