| name | dev-review-loop |
| scope | emploke |
| description | Strategy skill for emploke/coordinator — the dev → review+designer iterate-to-clean orchestration: case bank, brief templates, placeholder resolution, stop condition, failure-mode coverage |
| version | 1.0.0 |
Emploke Development-Loop Strategy Skill
Strategy: dispatch a single emploke/dev worker; on success, fan out to parallel emploke/review + emploke/frontend-designer reviewers; finish succeeded if both verdicts come back clean (APPROVE with at most minor findings), else dispatch the next emploke/dev iteration with prior verdicts available, and loop. Loaded by the emploke/coordinator agent alongside the generic emploke/coordinator skill at every coord wake-up.
Case bank
Match own direct parents against these cases — exactly one matches per wake-up (the case bank is total over expected shapes; see "Failure-mode coverage"). Unexpected shapes (3 parents, an unknown agent FQN, etc.) are bugs in workflow construction; terminate with workflow finish --outcome failed --message "coord saw unexpected DAG shape under emploke/dev-review-loop: <describe>".
CASE "no parents" (initial coord node):
addSubgraph:
dev (worker, agent=emploke/dev, brief=<template-dev-iter-1>)
parents = [self]
next-coord (coordinator, agent=emploke/coordinator)
parents = [dev]
exit
CASE "one parent, worker, agent=emploke/dev, status=succeeded":
addSubgraph:
review (worker, agent=emploke/review, brief=<template-review>)
parents = [self]
designer (worker, agent=emploke/frontend-designer, brief=<template-designer>)
parents = [self]
next-coord (coordinator, agent=emploke/coordinator)
parents = [review, designer]
exit
CASE "one parent, worker, agent=emploke/dev, status in (failed, cancelled)":
finishWorkflow(failed, "dev iteration ended in {status}")
exit
CASE "two parents, both worker, agents in {emploke/review, emploke/frontend-designer}":
for each parent:
fetch task: emploke task show --tid <parent.taskId> --json
fetch verdict: read <task-workdir>/verdict.json (parse per §C of generic skill)
blockers_and_majors = [
f for v in verdicts for f in v.findings if f.severity in ('blocker', 'major')
]
if blockers_and_majors is empty:
finishWorkflow(succeeded, summary={
"iterations": <count of dev nodes in DAG>,
"minor_findings_remaining": <count if any>
})
else:
addSubgraph:
dev (worker, agent=emploke/dev, brief=<template-dev-iter-2-plus>)
parents = [self]
next-coord (coordinator)
parents = [dev]
exit
CASE "two parents, both worker, any status in (failed, cancelled)":
finishWorkflow(failed, "reviewer iteration ended in {status}; coord cannot decide without verdict")
exit
Use the §B "Batch-mutate the DAG atomically" add-subgraph payload shape from the generic skill; substitute <self-node-id> with the actual id from the DAG snapshot.
Brief templates
Coord at dispatch time substitutes ${PLACEHOLDER} slots per the resolution table below; do not paraphrase template prose — workers receive these as their primary contract.
Template: template-dev-iter-1 (the initial dev call)
Implement the feature per the workflow brief.
# Workflow context
- Workflow id: ${WORKFLOW_ID}
- Workflow brief (verbatim from creator):
${WORKFLOW_BRIEF}
- Workflow details (verbatim from creator):
${WORKFLOW_DETAILS}
# Output expectations
Follow your normal dev workflow (branch, PR, etc.). The coord will pick
up your PR via the workflow DAG; no special output protocol on your end.
Template: template-dev-iter-2-plus (after a review round)
Re-implement the feature for iteration ${ITERATION_NUMBER}.
# Workflow context
- Workflow id: ${WORKFLOW_ID}
- Workflow brief (verbatim):
${WORKFLOW_BRIEF}
- Workflow details (verbatim):
${WORKFLOW_DETAILS}
# Prior iteration outputs (you must fetch these yourself)
- Prior review verdict + narrative:
emploke task show --tid ${PRIOR_REVIEW_TASK_ID} --json
then read <task-workdir>/verdict.json and <task-workdir>/review.md
- Prior designer verdict + narrative:
emploke task show --tid ${PRIOR_DESIGNER_TASK_ID} --json
then read <task-workdir>/verdict.json and <task-workdir>/review.md
# What to do
Address every finding marked severity=blocker or major. Apply your own
judgment on severity=minor findings — fix what you'd fix as a
professional.
Keep working on branch ${BRANCH_NAME} (already-pushed prior iteration
commits are visible; rebase / amend as you see fit).
Template: template-review (same every iteration)
Review the latest dev iteration in this workflow.
# Workflow context
- Workflow id: ${WORKFLOW_ID}
- Workflow brief (verbatim):
${WORKFLOW_BRIEF}
# What to review
The dev node immediately preceding you in the workflow DAG. Find it via:
emploke workflow dag --wfid ${WORKFLOW_ID} --json
The dev node is your direct parent. Read dev's task via its taskId, see
what changed, apply your normal review standards.
