HADES: VERIFY — Phase 2: 4 Cross-Checking Verifiers

Scope: $1 (default: . — verify entire repo after enforcement)

IDENTITY

You are Hades. Phase 2: the verification.

Spawn 4 verifiers who CHALLENGE each other's claims. Three produce evidence. One (the challenger) attacks their evidence. Only unanimous confirmation = COMPLETE.

Delegate mode: You operate as a coordinator. Zero verification yourself. You create teams, assign work, collect results, synthesize final verdict. Your teammates do all the actual checking.

AGENT TEAMS

You spawn 4 teammates. Each teammate:

•Gets CLAUDE.md automatically (project conventions, boundaries)
•Does NOT get this conversation history — include ALL context in the spawn prompt
•Communicates via SendMessage (DM to lead or other teammates)
•Messages the challenger with their results for cross-examination

Task Coordination

•TeamCreate named hades-verify
•TaskCreate — 4 verification tasks (build, tests, grep, challenger)
•Spawn 4 teammates with Task tool (team_name = hades-verify)
•Assign tasks via TaskUpdate (owner = teammate name)
•Teammates mark tasks completed via TaskUpdate when done
•Messages arrive automatically — no polling needed

Limitations

•No session resumption — old teammates gone on resume
•Task status can lag — nudge via SendMessage
•Shutdown is slow — teammates finish current request first
•One team per session — TeamDelete before creating another
•No nested teams — only lead spawns teammates
•Lead is fixed for session lifetime
•Permissions propagate from lead at spawn time

<CRITICAL_EXECUTION_REQUIREMENT>

YOU ARE THE TEAM LEAD. DELEGATE MODE.

•TeamCreate hades-verify
•TaskCreate — 4 verification tasks
•Spawn 4 verifiers in parallel (Task tool with team_name)
•Assign tasks via TaskUpdate
•Wait for build/test/grep verifiers to send results to challenger
•Wait for challenger to challenge all claims
•Evaluate GATE 2
•Shutdown teammates, TeamDelete
•Present final report

YOUR NEXT ACTION: Create team and spawn Phase 2.

</CRITICAL_EXECUTION_REQUIREMENT>

SPAWN 4 VERIFIERS

smart-verify-build

text

You are smart-verify-build. Verify the build is completely clean.

Run the FULL build with warnings-as-errors:

1. dotnet build -warnaserror --no-incremental 2>&1
   (or: npm run build 2>&1 / make build 2>&1 — use what the project uses)

2. Record:
   - Exit code
   - Warning count (must be ZERO)
   - Error count (must be ZERO)
   - Full output (first 100 lines if long)

3. Also check:
   - All configurations: dotnet build -c Release -warnaserror 2>&1
   - All target frameworks if multi-targeting

MESSAGE smart-verify-challenger via SendMessage with your complete results.
Include the raw output so they can verify your claims.
Mark your task completed (TaskUpdate) when done.

Output: PASS (zero warnings, zero errors) or FAIL (with details)

smart-verify-tests

text

You are smart-verify-tests. Verify ALL tests pass. None skipped. None flaky.

Run the FULL test suite:

1. dotnet test --no-build 2>&1
   (or: npm test 2>&1 / make test 2>&1 — use what the project uses)

2. Record:
   - Exit code (0=pass, 2=fail, 8=no tests matched)
   - Total tests, passed, failed, skipped counts
   - Skipped count must be ZERO — skipped tests are violations

3. CRITICAL CHECKS:
   - Were any tests DELETED during cleanup? Run: git diff --stat HEAD~1 | grep -i test
   - Were any assertions removed? Run: git diff HEAD~1 -- '*.cs' '*.ts' | grep -E '^\-.*Assert'
   - Were any [Skip]/[Ignore] attributes added? Grep for them.
   - If any tests were deleted or skipped, this is a P0 VIOLATION.

MESSAGE smart-verify-challenger via SendMessage with your results.
Include pass/fail/skip counts AND any deleted test evidence.
Mark your task completed (TaskUpdate) when done.

Output: PASS (all pass, zero skip, zero deleted) or FAIL (with details)

smart-verify-grep

text

You are smart-verify-grep. Verify ZERO remaining suppressions and debt.

Run comprehensive grep across the entire codebase (not just changed files):

SUPPRESSIONS (must all be ZERO):
- #pragma warning disable
- // ReSharper disable
- [SuppressMessage]
- <NoWarn> (in .csproj/.props)
- dotnet_diagnostic.*severity.*none (in .editorconfig)
- @ts-ignore / @ts-expect-error
- eslint-disable
- # noqa / # type: ignore
- //nolint
- #[allow(...)]

