Developer

Developer (orchestrator)

Drives the triage → build → review → fix → merge pipeline across isolated subagent workers, looping over every sub-issue of a spec unattended. Each worker gets a clean context — the only thing it knows is the arguments you pass in its prompt. You (the orchestrator) hold the state between steps.

Invoke

/developer <issue>              # spec with sub-issues → deliver them all
                                # plain issue → deliver just that one
/developer <spec> <subissue>    # deliver a single specific sub-issue

Flags (override the repo defaults — see Run configuration):
  --parallel | --sequential     # spec mode: waves vs one-at-a-time
  --auto-merge | --no-auto-merge  # merge CLEAN PRs vs leave them ready

If no issue number is given, ask for it and stop. Do not guess issue numbers. The execution flags only change spec mode; in single mode they are a no-op. Accept the bare words parallel / sequential as synonyms for the flags.

Namespacing. Installed as a Claude Code plugin, skills and agents carry the plugin prefix: the skills appear as developer-skills:<name> and the subagents as developer-skills:dispatcher / developer-skills:code-author / developer-skills:diff-reviewer. Use the names exactly as they appear in your available-skills and available-agents lists; the short names below refer to whichever form is installed.

Contract docs (tracker + code host)

The pipeline is agnostic about where issues and changes live. Two committed docs define the mechanics for this repo, and every worker reads them in its own context:

• docs/agents/issue-tracker.md — issue operations (read an issue, enumerate children of a parent, check a blocker, comment, label, close) in its ## Delivery operations section.
• docs/agents/code-host.md — change operations (publish, check out in a worktree, review, mark ready, reply, merge, auto-close semantics).

Read both once at the start (they are short — an allowed exception to "never read bodies yourself"). Every command block below shows the GitHub factory default (gh); when a contract doc defines a different mechanic for the same operation, the doc wins. If a doc is missing, the GitHub defaults apply as-is — suggest /setup-developer-skills if that looks wrong.

Note one capability flag from docs/agents/code-host.md: issue auto-close on merge? If not (e.g. issues on a tracker the code host can't close), the orchestrator closes the delivered issue itself per the tracker ops right after verifying the merge.

If either doc says the tracker or the code host is local (files in the repo, no remote), read LOCAL-HOST.md now, before anything else. It holds every standing adjustment a local host or tracker needs — capability overrides, tracker writes, branch discipline, cleanup and wrap-up. A run on a remote host never loads it.

Two more files are read on demand, never at the start: MERGE-FIX.md at the first merge conflict, and WRAP-UP.md when the loop ends.

(All three live next to this SKILL.md, in the skill's own directory — under the plugin root when installed as a plugin, the same place scripts/cleanup-worktrees.sh comes from. They are part of this skill: a step that says to read one is not optional, it is that step's other half.)

Run configuration

Two knobs govern a run. Resolve each one before mode detection, in this precedence order: CLI flag > repo default > factory default.

Knob	Values	Factory default
execution	parallel / sequential	parallel
merge	auto / manual	manual

Repo defaults live in docs/agents/developer-defaults.md, written by /setup-developer-skills. Read it once at the start (it is short — this is an allowed exception to "never read bodies yourself"); if it is missing or a knob is absent, fall back to the factory default. State the resolved configuration in one line before starting, e.g. Run config: execution=parallel, merge=manual (repo defaults).

What merge means:

• auto — a CLEAN verdict triggers the code host's merge operation (Merge step; gh pr merge on GitHub). The committed merge: auto line in docs/agents/developer-defaults.md is the user's standing authorization for these merges.
• manual — the pipeline stops at CLEAN: you already marked the PR ready after the review, so record the sub-issue as ready-to-merge and leave the merge to the human. Because sub-issues only close on merge (Closes #N), anything Blocked by a ready-to-merge sub-issue stays blocked for the rest of the run — expected, not an error; it lands in the wrap-up as the human's queue.

Workers (subagents)

Step	Subagent	Model	Isolation	Skill it runs
triage	dispatcher	sonnet (pinned)	—	(reads the issue)
build	code-author	chosen by triage	worktree	implement-issue
review	diff-reviewer	opus (pinned)	worktree	review-pr
fix	code-author	escalates per cycle	worktree	fix-pr
harvest	code-author	sonnet (pinned)	worktree	(reads PR bodies)

Spawn each via the Agent tool with the matching subagent_type. Pass isolation: "worktree" to every code-author and diff-reviewer spawn. Pass model explicitly to code-author spawns (triage decides the tier). Never run the skills yourself in the main context — the point is isolation.

Every spawn is run_in_background: true, dispatchers included, in both execution modes. A foreground spawn holds your turn open for the worker's whole run, and anything that interrupts that turn — a Ctrl-C, a dropped connection — takes the worker down with it: its context, its worktree and its commits are gone for good. The same interruption leaves a background worker running and still reachable. Sequential mode is not an exception to this: it means wait for this worker's result before spawning the next, not spawn it in the foreground.

