verify-completion

Verification Before Completion

Overview

Core principle: Evidence before claims, always.

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

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

BEFORE claiming any status or expressing satisfaction:

1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
   - If NO: State actual status with evidence
   - If YES: State claim WITH evidence
5. ONLY THEN: Make the claim

Skip any step = lying, not verifying

Common Failures

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

Goal-Backward Verification

The gate function above stops the most common failure — claiming without running the command. It does not stop the next-most-common one: running the wrong command. A test-suite pass is evidence for "these tests pass", not for "the user's stated goal is met", and confusing the two is how a green suite still ships a broken feature.

Work backward from what the user asked for:

1. Restate the goal in a single testable sentence. If the user said "make the export button work", the sentence is not "the button click handler runs" — it is "clicking Export downloads a file with today's data".
2. Name the observation that would prove that sentence true. Not a proxy. Not "the network call fires". The end-to-end artifact the user would look at: a downloaded CSV with today's rows, a rendered page, a deployed URL returning 200.
3. Choose a verification command that produces THAT observation. A unit test is evidence for a unit; the goal-observation is what needs a command of its own — a manual click, a curl, an end-to-end run, a screenshot.
4. Only then run the local sanity checks. Tests, lint, build. They are necessary and insufficient: they catch regressions in the pieces, but a green suite over the wrong pieces is silent.

Verification is not permission to manufacture a clean slate. Inspect the target before acting. If an existing file, process, or remote object could make the observation ambiguous, use an isolated destination or ask before changing it; never delete or overwrite user-owned state merely to make the verification command easier to interpret.

The failure to catch: every step of the plan passed its own tests, no step's evidence was the goal, and the finished stack does not do the thing. Fixture: tests/goal-backward-scenario.md.

Local-check evidence	Goal-backward evidence
"The 34 unit tests pass"	"I clicked Export and a CSV with today's 12 rows downloaded"
"The build exits 0"	"The deployed URL returns 200 and renders the new field"
"The migration ran"	"SELECT on the new column returns the expected shape"
"The API test passes"	"The UI that calls that API shows the new value"

Both belong in the evidence you cite. The local checks catch regressions; the goal-backward observation catches the wrong-thing-passing-its-tests failure mode a green suite cannot see.

Red Flags - STOP

• Using "should", "probably", "seems to"
• Expressing satisfaction before verification ("Great!", "Perfect!", "Done!", etc.)
• About to commit/push/MR 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

Rationalization Prevention

Excuse	Reality
"Should work now"	RUN the verification
"I'm confident"	Confidence ≠ evidence
"Just this once"	No exceptions
"Linter passed"	Linter ≠ compiler
"Agent said success"	Verify independently
"I'm tired"	Exhaustion ≠ excuse
"Partial check is enough"	Partial proves nothing
"Different words so rule doesn't apply"	Spirit over letter

Key Patterns

Tests:

✅ [Run test command] [See: 34/34 pass] "All tests pass"
❌ "Should pass now" / "Looks correct"

Regression tests (TDD Red-Green):

✅ Write → Run (pass) → Revert fix → Run (MUST FAIL) → Restore → Run (pass)
❌ "I've written a regression test" (without red-green verification)

Build:

✅ [Run build] [See: exit 0] "Build passes"
❌ "Linter passed" (linter doesn't check compilation)

Requirements:

✅ Re-read plan → Create checklist → Verify each → Report gaps or completion
❌ "Tests pass, phase complete"

Agent delegation:

✅ Agent reports success → Check VCS diff → Verify changes → Report actual state
❌ Trust agent report

When To Apply

ALWAYS before:

• ANY variation of success/completion claims
• ANY expression of satisfaction
• ANY positive statement about work state
• Committing, MR creation, task completion
• Moving to next task
• Delegating to agents

Rule applies to:

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

How this is watched

The moe-completion-evidence Stop hook (default-on) reads the transcript window for this turn only — bounded by the last human-user entry, not a type:"user" tool-result row — and writes an audit record to ~/.config/moe/core/audit/<project-key>/ (never the worktree; a session's evidence can run into the tens of megabytes). If it sees a completion-claim phrase in this turn's assistant text with no verification command in this turn's tool_uses, the record carries a warning and stderr says so. The hook never blocks a stop; it makes the silence falsifiable. Set MOE_EVIDENCE_DISABLED=1 to opt out.