Verify Task

Verify the task contract independently from the current repository and your own command evidence.

Preconditions and inputs

• Work in a fresh reviewer context separate from implementation, read-only except build/test/lint commands. Report BLOCKED if reviewer independence or required inputs are missing.
• Required inputs: task-file path and implementation report. Read the branch's exploration log when present.
• Treat the report and exploration notes as leads; establish every conclusion from the repository and command output. Return failures for correction with the evidence and gaps.

Process

Create this TODO verbatim and mark each item done as it completes:

TODO:

- [ ] Phase 1: DONE WHEN verification (conditions + scope + standards + build/test/lint)
- [ ] Phase 2: Adversarial review (fault injection, contracts, regression, security)
- [ ] Report verdict (PASS/FAIL with evidence)

Establishing the change surface

Use git status --short and git diff --stat HEAD to establish the complete working-tree change list, including files omitted from the implementation report. Read untracked files directly. Inspect git diff HEAD -- <file> for acceptance conditions and review leads; use the full diff only when its total churn is small.

Phase 1: DONE WHEN Verification

• Read DONE WHEN, VERIFY, and all available scope/contract sections. For investigation files, Regression Test supplies DONE WHEN and Proposed Fix supplies STEPS.
• Prove every DONE WHEN condition with file:line evidence or the required command result. Mark an unprovable condition FAIL with its gap.

Scope and constraints

Compare the full status list against the task:

• STANDARD/EPIC permits CONTEXT files, DOC UPDATE paths, and co-located or covering tests. CONTEXT docs are read-only references.
• SIMPLE permits the changes implied by TASK and their tests.
• The branch's docs/tasks/{branch-slug}/exploration.md permits append-only findings.

Flag every out-of-scope change, NON-GOALS implementation, or unsatisfied CONSTRAINTS line.

Standards

Load each applicable standard and apply its checklist to the diff:

Commands and QA

• Run configured build/test/lint from docs/PROJECT_CONTEXT.md, then task VERIFY commands in order. Require project exit 0 and every task-specific expected output/status; a mismatch is FAIL.
• Reuse results you produced in this run only for identical commands with confirmed matching working directory, environment, and unchanged state. Apply every condition's expectations to reused evidence.
• Execute explicit repeats and stateful sequences; rerun when equivalence is uncertain.
• Report unconfigured project commands (empty, placeholder, N/A) as skipped with a warning.
• When QA mode is active, verify the QA file against skills/task-generation/templates/qa-files.md.

Phase 2: Adversarial Review (per-task)

After Phase 1 passes, examine the task's changed paths. Every Phase 2 finding makes the verdict FAIL.

2.1 Fault Injection (mental fuzzing)

Probe each changed method/path with null, empty, and whitespace inputs; zero, negative, maximum, and collection boundaries; and failed API, database, or file operations. Identify incorrect results and swallowed failures.

2.2 Implicit Contract Changes

Trace changed return types, nullability, signatures, and removed/renamed symbols through their callers. Flag broken callers and dead references. Confirm INTERFACES Produces names, parameters, and return types in the resulting API/tree, including pre-existing or generated signatures; confirm Consumes is used as specified.

2.3 Regression Risk

Inspect shared paths, existing tests beyond the task's scope, and integration points that depend on prior behavior.

2.4 Security Quick Check

For auth, data handling, or user-input changes, examine trust boundaries for injection, unauthorized access, and leaked internals in errors.

Output

Return PASS only with evidence for every condition, applicable standard, configured command, and adversarial check. Return FAIL with gaps when an obligation fails, then hand off for correction.

PASS

VERIFY: PASS

Phase 1 - DONE WHEN:
- [ ] DONE WHEN condition 1 - verified at file:line (diff-checkable)
- [ ] DONE WHEN condition 2 - verified by command output (command-checkable)
- [ ] Scope clean - STANDARD/EPIC: only CONTEXT + DOC UPDATE + test files (+ exploration.md); SIMPLE: only files implied by TASK + tests (+ exploration.md)
- [ ] Contract clean - CONSTRAINTS satisfied, nothing under NON-GOALS touched, INTERFACES `Produces` present
- [ ] Standards gates - all applicable gates passed
- [ ] Build passes - exit code 0 (or skipped - not configured)
- [ ] Tests pass - N tests, 0 failures (or skipped - not configured)
- [ ] Lint passes - 0 errors (or skipped - not configured)

Phase 2 - Adversarial:
- [ ] Fault injection - no unhandled null/empty/boundary cases found
- [ ] Contract changes - no broken callers
- [ ] Regression risk - no shared paths affected without tests
- [ ] Security - N/A or no issues found

FAIL

VERIFY: FAIL

Phase 1 gaps:
- DONE WHEN condition N - NOT MET: <explanation>
- Scope violation - <file> modified but not in CONTEXT + DOC UPDATE + test files
- Standards gate - <specific gate> failed: <details>
- Build/test/lint - <which> failed: <error>

Phase 2 findings:
- Fault injection - <what breaks with null/empty input at file:line>
- Contract change - <caller at file:line not updated>
- Regression risk - <shared path at file:line affected>
- Security - <issue at file:line>

Handoff

• Report: your verdict - VERIFY: PASS or VERIFY: FAIL with gaps.
• TODO update: PASS → none. FAIL → Fix - dispatch execute-task subagent, then Re-verify - dispatch verify-task subagent.