---

name: openagent-verification-before-completion description: Use when about to claim work is complete, fixed, or passing with OpenAgent approval gates - requires requesting permission before verification, reporting evidence after verification, and approval before any completion claims or commits

Verification Before Completion - OpenAgent Version

Overview

Claiming work is complete without verification is dishonesty, not efficiency.

Core principle: Evidence before claims, always. With approval gates before verification, after verification, and before any completion claims.

Violating the letter of this rule is violating the spirit of this rule.

OpenAgent integration: This skill integrates approval gates from OpenAgent's safety-first philosophy. You will request approval before running verification commands, report evidence after verification, and request approval before making any completion claims or git operations.

The Iron Law

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

If you haven't run the verification command in this message, you cannot claim it passes.

The Gate Function with Approval Gates

BEFORE claiming any status or expressing satisfaction:

1. ⏸️ REQUEST APPROVAL: "May I verify [what] by running [command]?"
2. IDENTIFY: What command proves this claim?
3. RUN: Execute the FULL command (fresh, complete)
4. READ: Full output, check exit code, count failures
5. ⏸️ REPORT EVIDENCE: "Verification complete. Evidence: [show full output]. Status: [actual state]"
6. VERIFY: Does output confirm the claim?
   - If NO: State actual status with evidence
   - If YES: Proceed to next gate
7. ⏸️ REQUEST APPROVAL: "May I claim [completion/success] based on this evidence?"
8. ONLY THEN: Make the claim (after approval)

Skip any step = lying, not verifying
Skip any approval gate = violating OpenAgent safety

Approval Gates in Verification Workflow

digraph verification_openagent {
    rankdir=TB;
    need_claim [label="Need to claim\ncompletion/success", shape=ellipse];
    approval_verify [label="⏸️ REQUEST\nAPPROVAL\nto verify", shape=diamond, style=filled, fillcolor="#ffffcc"];
    identify [label="IDENTIFY\nVerification command", shape=box];
    run [label="RUN\nFull command", shape=box];
    read [label="READ\nOutput + exit code", shape=box];
    report_evidence [label="⏸️ REPORT\nEVIDENCE\n[show output]", shape=box, style=filled, fillcolor="#ffffcc"];
    check [label="Does output\nconfirm claim?", shape=diamond];
    state_failure [label="STATE\nActual status\nwith evidence", shape=box];
    approval_claim [label="⏸️ REQUEST\nAPPROVAL\nto claim success", shape=diamond, style=filled, fillcolor="#ffffcc"];
    make_claim [label="MAKE CLAIM\n(after approval)", shape=box];
    approval_commit [label="⏸️ REQUEST\nAPPROVAL\nfor commit/PR", shape=diamond, style=filled, fillcolor="#ffffcc"];
    commit [label="COMMIT/PR\n(after approval)", shape=box];
    done [label="Done", shape=ellipse];

    need_claim -> approval_verify;
    approval_verify -> identify [label="approved"];
    identify -> run;
    run -> read;
    read -> report_evidence;
    report_evidence -> check;
    check -> state_failure [label="NO"];
    check -> approval_claim [label="YES"];
    state_failure -> done;
    approval_claim -> make_claim [label="approved"];
    make_claim -> approval_commit;
    approval_commit -> commit [label="approved"];
    commit -> done;
}

Common Failures

Claim	Requires	Not Sufficient
Tests pass	Approval → Test command output: 0 failures → Report evidence → Approval	Previous run, "should pass"
Linter clean	Approval → Linter output: 0 errors → Report evidence → Approval	Partial check, extrapolation
Build succeeds	Approval → Build command: exit 0 → Report evidence → Approval	Linter passing, logs look good
Bug fixed	Approval → Test original symptom: passes → Report evidence → Approval	Code changed, assumed fixed
Regression test works	Approval → Red-green cycle verified → Report evidence → Approval	Test passes once
Agent completed	Approval → VCS diff shows changes → Report evidence → Approval	Agent reports "success"
Requirements met	Approval → Line-by-line checklist → Report evidence → Approval	Tests passing

Red Flags - STOP

• Using "should", "probably", "seems to"
• Expressing satisfaction before verification ("Great!", "Perfect!", "Done!", etc.)
• About to commit/push/PR without verification
• Trusting agent success reports
• Relying on partial verification
• Thinking "just this once"
• Tired and wanting work over
• ANY wording implying success without having run verification
• Skipping approval gates to "save time" or "be efficient"
• Running verification without requesting approval first
• Making completion claims without reporting evidence first

Rationalization Prevention

Excuse	Reality
"Should work now"	RUN the verification with approval
"I'm confident"	Confidence ≠ evidence
"Just this once"	No exceptions
"Linter passed"	Linter ≠ compiler
"Agent said success"	Verify independently with approval
"I'm tired"	Exhaustion ≠ excuse
"Partial check is enough"	Partial proves nothing
"Different words so rule doesn't apply"	Spirit over letter
"Approval gates slow me down"	Safety > speed. False claims waste more time.
"I already know it works"	Evidence ≠ assumption. Request approval and verify.
"User is waiting"	User wants truth, not false claims.
"Approval is obvious"	Request explicitly. Never assume.