If this is review iteration 2 or later, fetch the prior review node
(same agent as you, lower phase in the DAG) — read its verdict.json
to confirm previously-flagged findings are now addressed.
# Required output protocol
Write to <task-workdir>/verdict.json:
{
"verdict": "APPROVE" | "REQUEST_CHANGES",
"findings": [
{
"id": "F1", // unique within this verdict
"severity": "blocker" | "major" | "minor",
"summary": "<≤200 chars, single line>",
"detail": "<free-form, any length>"
}
]
}
Validation rules:
- verdict == "APPROVE" ⇒ findings MAY be [] OR contain only "minor" items
- verdict == "REQUEST_CHANGES" ⇒ findings MUST contain ≥1 "blocker" or "major"
- findings[].id must be unique within this verdict
Coord decision rule (so you understand the impact):
- All verdicts APPROVE with only minor findings → workflow finishes succeeded
- Any blocker/major → next dev iteration dispatched, current findings
propagated to next dev brief (it will read your verdict.json itself)
Optionally write <task-workdir>/review.md for free-form narrative. The
next dev iteration will read it for context if produced.
Template: template-designer (same every iteration)
Review the latest dev iteration's UI / UX in this workflow.
# Workflow context
- Workflow id: ${WORKFLOW_ID}
- Workflow brief (verbatim):
${WORKFLOW_BRIEF}
# What to review
The dev node immediately preceding you in the workflow DAG. Find it via:
emploke workflow dag --wfid ${WORKFLOW_ID} --json
Apply your normal frontend / design review standards.
If this is designer iteration 2 or later, fetch the prior designer node
(same agent as you, lower phase) and confirm previously-flagged findings
are resolved.
# Required output protocol
Identical to review's protocol (verdict.json + optional review.md).
See template-review above for the schema and validation rules.
Placeholder resolution table
Plain string replacement; placeholders with no value (e.g. ${WORKFLOW_DETAILS} when the creator passed nothing) substitute the empty string rather than leaving the literal ${...} in the dispatched brief.
| Placeholder | Source | Notes |
|---|
${WORKFLOW_ID} | workflow.id from emploke workflow show | string |
${WORKFLOW_BRIEF} | workflow.brief | verbatim, no rewriting |
${WORKFLOW_DETAILS} | workflow.details (may be null → emit empty string) | verbatim |
${ITERATION_NUMBER} | count of emploke/dev worker nodes already in the DAG, +1 | integer; template-dev-iter-2-plus only |
${PRIOR_REVIEW_TASK_ID} | taskId of the most recent agent=emploke/review worker parent of the prior coord | string; template-dev-iter-2-plus only |
${PRIOR_DESIGNER_TASK_ID} | taskId of the most recent agent=emploke/frontend-designer worker parent of the prior coord | string; template-dev-iter-2-plus only |
${BRANCH_NAME} | branch the prior dev node pushed (derive from prior dev task's terminal result or its <task-workdir>/branch.txt) | string; template-dev-iter-2-plus only |
${PRIOR_*_TASK_ID} lookups use the "Find prior-iter siblings" snippet from the generic skill §B (same agent FQN, lower phase).
Stop condition
Trigger finishWorkflow(succeeded, ...) in the "two parents, both reviewers" case when both verdicts parse cleanly per §C and the union of findings filtered to severity in ('blocker', 'major') is empty. The success summary records iterations (count of emploke/dev nodes in the final DAG) and minor_findings_remaining (visibility — the work ships with them outstanding). No iteration cap in v1; an iteration cap is deferred per coord-design.md §7 and will land as a sibling case in the case bank when added.
Failure-mode coverage
Every (parent role, parent terminal status) cell on every expected parent role matches exactly one case in the case bank — verify here when editing.
| Coord wake-up shape | Parent role | Parent status | Matched case | Action |
|---|
| 0 parents (initial coord node) | — | — | "no parents" | addSubgraph dev + next-coord |
| 1 parent | emploke/dev worker | succeeded | "one parent, dev, succeeded" | addSubgraph review + designer + next-coord |
| 1 parent | emploke/dev worker | failed | "one parent, dev, failed/cancelled" | finish(failed, "dev iteration ended in failed") |
| 1 parent | emploke/dev worker | cancelled | "one parent, dev, failed/cancelled" | finish(failed, "dev iteration ended in cancelled") |
| 2 parents | both reviewers (review + designer) | both succeeded AND verdict.json parseable | "two parents, both reviewers" | finish(succeeded) OR addSubgraph next dev iter, per blockers/majors count |
| 2 parents | reviewer | failed | "two parents, any failed/cancelled" | finish(failed, "reviewer iteration ended in failed") |
| 2 parents | reviewer | cancelled | "two parents, any failed/cancelled" | finish(failed, "reviewer iteration ended in cancelled") |
| 2 parents | reviewer | succeeded but verdict.json missing / unparseable | "two parents, both reviewers" → §C parse failure | finish(failed, "reviewer did not produce valid verdict.json") |
