name: integrity-check description: End-to-end 8-dimension verification — requirements through delivery, all aligned and correct triggers: [integrity, verify all, are we done, is it complete, end-to-end check, check everything, full verification] tags: [core] context_cost: high
Integrity Check Skill
Goal
Verify the ENTIRE project is consistent and correct from requirements to delivery. Catch regressions, drift, and gaps across 8 dimensions. Only orch-judge or human may sign off as "INTEGRITY VERIFIED."
8 Dimensions of Integrity
Dimension 1 — Requirements Integrity
- Does spec.md (or SPEC_TEMPLATE.md) cover ALL EARS requirements from PRD.md?
- Run spec-analyze.skill to check alignment
- Check: no requirements without tests
- Result: GREEN (all covered) | YELLOW (minor gaps) | RED (requirements missing)
Dimension 2 — Architecture Integrity
- Run agentic-linter.skill: zero architecture boundary violations
- Run
npx madge --circular src/orvendor/bin/deptrac: zero circular dependencies - Check: all architectural decisions have ADRs in
docs/architecture/decisions/ - Result: GREEN / YELLOW / RED
Dimension 3 — Implementation Integrity
- All tasks in project/TASKS.md have status: DONE
- No tasks with status TODO, IN_PROGRESS, or BLOCKED
- Every feature from PRD.md has corresponding code
- Result: GREEN / YELLOW / RED
Dimension 4 — Test Integrity
[test command] # Run all tests — MUST be 100% green
[coverage command] # Coverage MUST be >= 99%
- Zero failing tests (not even flaky ones currently failing)
- Zero skipped tests without documented reason
- Integration tests exist for all API endpoints
- Result: GREEN / YELLOW / RED
Dimension 5 — Security Integrity
[security_scan command] # dependency audit — zero critical/high CVEs
- SECURITY_CHECKLIST.md fully completed (not partially)
- Threat model exists for all security-sensitive features
- No secrets detected by gitleaks
- OWASP Top 10 items checked
- Result: GREEN / YELLOW / RED
Dimension 6 — Tech Debt Integrity
- No uncatalogued TODO/FIXME/HACK markers (all have task tickets)
- Cyclomatic complexity < 10 for all modified files
- No dead code in production branches
- TECH_DEBT_TEMPLATE.md entries are current
- Result: GREEN / YELLOW / RED
Dimension 7 — Documentation Integrity
- All public APIs have docstrings/JSDoc
- README reflects current state (not aspirational)
docs/api/openapi.yamlis in sync with actual implementation (if exists)- C4 diagrams reflect current architecture (if docs/architecture/ exists)
- Result: GREEN / YELLOW / RED
Dimension 8 — SDLC Integrity
- SDLC_TRACKER.md phase matches actual state of the project
- PROJECT_STATE.md current phase is accurate
- All SESSION_SNAPSHOT files exist for completed phases
- AUDIT_LOG.md has entries for all major decisions
- Result: GREEN / YELLOW / RED
Steps
-
Run all 8 dimension checks (in order of priority)
- Each dimension: run the relevant commands, read the relevant files
- Rate each: GREEN / YELLOW / RED with specific evidence
-
For each RED dimension: list specific items that must be fixed
-
For each YELLOW dimension: list items to address before production
-
Generate Integrity Report
## Integrity Report Date: [date] Project phase: [current SDLC phase] Run by: [agent/human] | Dimension | Status | Issues Found | |---|---|---| | 1. Requirements | GREEN | None | | 2. Architecture | RED | 2 circular deps in users module | | 3. Implementation | YELLOW | 1 task still IN_PROGRESS | | 4. Tests | GREEN | 247 passing, 84% coverage | | 5. Security | GREEN | SECURITY_CHECKLIST 100% | | 6. Tech Debt | YELLOW | 3 uncatalogued TODOs | | 7. Documentation | YELLOW | /api/users endpoint missing JSDoc | | 8. SDLC | GREEN | All snapshots current | ### OVERALL STATUS: RED (1 RED dimension = not release-ready) ### Blocking Issues (must fix before release) - [ARCH] Circular dependency: UserService <-> OrderService Fix: Extract IUserQueryPort interface ### Non-Blocking Issues (address before next sprint) - [IMPL] CreatePaymentTask still IN_PROGRESS - [DEBT] 3 TODO markers without task tickets - [DOCS] Missing JSDoc on 2 public functions ### INTEGRITY VERIFIED: NO Ready for production: NO Ready for staging: YES (if blocking issues fixed) -
Recommended actions (priority ordered)
- RED items: fix immediately (blocking)
- YELLOW items: fix before next gate
- GREEN items: monitor
-
Sign-off (only orch-judge persona or human may sign)
- When all dimensions GREEN: "INTEGRITY VERIFIED — READY FOR [next gate]"
Constraints
- NEVER mark integrity as verified if any dimension is RED
- NEVER skip dimensions — partial verification is not verification
- Security dimension RED = hard block on production deployment
- Run this check before every SDLC checkpoint and before any production deploy
Output Format
8-dimension integrity report with GREEN/YELLOW/RED per dimension + OVERALL STATUS + prioritized action list.
Security & Guardrails
1. Skill Security (Integrity Check)
- Check-Bypass Prevention: The integrity checking script/logic must be cryptographically pinned or read-only, preventing a compromised process from modifying the script to universally return
GREEN. - Secret Scanning Isolation: The phase detecting secrets (
gitleaks) must operate in a secure, ephemeral pipeline environment that aggressively wipes memory after execution so the detected secrets themselves aren't cached or leaked.
2. System Integration Security
- Hard-Blocking Deployments: The results of Dimension 5 (Security Integrity) must be integrated directly into the CI/CD pipeline as a non-overridable failure block. If
RED, the deployment system absolutely must refuse to release the artifact. - Comprehensive Dependency Auditing: Ensure the integrity check includes transitive dependencies (SCA), not just direct imports, detecting supply-chain attacks deep within the dependency graph.
3. LLM & Agent Guardrails
- Sign-Off Forgery Defense: The "INTEGRITY VERIFIED" sign-off must require a cryptographic signature or an explicit, out-of-band human 2FA approval step when transitioning from S09 to S10 (Production), preventing an LLM from hallucinating an approval.
- Confirmation Bias Mitigation: The
orch-judgepersona evaluating the 8 dimensions must not rely solely on self-reported agent logs; it must independently execute the underlying validation commands (tests, linters, scanners) to guarantee truth.