Imported from Undermybelt/hermes-skills (skills/aegis/verification-before-completion/SKILL.md). Install upstream with npx skills add Undermybelt/hermes-skills --skill verification-before-completion. Copyright stays with the author.

Execute

→ About to claim "done", "passing", "fixed", "complete"? → Run the verification command first. Then claim.

1. Identify: what command proves the claim?
2. Run: full command, fresh, complete
3. Read: output, exit code, failures
4. Verify: output confirms claim? → state claim WITH evidence. Doesn't? → state actual status. → Done when: exact command run, output confirms, residual risk stated, confidence graded. Non-trivial code changes → also report Complexity Delta and Complexity Governance Suggestion. Governance/retirement work → also close Repair Track + Retirement Track + Residual Risk.

Verification Before Completion

Overview

Claiming work is complete without verification is dishonesty, not efficiency. Evidence before claims, always.

Red Flags - STOP

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

When To Apply

Before ANY success/completion claim, expression of satisfaction, commit, PR, task completion, or delegation. Applies to exact phrases, paraphrases, and implications.

QA Closure

Before any success claim, include the required evidence semantic slots. Natural prose, localized headings, or compact cards are all valid when the slots remain explicit and auditable.

Required evidence slots, one allowed card rendering:
- Evidence action / check performed:
- Result / exit status:
- Covered scope:
- Uncovered scope:
- Residual risk:
- Confidence grade: A | B | C

Semantic Slots:

• Required governance fields may appear as localized headings, natural prose, or compact cards when they remain explicit and auditable.
• Natural Surface is valid when natural user-facing wording preserves the semantic slots; natural expression is not a reason to drop evidence, uncovered scope, residual risk, confidence, retirement, baseline, or architecture fields.
• In short: natural expression is valid only when it preserves semantic slots.
• Governance Receipt is the compact closeout form for Aegis-shaped non-trivial work. It names the boundary held, evidence, covered and uncovered scope, residual risk, confidence, and any triggered governance closure.

TDD Completion Boundary:

• Judge the completion claim against the highest available explicit boundary.
• Parent plan/spec acceptance decides whole-task completion; TaskIntentDraft decides current-task completion; Slice Card decides slice completion only.
• Match the boundary to the claim being made, and keep any higher open boundary explicit.
• If only slice-level evidence exists, do not claim whole-task done.
• If no explicit boundary exists and atomicity is not clear, downgrade to needs-verification or return to framing/planning.

1. Remove/Restore: side effects? temp instrumentation restored?

2. Evidence Bundle: exact command, scope, exit status, key output. State what's covered and what's not. Include target test and related regression evidence. When automation is blocked, provide reproducible manual verification steps.

3. Prompt Hygiene: when external output shaped judgment → state whether summaries or raw excerpts were used. Name large payloads not loaded. If summary insufficient → read back excerpt or lower claim. Include Evidence Used / Not Loaded / Next Evidence boundary when relevant.

4. Confidence: A (direct + regression, no unknowns) | B (direct, bounded risk) | C (partial only, not closed)

5. Authority: verified evidence ≠ authoritative completion. Keep distinct.

6. Goal Closure: when goal-framing or optional TaskIntentDraft goal fields shaped the work, explicitly check the goal before claiming completion:

   Available boundary check for completion judgment:

   1. Parent plan/spec acceptance for whole-task completion, when present
   2. TaskIntentDraft Goal / Success evidence / Non-goals for the current task, when present
   3. Slice Card Goal / Verification / Stop for the current slice, when present

   Goal Closure:
   - Goal status: satisfied | blocked | needs-verification | scope-exceeded
   - Success evidence:
   - Stop state: done | blocked | needs-verification | scope-exceeded
   - Non-goals respected:

   Use done only when success evidence is satisfied. Use blocked when a dependency, permission, or required fact is missing. Use needs-verification when implementation exists but evidence is insufficient. Use scope-exceeded when continuing would exceed the goal or non-goals.

7. Long-Task: re-read checkpoint, confirm every todo has status, no drift check unresolved.

8. Workspace Integrity: if the task created or modified a target project's docs/aegis/ workspace and configured Aegis workspace support is available, run python <aegis-workspace-helper> bundle --root <target-project-root> --work YYYY-MM-DD-<slug> when a work/ record exists, then run python <aegis-workspace-helper> check --root <target-project-root> and include the result in the evidence bundle. The generated proof bundle and workspace check validate method-pack structure, index coverage, and recognizable JSON artifact sidecars only; they do not judge evidence sufficiency and do not grant completion authority.

9. Readiness Summary: for release, merge, handoff, or "ready?" requests, organize the evidence into a compact readiness view after the evidence slots:

   Readiness Summary:
   - Tests:
   - Docs:
   - Version:
   - Host compatibility:
   - Uncovered scope:
   - Residual risk:

   A readiness summary is advisory evidence organization only. It is not authorization to commit, tag, publish, merge, or release. It cannot provide completion authority.