Three things at every spawn, each cheap now and each the difference between a resume and a rebuild later:

• Give the Agent tool a description that names the job and the sub-issue — Build #<N>, Review PR #<PR>, Fix #<N> cycle 2. It is how the worker is identified in the agent list and in the user's view of the run.

• Keep the agentId the tool returns for as long as that worker runs: it is the handle that picks the worker back up instead of starting it over (see Resuming the orchestrator). Drop it when its RESULT arrives.

• On a BUILD, FIX or HARVEST spawn — the jobs that hold uncommitted work and are expensive to lose — append a spawn row to this run's log as soon as the tool hands you the agentId, so the job leaves a trace that outlives your context:

  mkdir -p .scratch
  echo "$(date +%F) spec=#<spec> sub=#<N> event=spawned job=<build|fix|harvest> agent=<agentId> model=<tier>" \
    >> .scratch/developer-run-<spec>.log
  

  (sub=none on the harvest, which belongs to the whole run.) Triage and review spawns skip the row entirely: both are cheap to repeat, and pipeline step 0 reconstructs where a PR stands without them. Step 7 writes the other kind of row, the terminal one — a resume reads both, the wrap-up reads only the terminal ones.

Context economy

The loop may cover many sub-issues; your context must survive all of them.

• Never read issue or PR bodies yourself. Workers read them in their own disposable contexts. You only run the cheap listing commands below.
• Never read CI logs yourself either — gh run view --log-failed and anything like it. The Merge step's --json classification is the whole diagnosis the orchestrator gets; the rest is the fixer's, in its context.
• A worker's whole final message is its RESULT line — the agents require it and every spawn prompt below restates it. A worker that reports prose before its line is spending your context, not its own; nothing it says there survives the run, so anything worth keeping belongs on the PR or the issue.
• Track per sub-issue: number, task id, chosen model, PR number, verdict, fix cycles, wave (parallel mode), the agentId of the worker running on it right now (dropped the moment its RESULT arrives), outcome (merged / ready-to-merge / escalated / blocked) — and write the row to the run log the moment the sub-issue goes terminal (delivery pipeline step 7), so the wrap-up reads facts instead of recalling them. Also keep the dispatcher's touches/hints just long enough to forward hints into that sub-issue's Build step — discard both once the build is spawned, they have no use after that.

Resuming the orchestrator

A run outlives your turn: workers keep going in the background, their notifications can arrive late, and sessions get interrupted, compacted or restarted. So while a run is in flight, every prompt that reaches you is a resume — including a bare Continue from where you left off., an empty continuation, or a notification you believe you have already handled. There is no state in which the right answer is "no response requested": either work is pending and you take its next step, or nothing is, and you go to Wrap-up. Silence is the one failure mode this pipeline cannot recover from on its own.

(This is the orchestrator's own resume. Pipeline step 0 is the per-sub-issue one — it is what step B below runs.)

On any such prompt:

1. Rebuild the picture from the three places that survive a dead context, never from recall:

   • the progress board (TaskList) — which sub-issues are in_progress;
   • .scratch/developer-run-<spec>.log — the terminal rows already recorded, and the event=spawned rows naming the worker that was running on each sub-issue still in flight. If a wrap-up already ran this spec, the rows from before it are in .scratch/archive/developer-run-<spec>-*.log;
   • ListAgents — which of those workers are still alive.

2. Recover each non-terminal sub-issue in this order, stopping at the first that works:

   A. Its worker is still alive → SendMessage to its agentId and ask it to report. Its context, its worktree and its commits are all intact, so this continues the job rather than repeating it — by far the cheapest recovery, and the only one that does not throw away work already paid for. A worker that finished while you were not looking answers here too, with the RESULT whose notification you missed.

   This works for code-author. It usually does not for diff-reviewer: a review changes no files, so its worktree is removed when it ends and a reviewer missing from ListAgents is missing for good. Try it if it is listed; otherwise go straight to B — a review is cheap to repeat, a build is not.

   B. Its worker is gone → run pipeline step 0 on that sub-issue exactly as written. It asks the code host rather than your memory, and routes the sub-issue to Review, to the Fix cycle, or back to Triage when nothing was ever opened for it.

3. Re-enter the loop: recompute the unblocked set (spec loop step 1, or the wave in parallel mode) and carry on. Nothing left → Wrap-up.

Say in one line what you recovered and how, before continuing. During an unattended run the board and that line are the user's whole window into it.

Step 0 — Publish context docs before anything else

Workers branch from origin/main, so any domain-context file that is not committed and pushed is invisible to them. Grilling/spec sessions edit these files but do not commit them. Before dispatching any worker:

git status --porcelain -- CONTEXT-MAP.md '**/CONTEXT.md' docs/adr docs/agents AGENTS.md CLAUDE.md

If anything shows up, stage only those paths (never the user's unrelated work-in-progress), commit on the current branch (must be main — if not, stop and tell the user), and push:

git add CONTEXT-MAP.md '**/CONTEXT.md' docs/adr docs/agents AGENTS.md CLAUDE.md
git commit -m "docs(domain): publish context map and ADR updates"
git push origin main

If the push is rejected, stop and report — do not rebase or force anything. This is the flow's start, before going unattended; the user is still there to resolve it.