Key Patterns with Approval Gates

Tests:

⏸️ "May I verify all tests pass by running npm test?"
[After approval]
✅ [Run test command] [See: 34/34 pass]
⏸️ "Verification complete. Evidence: 34/34 tests pass. May I claim all tests pass?"
[After approval]
"All tests pass"

❌ "Should pass now" / "Looks correct"
❌ "Tests pass" (without approval gates)

Regression tests (TDD Red-Green):

⏸️ "May I verify the regression test by running the red-green cycle?"
[After approval]
✅ Write → Run (pass) → Revert fix → Run (MUST FAIL) → Restore → Run (pass)
⏸️ "Verification complete. Evidence: [show all three outputs]. May I claim regression test works?"
[After approval]
"Regression test verified with red-green cycle"

❌ "I've written a regression test" (without red-green verification)
❌ Skipping approval to request red-green verification

Build:

⏸️ "May I verify the build by running npm run build?"
[After approval]
✅ [Run build] [See: exit 0]
⏸️ "Verification complete. Evidence: build exit 0, no errors. May I claim build passes?"
[After approval]
"Build passes"

❌ "Linter passed" (linter doesn't check compilation)
❌ "Build passes" (without approval and evidence)

Requirements:

⏸️ "May I verify all requirements are met by checking against the spec?"
[After approval]
✅ Re-read plan → Create checklist → Verify each → Report gaps or completion
⏸️ "Verification complete. Evidence: [checklist with status for each item]. May I claim phase complete?"
[After approval if all items checked]
"Phase complete - all requirements met"

❌ "Tests pass, phase complete"
❌ Claiming completion without checklist verification

Agent delegation:

⏸️ "May I verify the agent's work by checking the VCS diff?"
[After approval]
✅ Agent reports success → Check VCS diff → Verify changes → Report actual state
⏸️ "Verification complete. Evidence: [diff summary]. Status: [actual state]. May I claim agent completed successfully?"
[After approval if changes match expectations]
"Agent completed successfully"

❌ Trust agent report without verification
❌ Verify diff without approval gates

Before commit/PR:

⏸️ "May I verify all changes are ready by running [test command, linter, build]?"
[After approval]
✅ Run verification → Report evidence
⏸️ "Verification complete. Evidence: [test results, linter output, build status]. All checks pass. May I commit/create PR?"
[After approval]
Proceed with commit/PR

❌ Commit/PR without verification
❌ Commit/PR without approval after verification

Why This Matters

From 24 failure memories:

• your human partner said "I don't believe you" - trust broken
• Undefined functions shipped - would crash
• Missing requirements shipped - incomplete features
• Time wasted on false completion → redirect → rework
• Violates: "Honesty is a core value. If you lie, you'll be replaced."

OpenAgent addition: Approval gates prevent rushing to false claims. They create natural checkpoints where you pause, gather evidence, and ensure accuracy before proceeding.

When To Apply

ALWAYS before:

• ANY variation of success/completion claims
• ANY expression of satisfaction
• ANY positive statement about work state
• Committing, PR creation, task completion
• Moving to next task
• Delegating to agents
• ANY git operations (commit, push, PR)

Rule applies to:

• Exact phrases
• Paraphrases and synonyms
• Implications of success
• ANY communication suggesting completion/correctness

Approval gates required at:

1. Before running verification command
2. After verification complete (report evidence)
3. Before making completion claim
4. Before committing or creating PR

Example: Complete Workflow with Approval Gates

Scenario: Bug fix complete, ready to commit

❌ WITHOUT approval gates (WRONG):

"I've fixed the bug. Tests pass. Committing now."
[Skips verification, makes claim without evidence, commits without approval]

✅ WITH approval gates (CORRECT):

"May I verify the bug fix by running the full test suite?"

[After approval]
[Runs: npm test]
[Reads output: 42/42 tests pass, 0 failures, exit 0]

"Verification complete. Evidence:
- npm test output: 42/42 tests pass
- Exit code: 0
- No failures or errors

The bug fix appears to be working. May I claim the tests pass?"

[After approval]
"All tests pass. May I verify the build by running npm run build?"

[After approval]
[Runs: npm run build]
[Reads output: build successful, exit 0]

"Build verification complete. Evidence:
- npm run build: successful
- Exit code: 0
- No build errors

May I commit these changes?"

[After approval]
git add [files]
git commit -m "fix: [description]"

"Committed successfully."

OpenAgent Approval Gate Summary

Approval points in verification workflow:

1. ⏸️ Before running verification command - "May I verify [what] by running [command]?"
2. ⏸️ After verification complete - "Verification complete. Evidence: [full output]. Status: [actual state]"
3. ⏸️ Before making completion claim - "May I claim [success/completion] based on this evidence?"
4. ⏸️ Before committing/PR - "May I commit/create PR with these changes?"

Why approval gates matter:

• Prevents rushing to false completion claims
• Ensures user sees evidence before claims are made
• Aligns with OpenAgent's safety-first philosophy
• Creates natural checkpoints for review
• Reinforces "evidence before assertions always"
• Prevents "I'm confident it works" from bypassing verification

The Bottom Line

No shortcuts for verification. No skipping approval gates.

1. Request approval to verify
2. Run the command
3. Read the output
4. Report evidence
5. Request approval to claim
6. THEN claim the result (after approval)

This is non-negotiable.

Final Rule

Completion claim → approval requested → verification run → evidence reported → approval for claim → claim made
Otherwise → not OpenAgent verification

No exceptions without your human partner's permission.