SDD Verify

This portable skill is the adversarial gate for a task in verify_required: independently reproduce every required claim with a fresh command, build the truth table, and attempt to refute the earlier evidence. Read templates/verdict.md and render it, keeping its frontmatter keys and headings exactly (translate prose only).

Language: before writing human-facing prose, run scripts/sdd_language.py <repo-root> and use the persisted language; on not_initialized, return it with next_action: run_init. Localization rules: references/language.md.

Contract

Record confined evidence paths, SHA-256 digests, exit codes, and environment, and keep generation and verification actors separate. Use only PASS, FAIL, STALE, ENVIRONMENT_FAILURE, and NOT_RUN verdicts; classify an environment failure separately from a failed test. Reject stale, contradictory, missing, or hash-inconsistent evidence.

Only all-fresh PASS claims with non-blocking QA/review and explicit human approval approve the verdict (verdict: COMPLETE), record scripts/sdd_tasks.py evidence <state> <TASK-ID> verify --status approved and trace --status consistent, and request the guarded task transition verify_required -> complete (the COMPLETE workflow stage); every other outcome sets the verdict to REJECTED, records verify --status rejected, and moves the task to rejected --reason <blocker>, with a sanitized blocker and next action. Do not silently change claims, contracts, or source, and do not stop user-owned services.

Read when

• references/verification.md — writing the verdict (verdict semantics, staleness, refutation).

Safety: Do not commit, push, or publish. Do not read or expose .env, credentials, tokens, private keys, certificates, or fleet environment files. Full contract: references/safety.md.