Mode detection

Enumerate the children of the given issue per the tracker's Delivery operations. GitHub default — native sub-issues (infer OWNER/REPO from git remote -v):

gh api graphql -f query='
{
  repository(owner:"OWNER", name:"REPO") {
    issue(number: N) {
      subIssues(first: 50) {
        pageInfo { hasNextPage }
        nodes { number title state labels(first: 10) { nodes { name } } }
      }
    }
  }
}' --jq '.data.repository.issue.subIssues'

If hasNextPage is true, stop and report: a spec with more than 50 sub-issues is not sized for this pipeline — tell the user to split it and end the run. Never proceed on the first page alone: delivering 50 of 60 while reporting the spec complete is a silent failure, the one outcome worse than stopping.

Keep each sub-issue's labels from this query — the pick reads them (spec loop step 1). Where a tracker's enumeration carries no labels, get them per its read-labels operation instead.

(Throughout this skill, #<N> stands for the issue ref in the tracker's own format — a number on GitHub/GitLab, a file path on a local tracker — and #<PR> for the change ref in the code host's format.)

• Open sub-issues exist → spec mode: loop over all of them (below).
• No sub-issues → single mode: run the delivery pipeline once on the given issue, with the issue itself as spec (no separate parent spec).
• Two arguments given: run the delivery pipeline once on <subissue> with <spec> as the spec. Skip the loop. If the sub-issue ends merged (verified CLOSED), read WRAP-UP.md and run its Close the spec step (step 4) afterwards — it may have been the spec's last open sub-issue. That one step is all this mode needs from the wrap-up.

Progress board (spec mode — not optional)

The user follows the run through the harness task list. Keep it faithful at every transition; a stale board defeats its purpose.

1. Immediately after mode detection, create one task per open sub-issue with TaskCreate, in sub-issue order: subject #<N> <title>, activeForm Delivering #<N>. The whole plan must be on the board before the first worker spawns.
2. When the delivery pipeline starts on a sub-issue → TaskUpdate status: in_progress. In parallel mode every wave member goes in_progress as its build spawns, so the board shows exactly what is running concurrently.
3. Terminal transitions, the moment they happen:
   • merged (sub-issue verified CLOSED) → status: completed.
   • ready-to-merge (merge: manual, verdict CLEAN) → back to status: pending and rename the subject to #<N> <title> — ready to merge: PR #<PR>. Not completed — the human still has to merge it.
   • escalated → back to status: pending and rename the subject to #<N> <title> — escalated: <one-line reason>. Never mark an escalated sub-issue completed — unchecked items at the end are the human's queue.
4. Sub-issues that never became deliverable (blocked by an escalated one, or by a ready-to-merge one the human hasn't merged yet) stay pending; rename them #<N> <title> — blocked by #<M> at wrap-up.

Single mode (no sub-issues) skips the board.

Spec loop

Repeat while open sub-issues remain:

1. Pick the next unblocked sub-issue: for each open sub-issue (lowest number first), check its blockers without reading full bodies — the "check a blocker's state" operation from the tracker doc. Blockers may be wired as the tracker's native dependency links, as a Blocked by body section, or both (/to-tickets prefers native edges where the tracker has them) — check both. GitHub default:

   # native dependencies: count of OPEN blockers (0 or absent = clear)
   gh api repos/{owner}/{repo}/issues/<N> --jq '.issue_dependencies_summary.blocked_by // 0'
   # body fallback: every blocker listed in the section must be CLOSED
   gh issue view <N> --json body --jq '.body' \
     | awk '/^##[#]* *[Bb]locked by/{f=1;next} /^#/{f=0} f'
   gh issue view <BLOCKER> --json state --jq '.state'
   

   Extract the Blocked by section, never a fixed window around the heading: a grep -A<n> reads the wrong number of lines by construction — it drops the fifth blocker of a list of six and swallows the first lines of whatever section follows a list of two.

   Take the first open sub-issue whose blockers are all closed, and:

   • Skip any sub-issue carrying the ready-for-human triage label (the repo's own string for that role if docs/agents/triage-labels.md maps it differently) — from the labels the enumeration returned, plus the ones you applied yourself while escalating this run. That label is the escalation gate: someone already gave up on this sub-issue, and picking it up again buys three more fix cycles against the same wall. The gate is symmetric and it is the whole mechanism: removing the label re-queues the sub-issue, there is no other state to reset.
   • Whatever a gated sub-issue blocks stays blocked, as with any open one.
   • With merge: manual, sub-issues you already delivered as ready-to-merge count as done for your loop but their dependents stay blocked — skip both.

2. Run the delivery pipeline on it.

3. On merged or ready-to-merge → next iteration. On escalated/blocked → record it, next iteration.

4. When no deliverable sub-issue remains (all closed or ready-to-merge, or the rest are blocked by escalated/unmerged ones) → wrap-up.

