verify

Verification discipline that requires evidence before acting on claims. Every significant assertion must be verified against fresh output before being declared true. Prevents false confidence and rework from unverified assumptions.

<Use_When>

• User says "verify", "check if", "confirm", "evidence"
• A claim is made about code behavior, test results, or system state
• Before marking a story complete in ralph
• Before closing a bug as resolved </Use_When>

<Do_Not_Use_When>

• The claim is trivially verifiable (file exists, command output is visible)
• User wants analysis, not verification (use research instead)
• The claim is about future behavior (predictions can't be verified) </Do_Not_Use_When>

<Why_This_Exists> "Looks good" is not verification. Many failed implementations passed because someone declared them complete without running fresh evidence. Verify requires that every claim be tested before it's accepted. </Why_This_Exists>

<Execution_Policy>

• Every claim must cite verifiable evidence from a fresh run
• "Looks correct", "should work", "the code seems fine" are not evidence
• Run the actual command; read the actual output; cite the actual line
• Distinguish "passes locally" from "verified in CI" </Execution_Policy>

Phase 1: Parse the Claim

Extract the specific assertion to verify:

Claim: "<the exact statement being verified>"

Phase 2: Identify Verification Method

For the claim, determine:

• What command/operation produces evidence?
• What would a passing result look like?
• What would a failing result look like?

Phase 3: Execute and Collect Evidence

Run the verification command and capture output:

$ <verification command>
<actual output>

Phase 4: Evaluate

## Verification: <claim>

**Evidence:**
<quoted relevant output>

**Verdict:** [PASS — claim confirmed | FAIL — claim disproved | UNCERTAIN — insufficient evidence]

**Explanation:** <brief interpretation of what the evidence shows>

Phase 5: Report

Present the verdict with the evidence citation. If UNCERTAIN, state what additional evidence is needed.

<Tool_Usage>

• Use Bash to run verification commands
• Use Read to examine file contents as evidence
• Use Grep to trace code paths
• Use mcp__plugin_oh-my-claudecode_t__lsp_diagnostics for type-check verification </Tool_Usage>

``` ## Verification: "TypeScript compiles with no errors"

Evidence: $ cd plugins/oma && npx tsc --noEmit [no output — exit code 0]

Verdict: PASS — claim confirmed. tsc exited with code 0, no errors in output.

</Good>

<Bad>

"TypeScript should compile since we just fixed the types."

Why bad: "Should" is not evidence. Run tsc and cite the actual output.
</Bad>
</Examples>

<Final_Checklist>
- [ ] Claim parsed and stated explicitly
- [ ] Verification command executed
- [ ] Actual output captured and quoted
- [ ] Verdict is evidence-based, not assumption-based
- [ ] If fail/disprove: the evidence is stated alongside the verdict
</Final_Checklist>