Delegation Eval Loop
Generator–evaluator: the child session generates (implements, researches, edits); the parent session evaluates and alone decides acceptance. Never treat the child's self-reported success as the final grade.
For spawn, isolation, workspace inheritance, and tool naming, follow delegate first. This skill owns briefing contracts, layered grading, reject/re-steer, and bounded retries.
Core Rules
- Evidence over assertion: "all tests pass" from the child is a hypothesis. Run or inspect deterministic verification in the parent (or from parent-visible raw output).
- Layered grading order: Deterministic → Invariant → Trajectory/reliability → Semantic. Fail fast on earlier layers.
- Sprint contract before spawn: Negotiate objective, authorized paths, and checkbox acceptance criteria in the brief. Do not invent pass criteria after the child returns.
- Targeted feedback: On reject, send exact failure output via
agent__messageToSession — not "try again".
- Bounded loop: Cap at 3–5 cycles. Then escalate to the user with diagnostics; do not spin forever.
- Reliability hard-fail: Incomplete, cancelled, timed-out, circuit-broken, or evidence-free runs fail evaluation even if the prose claims completion.
The 4-Layer Evaluation Stack
| Layer |
Check |
Verification Method |
Action on Failure |
| 1. Deterministic |
Build, types, tests, lint |
Parent runs the agreed commands (or requires raw exit-0 output in evidence) |
Reject with compiler/test log |
| 2. Invariant Gate |
Path/policy boundaries |
git status / git diff --stat vs authorized paths |
Reject; force revert of forbidden edits |
| 3. Trajectory / reliability |
Real execution, no fake done |
Result text + actions: missing cmds, loops, cancel/incomplete signals |
Reject; demand missing evidence or stop |
| 4. Semantic |
User intent, edge cases |
Parent LLM vs original requirements (only after 1–3 pass) |
Reject with specific missed requirements |
Composite cheerleading is forbidden: Layer 1 pass does not override Layer 2/3 failure.
Workflow
1. Brief with a Sprint Contract
When calling agent__startSession (or assigning via agent__messageToSession), include:
- Exact Objective — concrete deliverable
- Authorized paths / FORBIDDEN paths — invariant gate inputs
- Acceptance Criteria — checkbox list of commands/checks that must pass (
- [ ])
- Deliverable Channel — status + raw evidence in final text (not child scratchpad)
See handoff-templates.md.
2. Poll and Capture Child Result
agent__checkSession until terminal idle/error (or use wait=true when appropriate)- Capture final text. If the session errored, cancelled, or returned empty/incomplete diagnostics → Layer 3 fail; do not soft-accept.
3. Execute Layered Evaluation
Do not present the result to the user until layers pass (or you escalate).
- Deterministic checks in the child's effective workspace (see Metadata
workspace:)
- Invariant gates (
git status / scope)
- Trajectory & reliability (claims vs evidence; incomplete/circuit-break)
- Semantic review only if 1–3 pass
Details: evaluation-protocol.md.
4. Reject & Re-steer
If any layer fails:
- Increment iteration counter (abort if over max, typically 4)
- Structured correction: which layer, exact error snippet, single next action
agent__messageToSession(sessionId, message)- Return to Step 2
Templates: handoff-templates.md.
5. Final Acceptance
When all layers pass, synthesize for the user with verified evidence (commands run, scope clean). Mark which criteria were checked.
1---2name: delegation-eval-loop3description: Generator-evaluator loop for delegated sub-agent work: parent evaluates, child generates. Use layered verification (Deterministic, Invariant Gate, Trajectory/reliability, Semantic), sprint contracts with checkbox acceptance criteria, and targeted re-steers via agent__messageToSession. Use when briefing with strict acceptance criteria, evaluating child results from agent__checkSession, preventing premature "done" without proof, or when delegate routes here for high-stakes handoffs. Triggers: generator-evaluator, generator evaluator, delegation eval, subagent evaluation, verify subagent, delegation loop, eval loop, proof before done, acceptance criteria, delegation-eval-loop.4---56# Delegation Eval Loop78**Generator–evaluator:** the child session **generates** (implements, researches, edits); the **parent** session **evaluates** and alone decides acceptance. Never treat the child's self-reported success as the final grade.910For spawn, isolation, workspace inheritance, and tool naming, follow **`delegate` first**. This skill owns briefing contracts, layered grading, reject/re-steer, and bounded retries.1112## Core Rules13141. **Evidence over assertion**: "all tests pass" from the child is a hypothesis. Run or inspect deterministic verification in the parent (or from parent-visible raw output).152. **Layered grading order**: Deterministic → Invariant → Trajectory/reliability → Semantic. Fail fast on earlier layers.163. **Sprint contract before spawn**: Negotiate objective, authorized paths, and checkbox acceptance criteria in the brief. Do not invent pass criteria after the child returns.174. **Targeted feedback**: On reject, send exact failure output via `agent__messageToSession` — not "try again".185. **Bounded loop**: Cap at 3–5 cycles. Then escalate to the user with diagnostics; do not spin forever.196. **Reliability hard-fail**: Incomplete, cancelled, timed-out, circuit-broken, or evidence-free runs fail evaluation even if the prose claims completion.2021## The 4-Layer Evaluation Stack2223| Layer | Check | Verification Method | Action on Failure |24|---|---|---|---|25| **1. Deterministic** | Build, types, tests, lint | Parent runs the agreed commands (or requires raw exit-0 output in evidence) | Reject with compiler/test log |26| **2. Invariant Gate** | Path/policy boundaries | `git status` / `git diff --stat` vs authorized paths | Reject; force revert of forbidden edits |27| **3. Trajectory / reliability** | Real execution, no fake done | Result text + actions: missing cmds, loops, cancel/incomplete signals | Reject; demand missing evidence or stop |28| **4. Semantic** | User intent, edge cases | Parent LLM vs original requirements (only after 1–3 pass) | Reject with specific missed requirements |2930Composite cheerleading is forbidden: Layer 1 pass does **not** override Layer 2/3 failure.3132## Workflow3334### 1. Brief with a Sprint Contract3536When calling `agent__startSession` (or assigning via `agent__messageToSession`), include:3738- **Exact Objective** — concrete deliverable39- **Authorized paths / FORBIDDEN paths** — invariant gate inputs40- **Acceptance Criteria** — checkbox list of commands/checks that must pass (`- [ ]`)41- **Deliverable Channel** — status + raw evidence in **final text** (not child scratchpad)4243See [handoff-templates.md](references/handoff-templates.md).4445### 2. Poll and Capture Child Result4647- `agent__checkSession` until terminal idle/error (or use `wait=true` when appropriate)48- Capture final text. If the session errored, cancelled, or returned empty/incomplete diagnostics → **Layer 3 fail**; do not soft-accept.4950### 3. Execute Layered Evaluation5152Do **not** present the result to the user until layers pass (or you escalate).53541. Deterministic checks in the child's effective workspace (see Metadata `workspace:`)552. Invariant gates (`git status` / scope)563. Trajectory & reliability (claims vs evidence; incomplete/circuit-break)574. Semantic review only if 1–3 pass5859Details: [evaluation-protocol.md](references/evaluation-protocol.md).6061### 4. Reject & Re-steer6263If any layer fails:64651. Increment iteration counter (abort if over max, typically 4)662. Structured correction: which layer, exact error snippet, single next action673. `agent__messageToSession(sessionId, message)`684. Return to Step 26970Templates: [handoff-templates.md](references/handoff-templates.md).7172### 5. Final Acceptance7374When all layers pass, synthesize for the user with verified evidence (commands run, scope clean). Mark which criteria were checked.
fritzprix