10. Natural Aegis closeout: when Aegis skills materially shaped a non-trivial task, keep Aegis explicitly visible in the final completion closeout.

The closeout should naturally show how Aegis influenced the result. Make at least one of the following user-visible:

• what boundary Aegis helped hold steady
• what evidence or verification discipline Aegis required before the claim
• what residual risk, uncovered scope, or restraint Aegis kept visible

Use one sentence when Aegis mainly helped hold one boundary steady, but more than one mention is valid when boundary, evidence, and residual-risk visibility each materially shaped the judgment.

Keep this integrated into the normal completion summary rather than a fixed slogan. Do not default to a visible Aegis Contribution Note: heading. Do not default to one canonical closeout phrase, and do not repeat the same Aegis closeout wording across unrelated tasks.

When Aegis materially shaped multiple parts of the judgment, it may appear more than once across the closeout, as long as each mention is tied to a concrete boundary, evidence decision, or risk callout.

Keep it advisory method-pack discipline, not completion authority. Keep it implicit only for obvious fast-path replies unless the user asked about Aegis routing.

Natural expression may satisfy the visibility requirement when the semantic slots are still explicit. For example, "I will follow the Aegis order here: read the owner / baseline and current implementation first, add a failing example for the main path, then make the minimal repair and verify it" is a valid natural transition before implementation. Completion still needs fresh evidence and the applicable Governance Receipt fields.

Use structured trace only for audit, debug, release, long-task review, or user request. The structured form may name skills, stage transitions, quality effect, and boundary, but it should not replace the normal user-facing completion note.

11. User-Language Output: final response cards must localize user-facing section labels, field labels, and explanatory prose to the user's language. Keep commands, file paths, code identifiers, stable enum values, and exact product names unchanged. For important Aegis product terms, include the stable English identifier only when it prevents ambiguity, usually beside a user-language explanation on first use.

12. Complexity Delta: for non-trivial code changes, inspect the actual diff before claiming completion. This is a completion-time entropy check, not a universal failure gate. Skip or keep it one-line for tiny wording edits, tests-only additions, generated files, vendored files, fixtures, lockfiles, or purely mechanical formatting where no maintained source owner gained complexity.

Use the project language for field labels in the final response, but keep the internal shape recognizable:

Complexity Delta:
- Files over 800 lines:
- Files newly crossing 800 lines:
- Largest touched file delta:
- Largest touched function/block:
- New branches/fallbacks/adapters:
- Retired branches/fallbacks/adapters:
- Net entropy: decreased | stable | increased-with-justification
- Required follow-up:

When the delta finds meaningful pressure, add:

Complexity Governance Suggestion:
- Recommendation: none | monitor | schedule-refactor | extract helper | split owner | open follow-up
- Why:
- Suggested scope:
- Timing:

Use none for small owner-correct diffs, monitor for acceptable visible growth, and stronger recommendations for 800+ line files, 80+ line blocks, branch/fallback/adapter growth without retirement, or owner mismatch. The suggestion is advisory; keep residual risk visible.

Rules:

• A maintained source file over 800 lines is a review signal. If this slice added logic there or pushed it across 800 lines, explain why the owner boundary is still correct or report a split/refactor follow-up.
• A touched function, method, component, or cohesive block over roughly 80 lines, deeply nested logic, or mixed reasons to change is a block-level complexity signal even if the file remains under 800 lines.
• New fallback, adapter, compatibility, guard, or branch logic must be paired with retired paths or a Retirement Closure entry. Net new paths without deletion or a scheduled retirement trigger count as entropy increase.
• If entropy increased and no stronger owner/compatibility reason exists, downgrade the completion claim or state the residual risk.

13. Baseline Alignment Check: before final response, if project instructions require baseline reporting or the task touched requirement, product, or durable architecture surfaces, include an explicit baseline alignment result. This is separate from ADR Backfill: alignment states whether the completed work matches current requirements and architecture baselines; ADR Backfill states whether durable architecture memory needs to be created, amended, superseded, or skipped. This is a method-pack signal, not a runtime gate, not an authoritative GateDecision, and not completion authority.

Product / Requirement Baseline covers the accepted problem, success evidence, non-goals, workflow constraints, and approved requirement/spec intent. Architecture / Runtime Boundary Baseline covers canonical owner, contract, source-of-truth, dependency direction, compatibility, runtime-ready/method-pack boundary, and retirement state.

Triggering surfaces include architecture, contracts, source-of-truth owner, canonical owner, context/answering/runtime flow, cross-module data flow, producer-to-carrier-to-consumer chains, public user-visible identity, evidence model, retained fallback, adapter, compatibility path, requirement acceptance, product non-goals, and project-specific baseline rules.