Parallel mode (execution: parallel)

Parallel is the factory default (see Run configuration). The trade-off: sequential with merge: auto delivers one sub-issue fully before the next starts, so each PR branches from a main that already contains the previous one — no merge conflicts by construction. Parallel trades that guarantee for throughput: independent sub-issues are built concurrently, and conflicts between their PRs become expected work, resolved by extra merge-fix jobs. Note that with merge: manual sibling PRs all branch from the same main regardless of execution mode — sequential buys no conflict guarantee there, so parallel costs nothing extra.

Work in waves:

1. Wave = every open sub-issue whose blockers are all closed (same check as step 1 of the spec loop), minus the ones that step's gate excludes — ready-for-human above all, whether this run applied it or an earlier one did.

2. Run the delivery pipeline on each wave member concurrently, entry points first: the pipeline's step 0 resolves where each member starts, and only the ones with no open change get a dispatcher and a build. Spawn those dispatchers in one batch, then their code-author BUILD jobs in parallel (each in its own worktree, run_in_background: true) — minus any member its dispatcher scored oversized, which is escalated instead of built and simply leaves the wave. A resumed member goes straight into the review or fix stage alongside them. As each build reports its PR, spawn its diff-reviewer; as each reviewer reports, mark that PR ready (step 3 of the pipeline); fix cycles run per PR exactly as in the sequential pipeline. Cap concurrent build/review/fix workers at 3; queue the rest of the wave.

3. Merges stay strictly serial — never merge two PRs concurrently. With merge: auto, merge each PR as it reaches CLEAN. After every successful merge, refresh the wave's still-open PRs per the code host's update-branch operation — GitHub default, per open sibling:

   gh pr update-branch <PR>
   

   It is a remote operation — no local git. Each sibling branched from a main that did not contain this merge; left stale, its CI goes red for synchronization, not for a bug, and a full fix cycle ends up doing what this one call does. A sibling whose update fails on a conflict is left alone — its own Merge step's conflict path (MERGE-FIX.md) already owns that case.

   Every PR in the wave branched from the same main, so any PR merged after the first may conflict: on merge failure after the Merge step's checks gate passed, run the merge-fix job (MERGE-FIX.md) and retry once. With merge: manual there is nothing to serialize — each CLEAN PR just becomes ready-to-merge.

4. When every wave member is delivered (merged, ready-to-merge, or escalated), recompute the unblocked set → next wave. None left → wrap-up.

Everything else — context economy, escalation, wrap-up, rules — is unchanged.

Delivery pipeline (per sub-issue)

0. Entry point — resume, never rebuild

(Pipeline step 0, not the top-level Step 0 that publishes the context docs. This is the per-sub-issue resume, reached both on a fresh run and as step B of Resuming the orchestrator.)

A run can die at any point — a dead session, a compaction, a Ctrl-C — and the sub-issues it half-delivered are still open, so re-running /developer <spec> picks them right back up. What the tracker forgets is how far each one got: build from scratch again and you get a second PR for the same sub-issue and a second review paying for it. So before triaging, ask the code host whether a change already exists for this sub-issue, per its "open change for this issue" operation. GitHub default:

gh pr list --state open --search '"Closes #<subissue>" in:body' --json number,isDraft

• No open PR → nothing to resume: step 1 (Triage).
• One open PR, no unresolved review threads → the build landed but the review did not: keep its <PR>, skip Triage and Build, start at step 3 (Review).
• One open PR with unresolved review threads → a review landed and its fixes did not: start at step 4 (Fix cycle), counting from cycle 1 with the fixer at opus (the build tier died with the session that chose it).
• More than one open PR matches → escalate: two open changes for one sub-issue is a human's call, never a pick.

GitHub default for the unresolved-thread count:

gh api graphql -f query='
{
  repository(owner:"OWNER", name:"REPO") {
    pullRequest(number: <PR>) {
      reviewThreads(first: 50) { nodes { isResolved } }
    }
  }
}' --jq '[.data.repository.pullRequest.reviewThreads.nodes[]
          | select(.isResolved == false)] | length'

The fix-cycle budget starts fresh on a resume: a PR that already burned cycles in the dead run gets three more here. That is deliberate — the alternative is reconstructing a counter nothing ever recorded — and the ready-for-human gate is what stops a sub-issue looping forever across runs.

1. Triage

Spawn dispatcher with run_in_background: true:

Triage issue #<subissue>. Score its implementation complexity per your rubric. Your entire final message must be the RESULT complexity=… model=… touches=… hints=… reason=… line — nothing before it, nothing after it.

Parse complexity= first:

• complexity=oversized → the sub-issue does not fit in a single fresh context window. Do not build it. No model tier rescues a ticket that does not fit: the builder runs out of room, the review finds half a feature, and three fix cycles burn against the same wall. Go straight to Escalation, quoting the dispatcher's hints= — they carry the fault lines — and move to the next sub-issue. No BUILD, no review, no fix cycles are spent on it.
• Anything else → parse model=<tier> and continue to Build.

