Test Change

P0 testing skill for this repository (issue #890). Selects test layers from the change surface, runs the documented commands, and records what passed, skipped, and why. Does not replace CI.

Source of truth: repository root AGENTS.md §6. Executable matrix and red-test attribution live in /run-verification. Shared recipes: .claude/skills/references/test-command-recipes.md. Hard rules: .claude/skills/references/hard-rules.md.

Usage

/test-change [--scope backend|web|desktop|docs|workflow|ai-assets|auto]

Instructions

Step 1: Map change → layers

From git diff --name-only origin/main...HEAD (or the task boundary):

Path hit	Layers to run
main.py, src/**, data_provider/**, api/**, bot/**, tests/**	backend offline gate / path pytest; py_compile minimum
Config / .env.example / config_registry*	config registry guard + doc consistency (hard-rules §2)
apps/dsa-web/**	web lint, unit, build; i18n tests if copy/locales touched; smoke e2e only when UI journeys change
apps/dsa-desktop/**, desktop scripts	web build then desktop build
AGENTS.md, .claude/skills/**, copilot mirrors	python scripts/check_ai_assets.py
Docs only	no code tests required; verify every command/filename mentioned

Step 2: Execute via run-verification

Run /run-verification with the derived scope(s). Prefer full gates when risk is high; never invent alternate install or test entry points when a recipe already exists.

Config-touching changes must include:

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

Step 3: Evidence bar

Minimum evidence before "done" or PR:

• Commands actually run + results
• Pre-existing failures on origin/main (non-blocking) vs new failures (blocking)
• Not executed items with reason
• UI/report changes: screenshot policy is PR attachment only (never commit binary evidence)
• Reachability claims (user-visible mounts): support with hard-rules §3 grep evidence, not file existence alone

Step 4: Forbidden

• Weakening tests to greenwash
• Expanding config-registry debt baselines to green CI
• Claiming verification while new failures remain
• Auto git push / merge

Division of labor

• run-verification: owns how to run the matrix and attribute red tests; this skill is the issue #890 entry name and the pre-PR evidence checklist that calls it.
• develop-feature Step 4 / fix-issue Step 4: invoke this skill or run-verification equivalently.
• regression-scout: lists likely surfaces; this skill runs the checks.

Allowed Auto-Actions

• All non-destructive verification commands in the matrix and recipes

Never

• Auto push, merge, or secret logging
• Claiming CI replacement by agent self-attestation