Prompt file imported from dotnetpower/fdai (.github/prompts/verify.prompt.md). Fill in {{ARGS}} before use. Copyright stays with the author.

/verify - run focused FDAI verification

Run the narrowest executable check that can falsify the current change and report its summary.

Steps

1. Confirm the current working directory is the repo root (git rev-parse --show-toplevel). If not, cd there.
2. If a Python venv exists at .venv/, activate it so ruff and pytest are on PATH: source .venv/bin/activate.
3. If the user supplied a pytest path, run focused verification: bash scripts/verify.sh --full {{ARGS}}.
4. Otherwise select the smallest test file, node id, typecheck, linter, or structural checker for the task-owned paths. Do not substitute verify.sh --fast, verify.sh --all, or an unscoped package/repository suite.
5. If the user explicitly identifies a merge or release boundary, run make validation-all exactly once. Do not wait for or rerun the same validation.
6. Print the relevant command summary. If any gate failed:
   • Name the failing gate.
   • Point at the individual scripts/check-*.sh or the offending pytest path so the caller can rerun in isolation.
7. Do NOT commit anything from this prompt. Verification only.

Guardrails

• Never bypass a gate (no --no-verify, no gate skipping).
• Never edit the gate scripts to make them pass; treat a failure as a real finding.
• Do not touch untracked or WIP files while verifying.
• Do not validate one session by selecting unrelated files from a shared dirty worktree. Prefer a separate Git worktree for concurrent sessions.