On any malformed result, default to opus and build — a line you cannot parse is not an oversized verdict. Parse touches= and hints= too, defaulting each to none if the line predates these fields or omits them — never block the pipeline on a missing hint.

2. Build

Spawn code-author with model: <tier>, isolation: "worktree" and run_in_background: true, then log the spawn row (Workers):

BUILD job. Spec issue #<spec>, sub-issue #<subissue>. Run the implement-issue skill on the sub-issue. The sub-issue's ## Spec extract section carries the spec decisions that apply to it — read the full spec issue only if that section is missing. Triage found: <dispatcher's hints, verbatim, or omit this line entirely when hints=none>. Your entire final message must be the RESULT pr=… url=… line — no summary before it, nothing after it. Whatever deserves a record goes in the PR body, not in your reply. Report only a PR number you have confirmed exists.

• RESULT blocked … → escalate (see below) and move to the next sub-issue.

• RESULT pr=<PR> url=<URL> → confirm the PR exists, then keep <PR> and continue:

  gh pr view <PR> --json number,state,headRefName
  

  Never skip it. A build has reported pr=<N> for a number the host 404s on, with its whole implementation sitting uncommitted in its worktree because the publish step never ran at all. A worker's RESULT is a claim; this is the one cheap command that turns it into a fact, and the only thing standing between a fabricated line and a reviewer sent after a PR that was never opened. (Change metadata is read per docs/agents/code-host.md on another host.)

  On a 404, do not re-spawn the build — the work is almost certainly intact in the worker's worktree. Recover the worker per Resuming the orchestrator step A: its spawn row holds the agentId, ListAgents says whether it is still alive, and SendMessage tells it what you found and to run its publish step for real, reporting only a number it has verified. If the worker is gone, escalate naming its branch and worktree, so nobody rebuilds on top of work that still exists.

3. Review

Spawn diff-reviewer with isolation: "worktree" and run_in_background: true:

Review PR #<PR> by running the review-pr skill on it — its step 1 plus the repo's docs/agents/code-host.md give the exact checkout procedure for your worktree; follow them, not memory. Post the review (inline comments + summary) as a single COMMENT submission — never an approval event — and do not mark the PR ready or merge; those are orchestrator steps. Your entire final message must be the RESULT verdict=… line — the review itself is your output, your reply is not.

Then mark the PR ready yourself, whatever the verdict — per the code-host doc's mark-ready operation. GitHub default:

gh pr ready <PR>

Skip it whenever the PR is already ready — a re-review, or a resumed run whose step 0 found isDraft: false.

• verdict=CLEAN → go to Merge.
• verdict=NEEDS_FIXES → enter the fix cycle.
• RESULT blocked because the change branch is held by another worktree (the worker quotes git's "already used by worktree" error) → a previous worker's worktree wasn't cleaned: run Cleanup (step 6) and re-spawn the reviewer, once per sub-issue — if it blocks again, escalate.
• Any other RESULT blocked or malformed result → escalate, next sub-issue.

4. Fix cycle (max 3)

For cycle c = 1, 2, 3:

1. Fixer model: cycle 1 uses the build tier, each later cycle escalates one tier (sonnet → opus; opus stays opus). A sub-issue resumed straight into this step has no build tier — step 0 already set it to opus.

2. Spawn code-author with that model, isolation: "worktree" and run_in_background: true, then log the spawn row (Workers):

   FIX job. PR #<PR>. Run the fix-pr skill to address all review threads — its step 1 plus the repo's docs/agents/code-host.md give the exact checkout procedure for your worktree; follow them, not memory. Pushing the fixes and replying to the review threads are part of your delegated task. Your entire final message must be the RESULT pr=… url=… line — what you fixed belongs in the thread replies, not in your reply to me.

   When this cycle was triggered by the checks gate (Merge step) rather than by a review verdict, append one line to that prompt naming the failure — The PR's CI is red: <failing job URL>. Fix the failing checks too; there may be no review threads at all. — so the fixer does not go looking for threads that do not exist.

   RESULT blocked … → escalate, next sub-issue (a branch-held-by- worktree blocked gets the same one-shot Cleanup + re-spawn as in step 3).

3. Re-review: spawn diff-reviewer again (same prompt as step 3, mention it is a re-review after a fix pass).

   • CLEAN → Merge.
   • NEEDS_FIXES and c < 3 → next cycle.
   • NEEDS_FIXES and c = 3 → escalate (do NOT merge), next sub-issue.

5. Merge

With merge: manual (the factory default) there is nothing to merge: you already marked the PR ready after the review, so record the sub-issue as ready-to-merge, update its board task (— ready to merge: PR #<PR>), run Cleanup (step 6), record its row (step 7), and move on. The sub-issue stays open until the human merges, so its dependents remain blocked this run.

Say how to merge the moment a sub-issue becomes ready-to-merge — a bare "ready to merge" leaves the user asking what to do, especially off GitHub. State the code-host doc's merge operation concretely.

With merge: auto: this merge is pre-authorized — the user opted into merge: auto in docs/agents/developer-defaults.md (or passed --auto-merge this run), which is standing authorization to merge PRs whose review verdict is CLEAN. If the permission system still asks, say exactly that; if it denies, follow the denial rule under Rules (escalate, never retry).