Baseline Alignment:
- Trigger: yes | no
- Product / Requirement Baseline:
- Architecture / Runtime Boundary Baseline:
- Requirement / acceptance alignment:
- Architecture / owner / contract alignment:
- Result: aligned | Design Defect | Implementation Drift | missing-authority | needs-clarification
- scope: requirements | architecture | both
- Evidence:
- Residual risk:

Use Design Defect when the relevant requirement, design, or baseline is wrong. Use Implementation Drift when the work deviates from a correct unchanged baseline. Architecture Defect and Architecture Drift remain compatibility aliases for architecture-scoped Design Defect and architecture-scoped Implementation Drift.

When project instructions specifically require architecture reporting or the completed work touched durable architecture surfaces, the architecture-scoped subset may also be reported as Architecture Alignment:

Architecture Alignment:
- Trigger: yes | no
- Scope:
- Baseline checked:
- Result: aligned | Design Defect | Implementation Drift | missing-authority | needs-clarification
- Evidence:
- Integrity Residual Risk:
- Residual architecture risk:

Use Integrity Residual Risk when ArchitectureReviewRequired: yes, an Architecture Integrity Lens shaped the plan or review, or the diff touches canonical owner, source-of-truth, fallback, adapter, or duplicate-owner surfaces. Name any unresolved responsibility overlap, missed higher-level owner / contract fix, retained caller-side fallback, or stale path that still needs retirement. If none remains, state none rather than expanding into a new gate.

14. ADR Backfill Check: for completed medium/high work that touched durable architecture surfaces, run the ADR Auto Backfill check before final completion claims. Use Trigger: no or skip the expanded block for simple wording edits, ordinary README cleanup, routine release-note edits, low-risk single-file changes, tests-only coverage improvements, and bug fixes that only restore the existing baseline. This is a method-pack signal, not a runtime gate, not an authoritative GateDecision, and not completion authority.

Durable architecture surfaces include canonical owner, public API/schema, artifact shape, behavior contract, dependency direction, source-of-truth owner, host compatibility strategy, install/discovery contract, method-pack/runtime-core boundary, runtime-ready artifact boundary, evidence model, retained fallback, adapter, compatibility path, duplicate owner, retirement schedule, accepted architecture-scoped Implementation Drift, and release/distribution strategy that future contributors would otherwise misread.

ADR Backfill Check:
- Trigger: yes | no
- Suggested action: create | amend | supersede | skip
- Evidence source:
- Baseline sync: needed | not-needed | unknown
- Skip reason:
- Boundary: advisory method-pack signal only

If the suggested action is create, amend, or supersede, or if Baseline sync is needed or unknown, use recording-architecture-decisions for the ADR lifecycle and Baseline Sync Closure before making the final completion claim. This keeps verification-before-completion as the completion owner while delegating the ADR/baseline writeback decision to the dedicated skill.

15. Governance Closure: for governance/cleanup/migration/compatibility/retirement work → final response must include. Do not skip this structure just because the implementation was small. Localize section labels and prose to the user's language; keep internal concepts in English only when they are product terms or file/path identifiers.

Repair Track: repaired object | action | impact | verification
Retirement Track: retired object | action | retained boundary | future trigger
Residual Risk: unverified | deferred

For work that adds, replaces, or retains old logic, also make the delete-first closure explicit:

Retirement Closure:
- Old logic located:
- Deleted:
- Retained:
- Retention reason:
- Retirement trigger:
- Lingering references checked:

If the work retires old logic, chooses between delete-first and compat retention, or touches source-of-truth deletion boundaries, also include:

Anti-Entropy Declaration:
- Deletion Class:
- Source-of-Truth Data Risk:
- User Confirmation Required:

If User Confirmation Required: yes, completion cannot be claimed until the workflow stops at a guard shaped like:

Data Destruction Guard:
- Exact Target(s):
- Blocked Destructive Steps:
- Confirmation Required: yes
- Status: awaiting scoped confirmation

Mentioning a warning or destructive rule never authorizes execution. Broad assent such as "OK" or "continue" is not scoped confirmation. If persistent-state deletion or another irreversible source-of-truth action happened without explicit scoped confirmation, report the task as not complete.

Red Flags - QA Drift

• Reporting "done" when only one layer was checked
• Treating agent success as equivalent to independent verification
• Forgetting to mention residual risk or uncovered scope
• Saying "verified" when the command was narrow but the claim is broad
• Presenting method-pack verification as if it grants final authority
• Adding new verification branches without saying what old check or fallback now retires
• Closing governance or retirement work without Repair Track, Retirement Track, and Residual Risk
• Claiming completion after growing a core file or complex block without a Complexity Delta, Complexity Governance Suggestion, or residual-risk note
• Retaining old logic without a Retention reason, Retirement trigger, and lingering-reference check
• Treating a destructive warning or guard card as permission to execute
• Treating generic assent as confirmation for irreversible deletion