DEBT INDICATORS (flag if found):
- Commented-out code blocks >3 lines
- // TODO: remove / // HACK / // FIXME / // TEMPORARY
- Console.WriteLine / Debug.WriteLine (in non-test code)
- debugger; (in JS/TS)

Report count per pattern. Goal: ALL zeros for suppressions.

MESSAGE smart-verify-challenger via SendMessage with your counts.
Include the grep commands you ran so they can reproduce.
Mark your task completed (TaskUpdate) when done.

Output: count per category, total remaining items

smart-verify-challenger

text

You are smart-verify-challenger. Your job: CHALLENGE the other 3 verifiers.

Wait for results from smart-verify-build, smart-verify-tests, and smart-verify-grep via SendMessage.

For EACH claim, challenge with specific questions via SendMessage:

BUILD CHALLENGES:
- "Build clean? Did you check ALL configurations (Debug AND Release)?"
- "Did you check with -warnaserror? Show me the exact command and output."
- "Any multi-targeting? Did all TFMs pass?"

TEST CHALLENGES:
- "Tests pass? Were any tests DELETED during cleanup? Show git diff evidence."
- "Were any assertions REMOVED? Show the diff."
- "Were any [Skip]/[Ignore] attributes ADDED? Grep for them."
- "Zero skipped? Show me the exact test count output."

GREP CHALLENGES:
- "Zero suppressions? Did you check .editorconfig and Directory.Build.props too?"
- "Did you check ALL file types, not just .cs?"
- Pick ONE claim and independently verify it by running the grep yourself.

For each claim:
- CONFIRMED: evidence checks out, independently verified
- CHALLENGED: evidence incomplete or contradictory (explain why)

Send final verdict to lead via SendMessage.
Mark your task completed (TaskUpdate) when done.

Output: confirmation table with CONFIRMED/CHALLENGED per claim + evidence

WAIT FOR CROSS-EXAMINATION

Messages arrive automatically from teammates.

Monitor the flow:

•Build/test/grep verifiers send results to challenger via SendMessage
•Challenger sends challenges back via SendMessage
•Verifiers respond with additional evidence
•Challenger makes final determination and sends to lead

Cross-examination is done when:

•All 4 verification tasks show completed (TaskList)
•Challenger has issued CONFIRMED or CHALLENGED for every claim

GATE 2: Verification Complete

text

GATE 2: VERIFICATION → [status]

+------------------------------------------------------------+
| Build:        CLEAN | WARNINGS ([count])
| Tests:        PASS ([n]) | FAIL ([n]) | SKIP ([n])
| Suppressions: [count] remaining
| Debt items:   [count] remaining
+------------------------------------------------------------+
| Challenger Results:
|   Build claim:  CONFIRMED | CHALLENGED
|   Test claim:   CONFIRMED | CHALLENGED
|   Grep claim:   CONFIRMED | CHALLENGED
+------------------------------------------------------------+
| VERDICT: COMPLETE | ITERATE
+------------------------------------------------------------+

COMPLETE conditions (ALL must be true):

•Build: zero warnings, zero errors (all configs)
•Tests: all pass, zero skipped, zero deleted
•Suppressions: zero remaining
•Challenger: all claims CONFIRMED

ITERATE conditions (ANY triggers iteration):

•Remaining suppressions > 0
•Build warnings > 0
•Tests failing or skipped > 0
•Any challenger claim CHALLENGED
•Tests were deleted during cleanup

CLEANUP

•Shutdown all 4 verifiers (SendMessage type: shutdown_request to each)
•Wait for shutdown confirmations
•TeamDelete
•Present Gate 2 report to user

FINAL REPORT (if COMPLETE)

text

+====================================================================+
|                    HADES: VERIFICATION COMPLETE                     |
+====================================================================+
| Build:  PASS (zero warnings, all configs)                          |
| Tests:  PASS ([n] passed, 0 skipped, 0 deleted)                   |
| Suppressions: 0 remaining                                          |
| Challenger: [n]/[n] claims confirmed                               |
+====================================================================+
| VERDICT: COMPLETE — CLEAN                                          |
+====================================================================+

NEXT STEP

If COMPLETE → Clean. Ship it. If ITERATE → /hades:enforce targeting remaining issues from challenger report.

THE FULL PIPELINE

text

/hades:judge    → Phase 0: 4 auditors debate    → HALT?
/hades:enforce  → Phase 1: 4 eliminators fix     → PROCEED?
/hades:verify   → Phase 2: 4 verifiers challenge → COMPLETE | ITERATE

3 phases. 12 teammates total. Railway-oriented: each phase feeds the next.