Checks gate — never merge on red CI. If docs/agents/code-host.md declares a CI system, wait for the PR's checks and read their result before merging, per its "check the change's CI status" operation. GitHub default:

# 1. wait until CI has attached at least one check to the current head sha
for _ in $(seq 20); do
  [ "$(gh pr view <PR> --json statusCheckRollup --jq '.statusCheckRollup | length')" -gt 0 ] && break
  sleep 15
done
# 2. then wait for them to finish — non-zero here means a check actually failed
gh pr checks <PR> --watch --fail-fast

Step 1 is not optional. gh pr checks exits non-zero for two different reasons — a check failed, and no check is registered yet (no checks reported on the '<branch>' branch) — and nothing downstream can tell them apart: the classify step below needs a <run-id> that does not exist yet. The window is real and you will hit it, because gh pr update-branch (the BEHIND path below, and parallel mode's post-merge refresh) moves the head sha and CI takes a few seconds to attach runs to the new one. Waiting for the checks to appear turns that into a wait instead of a red.

Still no check after the loop's ~5 minutes, in a repo whose code-host doc declares CI → that is infra-red: nothing ever picked the change up. Take the infra-red branch below.

Never open a command with a bare sleep — the harness blocks it, here and anywhere else in this skill. Wait inside an until/for loop like the one above, or with the Monitor tool.

• Green (or the code-host doc declares no CI) → merge.

• Red → this is not a conflict. First check whether the branch is merely behind main — a sibling merged after this branch was cut:

  gh pr view <PR> --json mergeStateStatus --jq .mergeStateStatus   # BEHIND?
  

  On BEHIND, run the code host's update-branch operation (gh pr update-branch <PR> on GitHub) and re-run this gate — once per PR; the red was synchronization, not a bug, and no fixer is needed.

  Still red on an up-to-date branch → classify the red before paying for a fixer, per the code-host doc's classify-a-red operation. GitHub default (<run-id> comes from the failing check's link):

  gh run view <run-id> --json conclusion,jobs --jq '{run: .conclusion,
    failed: [.jobs[] | select(.conclusion != "success" and .conclusion != "skipped")
    | {name, steps: (.steps | length)}]}'
  

  • Code-red — a failed job executed steps (steps > 0): the change was exercised and failed. Before paying for a fixer, retry the run once:

    gh run rerun <run-id> --failed
    

    then re-enter this gate. Green → merge, and no fix cycle was spent. Red again → the failure is real: treat it as one more fix cycle (step 4), spawning the fixer with the failing job's URL appended to its prompt. The same three-cycle budget applies; exhausted → escalate.

    Once per PR, and never twice. A suite whose infrastructure wobbles — a service the tests dial refusing connections, an unhandled teardown error, a timeout — reds a change that is fine, and a fix cycle against it buys nothing; one retry is the cheapest way to find out, cheaper than a worker plus a review. But a retry loop merges a genuinely broken intermittent test by persistence, which is worse than spending the cycle. One retry, then believe it.

  • Infra-red — every failed job sits at steps: 0, the run concluded startup_failure, or no runner ever picked the job up: the code was never exercised, so there is nothing a fixer can fix. Spawn none. Escalate the sub-issue naming the cause (runner offline, CI minutes exhausted) and go to wrap-up: a CI that cannot start reds every later PR's gate identically, so continuing burns builds that cannot merge. The wrap-up's unblock question is one line: restore the CI (minutes, runner), then re-run /developer <spec>.

  • The code-host doc defines no classify operation (or the host cannot tell) → every red is code-red, as before.

  Never read CI logs in the main context. The --json query above is the whole diagnosis you are allowed: gh run view --log-failed, --log, and any grep over them dump raw job output straight into the context this design exists to protect — the same rule as "never read issue or PR bodies yourself" (Context economy), and the reason the fixer is handed the failing job's URL instead of your reading of it. If the --json classification is not enough to decide, the answer is the retry above, then the fixer — never a closer look.

  Merging a red PR is the one failure this gate exists to prevent, so never fall through to the merge command on red — not even when the failing check looks unrelated.

Without this gate a repo with no branch protection merges its own red build, and a repo with required checks fails the merge for a reason that is not a conflict — which is exactly what makes the merge-fix job (MERGE-FIX.md, and the failure branch further down this step) the wrong answer to it.

Never touch local git state — your checkout may be in use by the user. Merge remotely, per the code-host doc's merge operation. GitHub default:

gh pr merge <PR> --merge

Do not pass --delete-branch: it also tries to delete the local branch, which is always still checked out in the build worker's worktree, so it fails noisily every time. The remote branch is deleted in Cleanup (step 6), after the worktrees are gone.

Then make sure the sub-issue is closed. If the code host auto-closes linked issues (see docs/agents/code-host.md), just verify — GitHub default:

gh issue view <subissue> --json state --jq '.state'   # expect CLOSED

If there is no auto-close (issues on a different tracker than the code host, or a local tracker), close the sub-issue yourself per the tracker ops, with a comment naming the merged change.

