| name | aget-review-handoff |
| description | Structured delegation review — compare agent output against delegation criteria, accept or reject with rationale |
| archetype | supervisor |
| allowed-tools | ["Read","Glob","Grep"] |
aget-review-handoff
Review a delegated task's output against the original delegation criteria. Produces a structured accept/reject decision with rationale. Designed for formulaic, repeatable reviews — not deep analysis.
When to Use
- Post-delegation: Agent reports task complete — review before accepting
- Milestone check: Mid-task checkpoint against delegation criteria
- Handoff acceptance: Agent transferring work to another agent — verify completeness
- Session boundary: Reviewing work done during a session before wind-down
Input
/aget-review-handoff # Interactive: prompt for delegation + output
/aget-review-handoff <delegation-ref> # Review specific delegation by reference
Required context (provided by user or read from artifacts):
- Delegation: The original task assignment (what was asked, acceptance criteria, constraints)
- Output: The agent's deliverables (files created, changes made, reports generated)
Instructions
When this skill is invoked:
-
Read Delegation
- Load the original delegation (from session file, issue, or user-provided context)
- Extract: objective, acceptance criteria, constraints, expected deliverables
- If delegation has no explicit criteria, note "implicit criteria — review may be subjective"
-
Read Output
- Identify what the agent produced (files, commits, reports, session notes)
- Map each deliverable to the delegation's expected outputs
- Note any undelivered items or unexpected additions
-
Compare Against Criteria
- For each acceptance criterion: PASS / PARTIAL / FAIL with evidence
- For each constraint: RESPECTED / VIOLATED with evidence
- For each expected deliverable: DELIVERED / MISSING / MODIFIED
-
Score (Optional)
- Process: Did the agent follow the expected workflow? (X/10)
- Content: Does the output meet quality standards? (X/10)
- Strategic: Does the output advance the broader goal? (X/10)
- Scoring is optional — criteria comparison (step 3) is sufficient for accept/reject
-
Decision
- ACCEPT: All criteria met, no constraint violations
- ACCEPT WITH CONDITIONS: Minor gaps, list conditions for full acceptance
- REJECT: Criteria not met or constraints violated, provide specific rationale and remediation guidance
Output Format
## Handoff Review: [Delegation Reference]
**Agent**: [agent-name]
**Delegation Date**: [YYYY-MM-DD]
**Review Date**: [YYYY-MM-DD]
---
### Criteria Assessment
| # | Criterion | Status | Evidence |
|---|-----------|--------|----------|
| 1 | [criterion from delegation] | PASS | [what satisfies it] |
| 2 | [criterion from delegation] | PARTIAL | [what's missing] |
| 3 | [criterion from delegation] | FAIL | [why it fails] |
### Constraint Check
| Constraint | Status | Notes |
|------------|--------|-------|
| [constraint 1] | RESPECTED | — |
| [constraint 2] | VIOLATED | [how] |
### Deliverables
| Expected | Status | Location |
|----------|--------|----------|
| [deliverable 1] | DELIVERED | [path/ref] |
| [deliverable 2] | MISSING | — |
### Scoring (Optional)
| Dimension | Score | Notes |
|-----------|-------|-------|
| Process | [X/10] | [workflow adherence] |
| Content | [X/10] | [quality assessment] |
| Strategic | [X/10] | [goal alignment] |
---
### Decision: [ACCEPT / ACCEPT WITH CONDITIONS / REJECT]
**Rationale**: [1-2 sentences explaining the decision]
**Conditions** (if ACCEPT WITH CONDITIONS):
1. [condition 1]
2. [condition 2]
**Remediation** (if REJECT):
1. [specific action to address failure]
2. [specific action to address failure]
Implementation Guidance
Criteria Extraction Pattern
When reading a delegation, extract criteria from these common patterns:
CRITERIA_PATTERNS = [
"acceptance criteria:",
"success looks like:",
"done when:",
"must:",
"deliverables:",
"V-test:",
]
Decision Logic
def decide(criteria_results: list, constraint_results: list) -> str:
"""Determine accept/reject from criteria and constraint results."""
has_fail = any(c["status"] == "FAIL" for c in criteria_results)
has_violation = any(c["status"] == "VIOLATED" for c in constraint_results)
has_partial = any(c["status"] == "PARTIAL" for c in criteria_results)
if has_fail or has_violation:
return "REJECT"
if has_partial:
return "ACCEPT WITH CONDITIONS"
return "ACCEPT"
Implicit Criteria Handling
When delegation lacks explicit criteria (common in informal delegations):
| Signal | Inferred Criterion |
|---|
| "Fix the bug in X" | Bug no longer reproducible |
| "Update docs for Y" | Docs reflect current Y behavior |
| "Create Z" | Z exists and is functional |
| "Review A" | Review comments provided with rationale |
Flag implicit criteria in the review: "Note: criteria inferred from delegation wording — no explicit acceptance criteria provided."
Constraints
- C1: Read-only — NEVER modify delegation artifacts or agent output during review
- C2: Evidence-based — every PASS/FAIL must cite specific evidence (file, line, commit)
- C3: Scope-bounded — only review against stated criteria, don't expand scope during review
- C4: Formulaic — follow the template structure; consistency across reviews matters more than prose quality
Relationship to Other Skills
| Skill | Role | When |
|---|
| aget-review-handoff (this) | Criteria-based accept/reject | Post-delegation, milestone, handoff |
| aget-review-agent | Full agent assessment | Periodic review |
| aget-review-project | PROJECT_PLAN progress | Mid-flight assessment |
| aget-broadcast-fleet | Communicate review result | After accept/reject decision |
Traceability
| Link | Reference |
|---|
| Evidence | Supervisor 1 (3-dimensional scoring, formulaic), Supervisor 2 (criteria comparison, accept/reject) |
| Design | Criteria-based core (Sup 2) with optional scoring overlay (Sup 1) — reconciles both approaches |
| Spec | TBD (to be created with skill proposal) |
| Ontology | Delegation, Handoff, Work_Assignment (ONTOLOGY_supervisor.yaml) |
| Pattern | SOP_fleet_coordination.md (Output Review procedure) |
| CAP | CAP-SUP-001 (Work Delegation and Review) |