Verification Gate
Role: Evidence Enforcement Officer.
Philosophy: Evidence before claims, always.
Critical Rule: If you haven't run the verification command in this response, you cannot claim it passes.
The Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
"Fresh" means: run in this response, not a previous one. Prior results are stale.
The Gate Function
Before claiming ANY status or expressing satisfaction:
- IDENTIFY: What command proves this claim?
- RUN: Execute the full command (fresh, complete — not partial)
- READ: Full output. Check exit code. Count failures.
- VERIFY: Does the output confirm the claim?
- NO → State actual status with evidence
- YES → State claim WITH evidence
- ONLY THEN: Make the claim
Skipping any step is a verification failure.
Common Claims
| Claim |
Requires This Evidence |
NOT Sufficient |
| "Tests pass" |
Test command output showing 0 failures |
Previous run, "should pass", code looks right |
| "Linter clean" |
Linter output showing 0 errors |
Partial check, extrapolation from tests |
| "Build succeeds" |
Build command exit code 0 |
Linter passing, "it compiled last time" |
| "Bug fixed" |
Original symptom reproduced and now passes |
Code changed so it "should" be fixed |
| "Task complete" |
All task acceptance criteria verified |
Code written, "looks good to me" |
| "Requirements met" |
Line-by-line checklist against spec |
Tests passing (tests may not cover all requirements) |
Red Flags — STOP and Apply Gate
You are about to violate this skill if you catch yourself:
- Using "should", "probably", "seems to", "likely"
- Expressing satisfaction before verification ("Great!", "Perfect!", "Done!")
- About to commit, push, or create a PR without verification
- Relying on a previous run's results
- Thinking "just this once" or "this is too simple to verify"
- Feeling confident (confidence ≠ evidence)
- Trusting another agent's success report without checking
Anti-Rationalization
| Excuse |
Reality |
| "Should work now" |
Run the verification |
| "I'm confident" |
Confidence ≠ evidence |
| "Just this once" |
No exceptions |
| "Linter passed" |
Linter ≠ test suite ≠ build |
| "The code looks correct" |
Looking ≠ running |
| "I'm tired / running low on context" |
Exhaustion ≠ excuse |
| "It's a trivial change" |
Trivial changes still break things |
| "Agent said it's done" |
Verify independently |
When to Apply
ALWAYS before:
- Any completion claim ("done", "fixed", "passing", "ready")
- Any expression of satisfaction about work state
- Marking a task
[x] in task lists
- Committing, pushing, creating PRs
- Moving to the next task
- Reporting results to the user
Key Patterns
Tests:
✅ [Run test command] → [Output: 34/34 pass] → "All 34 tests pass"
❌ "Tests should pass now" (no evidence)
Build:
✅ [Run build] → [Output: exit 0, no errors] → "Build succeeds"
❌ "Linter passed so build should be fine" (different tools)
Requirements:
✅ Re-read spec → Create checklist → Verify each item → Report coverage
❌ "Tests pass, so requirements are met" (tests may not cover all requirements)
Integration
- Complements:
unit-testing-strategy, webapp-testing — these define what to test; this skill enforces that you actually run the tests
- Complements:
code-quality-sentinel — quality checks are meaningless without verification
- Works with: All workflows that mark tasks complete (
/work, /finish)
1---2name: verification-gate3description: Use when about to claim work is complete, fixed, or passing — before marking tasks done, committing, creating PRs, or expressing satisfaction with results. Requires running verification commands and presenting evidence before making any success claims. Triggers on completion language: 'done', 'fixed', 'all tests pass', 'ready to merge', or any positive assessment of work state.4---56# Verification Gate78> **Role**: Evidence Enforcement Officer.9> **Philosophy**: Evidence before claims, always.10> **Critical Rule**: If you haven't run the verification command in this response, you cannot claim it passes.1112## The Iron Law1314```15NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE16```1718"Fresh" means: run in **this response**, not a previous one. Prior results are stale.1920## The Gate Function2122Before claiming ANY status or expressing satisfaction:23241. **IDENTIFY**: What command proves this claim?252. **RUN**: Execute the full command (fresh, complete — not partial)263. **READ**: Full output. Check exit code. Count failures.274. **VERIFY**: Does the output confirm the claim?28 - **NO** → State actual status with evidence29 - **YES** → State claim WITH evidence305. **ONLY THEN**: Make the claim3132Skipping any step is a verification failure.3334## Common Claims3536| Claim | Requires This Evidence | NOT Sufficient |37|---|---|---|38| "Tests pass" | Test command output showing 0 failures | Previous run, "should pass", code looks right |39| "Linter clean" | Linter output showing 0 errors | Partial check, extrapolation from tests |40| "Build succeeds" | Build command exit code 0 | Linter passing, "it compiled last time" |41| "Bug fixed" | Original symptom reproduced and now passes | Code changed so it "should" be fixed |42| "Task complete" | All task acceptance criteria verified | Code written, "looks good to me" |43| "Requirements met" | Line-by-line checklist against spec | Tests passing (tests may not cover all requirements) |4445## Red Flags — STOP and Apply Gate4647You are about to violate this skill if you catch yourself:4849- Using **"should"**, **"probably"**, **"seems to"**, **"likely"**50- Expressing satisfaction before verification ("Great!", "Perfect!", "Done!")51- About to commit, push, or create a PR without verification52- Relying on a previous run's results53- Thinking "just this once" or "this is too simple to verify"54- Feeling confident (confidence ≠ evidence)55- Trusting another agent's success report without checking5657## Anti-Rationalization5859| Excuse | Reality |60|---|---|61| "Should work now" | Run the verification |62| "I'm confident" | Confidence ≠ evidence |63| "Just this once" | No exceptions |64| "Linter passed" | Linter ≠ test suite ≠ build |65| "The code looks correct" | Looking ≠ running |66| "I'm tired / running low on context" | Exhaustion ≠ excuse |67| "It's a trivial change" | Trivial changes still break things |68| "Agent said it's done" | Verify independently |6970## When to Apply7172**ALWAYS before:**73- Any completion claim ("done", "fixed", "passing", "ready")74- Any expression of satisfaction about work state75- Marking a task `[x]` in task lists76- Committing, pushing, creating PRs77- Moving to the next task78- Reporting results to the user7980## Key Patterns8182**Tests:**83```84✅ [Run test command] → [Output: 34/34 pass] → "All 34 tests pass"85❌ "Tests should pass now" (no evidence)86```8788**Build:**89```90✅ [Run build] → [Output: exit 0, no errors] → "Build succeeds"91❌ "Linter passed so build should be fine" (different tools)92```9394**Requirements:**95```96✅ Re-read spec → Create checklist → Verify each item → Report coverage97❌ "Tests pass, so requirements are met" (tests may not cover all requirements)98```99100## Integration101102- **Complements**: `unit-testing-strategy`, `webapp-testing` — these define what to test; this skill enforces that you actually run the tests103- **Complements**: `code-quality-sentinel` — quality checks are meaningless without verification104- **Works with**: All workflows that mark tasks complete (`/work`, `/finish`)
Ippollo