If the merge fails after the checks gate passed, it is a conflict with a previously merged change: read MERGE-FIX.md and dispatch the job it describes, then merge again. That file also covers the conflict a human hits on their own merge under merge: manual — the answer is the same job, never the main context.

6. Cleanup

The harness only auto-removes a worker's worktree when it is unchanged — build and fix workers always leave a branch, commits, and node_modules behind, so without this step every sub-issue leaks worktrees until the disk fills. Run it whenever a sub-issue finishes — merged, ready-to-merge, or escalated — everything is pushed by then, so nothing local is worth keeping.

All removal mechanics live in the bundled script scripts/cleanup-worktrees.sh (next to this SKILL.md — under the plugin root when installed as a plugin). Never improvise git worktree remove, git branch -D, or any other repair yourself — the script is the only sanctioned way to touch local git state here. It removes only the linked worktrees and local branches matching what you pass, refuses by construction to touch the primary checkout, keeps any worktree with uncommitted changes, and deletes a branch only when its commits are on a remote — nothing it deletes is ever the only copy of work. Every refusal is a KEPT line naming its reason: treat those like WARN lines — carry them into the wrap-up summary verbatim, and never re-run with broader flags or improvised git to force what the script declined. If it finds the primary in detached HEAD it prints a WARN line and leaves it alone (that is the fingerprint of a worker having escaped its worktree — carry the WARN into your wrap-up summary, do not "fix" the checkout).

BRANCH=$(gh pr view <PR> --json headRefName --jq .headRefName)   # skip if no PR
HEAD_SHA=$(gh pr view <PR> --json headRefOid --jq .headRefOid)
bash <skill-dir>/scripts/cleanup-worktrees.sh \
  --branch "$BRANCH" --branch "fix/pr-<PR>*" \
  --branch "agent/issue-<subissue>-*" --sha "$HEAD_SHA"

(The two gh pr view lines are the GitHub default for the change-metadata operation — on another host get branch and head sha per docs/agents/code-host.md.)

(A blocked build that never opened a PR has no $BRANCH/$HEAD_SHA — drop those flags; the agent/issue-<subissue>-* pattern still catches its worktree. Its never-pushed branch comes back KEPT (tip not on any remote) — that is the guard working, not a failure: the branch is the only copy of whatever the build did. Leave it and report it.)

If the sub-issue was merged, also delete the remote branch now (the merge deliberately skipped --delete-branch, and cleanup-worktrees.sh only ever deletes local branches):

if git ls-remote --exit-code --heads origin "$BRANCH" >/dev/null 2>&1; then
  git push origin --delete "$BRANCH"
else
  echo "remote branch already gone"
fi

Ask before you push: many hosts delete the head branch themselves on merge, and against one of those a bare git push origin --delete fails with remote ref does not exist on every merge of the run. That noise is indistinguishable from a delete that failed for a reason worth knowing, which is the whole cost of leaving it unguarded.

Matching strictly on this sub-issue's branches/sha is what makes this safe in parallel mode — other wave members' worktrees never match. On an escalated or ready-to-merge sub-issue the remote branch and open PR are untouched; only local state goes.

A review that ran before a fix cycle left its worktree detached at a sha the fixes have since superseded, so it never matches --sha here. That is expected: the wrap-up sweep removes those — do not chase them now, and do not improvise extra flags for them.

7. Record the row

The moment a sub-issue reaches its terminal state — merged, ready-to-merge, or escalated — append its ledger row to this run's log, before touching the next sub-issue:

mkdir -p .scratch
echo "$(date +%F) spec=#<spec> sub=#<N> model=<tier> effort=<effort> pr=#<PR> verdict=<CLEAN|—> cycles=<n> wave=<w|—> outcome=<merged|ready-to-merge|escalated>" \
  >> .scratch/developer-run-<spec>.log

(effort= is the reasoning effort the build ran at — the code-author definition pins it (medium today), so copy that value; it exists so rows stay comparable across runs if the pin ever changes. verdict=— / wave=— where the field does not apply — an escalated sub-issue that never got a CLEAN, sequential mode. pr=none for a build that never opened one, and model=none effort=none pr=none cycles=0 for a sub-issue escalated as oversized, which never reached a builder at all — that row is what lets the harvest notice a spec whose tickets are systematically too big.)

Write it here and the wrap-up reads facts instead of recalling them: a run that survives ten sub-issues, a context compaction, and a resume (step 0) still reports the exact tier, PR and cycle count of the first one. The row is the same one the wrap-up hands the harvest and the same one the chat summary tabulates — write it once, correctly, now.

This row and the event=spawned rows from the Workers section share the file, and the difference is the outcome= field: every row written here carries one, no spawn row does. That is what the wrap-up filters on when it hands the harvest the run's record, and what a resume filters on when it looks for sub-issues still in flight.

The log is a run artifact, not tracked work: never stage it. The context-docs publish (the top-level Step 0) and the local-tracker chore(tracker): commits both name their own paths, so neither picks it up.

