Verify

Check Done

No completion claims without fresh verification evidence. Period.

Before ANY claim that work is done, tests pass, a bug is fixed, or code is working — you MUST run the verification command, read the output, and confirm it matches the claim.

If the repo declares custom lifecycle hooks, read ../lifecycle-hooks.md and honor any phase-boundary hooks before and after verify.

The Gate

When reporting durable evidence, use the terse Verification report shape from artifact-templates.md.

For every positive claim about work state:

1. Identify the verification command (test suite, linter, build, curl, etc.)
2. Run it fresh — not from memory, not from a previous run
3. Read the actual output
4. Confirm the output supports the claim
5. Then make the claim

What Requires Evidence

| Claim | Required evidence | |---|---| | "Tests pass" | Test command output showing 0 failures | | "Linter clean" | Linter output showing 0 errors | | "Build succeeds" | Build command with exit code 0 | | "Bug is fixed" | Failing test now passes (red to green) | | "Feature works" | Test or manual verification output |

Proof Requirements

When a Work Contract includes proof_requirements, match verification evidence to that vocabulary before claiming done. A command log can satisfy command_gate, review output can satisfy review or fresh_eyes, CI status can satisfy ci_check, and a deck/screenshot can satisfy screenshot_show_me. Missing required proof means stop and report the missing proof or human interrupt; do not downgrade it to advisory in chat.

Red Flags

These words in your response without preceding evidence mean you're guessing:

• "should work", "probably works", "seems to be working"
• "I believe this fixes", "this should resolve"
• "tests should pass now"

Replace with: run the command, paste the output, state the fact.

After Subagent Work

Agent success reports are not evidence. After a subagent completes:

1. Check the VCS diff
2. Run verification independently
3. Then confirm

Visual Proof

If the user explicitly asks for a richer visual proof deck, invoke show-me. If verification would be clearer as screenshots, videos, diffs, command logs, or a browsable HTML report, suggest show-me and wait for a direct user request. Do not run show-me automatically; it is a manual/user-requested escalation in v1.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sandsower
• Source: sandsower/beislid
• License: MIT

Install and usage instructions live in the source repository linked above.