/supergraph:verify

Requires healthy index_status for CBM_PROJECT (see references/codebase-memory-contract.md); then codebase-memory-mcp detect_changes, trace_path, recipes cycles/hubs/bridges/test-gaps. Block on degraded/new cycle/unapproved hub/missing tests.

Fresh verification gate before completion claims.

Iron Law: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

When to Use

Before saying/implying: done, fixed, passing, clean, ready, complete, works, verified, successful. Also: before marking Status: completed, committing, PR'ing, accepting agent reports, moving to next phase.

Usage

/supergraph:verify | tests | lint | build | plan auth-login task 2 | claim "login fixed"

Evidence Mapping

Claim	Required fresh evidence
Tests pass	$TEST_CMD exits 0, zero failures
Lint clean	$LINT_CMD exits 0, zero errors
Build succeeds	$BUILD_CMD exits 0
Bug fixed	Original failing symptom/test now passes
Regression covered	RED failed before fix, GREEN passes after
TDD complete	RED failure valid, GREEN passes, refactor verified
Task completed	Acceptance criteria met + verification commands pass
Agent completed	Diff inspected + independent tests/lint
Review passed	Reviewer + graph review + tests/lint pass
Ready to merge	Tests + lint + build + review PASS

Steps

0. Announce

"✅ /supergraph:verify — verifying [claim] with fresh evidence..."

1. Identify Claim

Map claim to required proofs from evidence mapping.

2. Select Plan Context (if applicable)

0 plans → skip | 1 → use | >1 → ask | plan <slug> → match.

3. Get Commands

Read from plan ## Environment Context or .supergraph-env (set by /supergraph:scan). Missing → STOP, run scan first. No command can prove claim → STOP: "Required check is unknown."

3b. Serena sweep (optional): See serena/SKILL.md:Setup — get_diagnostics_for_file per touched file; STOP on type errors before test suite. Skip if SERENA_ACTIVE=false or unavailable.

4. Run Fresh Verification (NOW — no reuse of old output unless freshness guard passes)

For agent output: git diff --stat && git diff --name-only. Then run relevant tests/lint/build locally. Freshness guard (dedup suite, preserves Iron Law): Reuse of immediately prior evidence is allowed only if HEAD_SHA unchanged and git diff --name-only HEAD unchanged and last PASS timestamp <120s ago (same commit). Otherwise must rerun $TEST_CMD/$LINT_CMD fresh NOW. Guard respects CBM_INDEXED_AT TTL from /supergraph:scan but never skips verification before merge/PR — if HEAD_SHA changed, reuse is invalid.

5. Read Output

Check: exit code, failure count, error count, warnings. Don't confuse lint ≠ build, targeted ≠ suite.

6. Report Evidence

Pass:

## Verification Evidence
- Claim: [claim] | Command: `[command]` → exit 0 → PASS
- Evidence: [output summary] | Timestamp: [now]

Fail:

## Verification Failed
- Claim: [claim] | Command: `[command]` → exit [N] → FAIL
- Failure: [summary] | Next: /supergraph:fix

Cannot verify → say so explicitly. No completion language.

7. Update Plan Status

Only after evidence passes: in_progress → completed. If failed → keep status or mark stuck if retries exhausted.

For user-facing confirmation: Announce task/plan completion in the user's language. If plan context exists, show: "Task N completed — [user-facing summary]"

8. Report

✅ /supergraph:verify complete
- Claim: [claim] | Evidence: [command] → PASS|FAIL
- Next: [recommendation]

Anti-Patterns — Block These

• "should work" / "seems fixed" / "probably passes" / "looks good"
• Trusting agent reports without independent check
• Stale command output | No acceptance criteria check
• Commit/PR without fresh tests+lint+build+review evidence
• Hiding skipped checks behind positive wording

Evidence Rules

Evidence before claims, always. Fresh only — old runs don't count. Read output before reporting. No positive language without proof. Agent output requires independent verification.