Escalation

When a sub-issue is blocked, triaged oversized, non-convergent after 3 fix cycles, or unmergeable, apply the ready-for-human triage label to the sub-issue and comment on both the sub-issue and the spec, per the tracker ops. GitHub default:

gh issue edit <subissue> --add-label "ready-for-human"
gh issue comment <subissue> --body "Escalated by /developer: <reason>. PR: <url or none>."
gh issue comment <spec> --body "Sub-issue #<subissue> escalated: <one-line reason>."

Escalating an oversized sub-issue, the comment must carry the fault lines, not just the verdict: the dispatcher already saw where the work splits and said so in hints=. Pass those hints through verbatim — as the seed of the re-cut, never as the partition itself: splitting a ticket is design work, and the comment routes it to /to-tickets in a fresh session with a high-tier model and high effort, the conditions the original cut was made under. Say PR: none — nothing was built. GitHub default:

gh issue comment <subissue> --body "Escalated by /developer: oversized — does not fit in a single fresh context window. <reason, from the dispatcher>.

Fault lines, from triage (a starting point, not the split): <the dispatcher's hints, verbatim>

To split it: run /to-tickets on this issue in a fresh session with a high-tier model and high effort, starting from the fault lines above. Then remove the \`ready-for-human\` label to put the new sub-issues in play. PR: none."

Leave the PR open (never merge an unclean PR). Run the Cleanup step (step 6) — the local worktrees go, the remote branch and PR stay — record the row (step 7), then continue the loop with the next unblocked sub-issue. An oversized sub-issue has no PR and no worktree: record its row with pr=none and skip Cleanup.

The label is what makes the escalation outlive this session: the pick (spec loop step 1) skips a ready-for-human sub-issue on every future run, until a human removes it.

Wrap-up

When no deliverable sub-issue remains — all closed or ready-to-merge, or the rest blocked by escalated/unmerged ones — read WRAP-UP.md and follow it. It holds the seven closing steps (reconcile the board, harvest + ledger, final sweep, close the spec, push notification, chat summary, execution report). It is read once, here, at the end of the run.

Rules

• Resolve the run configuration (execution + merge) once, before mode detection, and stick to it for the whole run — flags > repo defaults > factory defaults (parallel, manual).
• In sequential mode, one sub-issue is fully delivered (merged, ready-to-merge, or escalated) before the next starts. In parallel mode, builds/reviews/fixes may overlap, but merges are always one at a time.
• Never run the merge operation (gh pr merge, glab mr merge, …) when the resolved config says merge: manual — ready + CLEAN is the terminal state there, even if merging seems convenient.
• Unattended: never stop to ask the user anything mid-loop. Escalate via labels/comments and keep going.
• Every worker spawn is run_in_background: true, in both execution modes. Never hold your turn open waiting for a worker to finish.
• While a run is in flight, no prompt that reaches you is a no-op — a bare "continue" is a resume, not a question. Reconstruct and take the next step per Resuming the orchestrator; never answer that no response is needed.
• Before rebuilding anything, check whether its worker is still alive (ListAgents) and resume it with SendMessage. Re-spawning a live worker's job pays twice for work that was never lost.
• Never act on a RESULT pr=… you have not confirmed exists — one gh pr view after every build, before the reviewer is spawned. A worker's report is a claim until you check it, and this is the only claim you can check for free.
• Each worker starts stateless: pass everything it needs in its prompt; never assume a fresh spawn can see prior steps. A worker resumed with SendMessage is the one exception — it still holds its own context.
• Never run git checkout, git pull, or any state-changing git command in the main context — the only exceptions are Step 0's scoped commit+push of context docs, the cleanup-worktrees.sh script (steps 6 and wrap-up), the merged-branch git push origin --delete, and — local tracker only — the scoped .scratch/ tracker-write commits from LOCAL-HOST.md. If the script warns that the primary checkout is detached, report it — never repair it.
• Only spawn the fix worker when the review said NEEDS_FIXES or the checks gate found the change's CI code-red and the one retry it allows came back red too — an infra-red (the failing job never executed) escalates and ends the run instead.
• Retry a red CI run exactly once per PR, never twice, and never diagnose it by reading its logs in the main context.
• Never spawn a build for a sub-issue triaged oversized — escalate it with the fault lines instead. Buying it a stronger model is the one thing that does not work.
• Never merge a change whose CI checks are red, whatever the merge policy and however unrelated the failing check looks.
• Never resolve merge conflicts in the main context — not even when the user hands you one interactively. That is always the merge-fix job's work, in its own worktree.
• Marking ready and merging are yours, never a worker's; posting the review is the reviewer's, never yours. Never author, edit, or amend review content in the main context.
• If a permission is denied — a worker reports one (posting the review, pushing, commenting), or your own code-host write is denied (marking ready, merging, …) — never re-run the denied action yourself or re-shape it into a different command: that is tunneling around the denial and will also be blocked. Treat the sub-issue as blocked: escalate it and continue the loop.