@code-yeongyu/lazycodex 0.1.1

package/packages/omo-codex/plugin/skills/start-work/SKILL.md

@@ -0,0 +1,129 @@
+---
+name: start-work
+description: "Execute a Prometheus work plan in Codex with Boulder state, evidence ledger updates, worktree discipline, parallel subagents, and Stop-hook continuation. Use after planning when the user says start work, execute plan, continue plan, resume plan, or asks to run a .omo/plans plan."
+---
+
+## Codex Harness Tool Compatibility
+
+This skill ports the OpenCode `/start-work` flow onto Codex. Any OpenCode-only tool name in an inherited example must be translated to its Codex equivalent:
+
+| OpenCode example | Codex tool to use |
+| --- | --- |
+| `task(subagent_type="explore", ...)` | `spawn_agent(agent_type="explorer", task_name="...", message="...")` |
+| `task(subagent_type="librarian", ...)` | `spawn_agent(agent_type="librarian", task_name="...", message="...")` |
+| `task(subagent_type="plan", ...)` | `spawn_agent(agent_type="plan", task_name="...", message="...")` |
+| `task(subagent_type="oracle", ...)` for final verification | `spawn_agent(agent_type="codex-ultrawork-reviewer", task_name="...", message="...")` |
+| `task(category="...", ...)` for implementation or QA | `spawn_agent(agent_type="worker", task_name="...", message="...")` |
+| `background_output(task_id="...")` | `wait_agent(...)` |
+| `dispatchInternalPrompt(...)` | the `Stop` hook emits `{"decision":"block","reason":"<prompt>"}` automatically; see Continuation |
+| `team_*(...)` | `spawn_agent` + `send_message` + `followup_task` + `wait_agent` + `close_agent` |
+
+When translating `load_skills=[...]`, name the skills inside the spawned agent's `message`. If a code block below conflicts with this section, this section wins.
+
+# start-work
+
+Execute a Prometheus work plan until every top-level checkbox is complete. This skill pairs with the Codex `Stop` / `SubagentStop` continuation hook in `components/start-work-continuation`, which re-injects the next turn while `.omo/boulder.json` says the current `codex:<session_id>` still has unchecked plan work.
+
+## Usage
+
+```text
+$start-work [plan-name] [--worktree <absolute-path>]
+```
+
+- `plan-name` is optional. It may be a full or partial file stem under `.omo/plans/`.
+- `--worktree` is optional. Use it only when the user explicitly asks to work in a separate git worktree.
+
+## Phase 1: Select the plan
+
+1. Read `.omo/boulder.json` if it exists.
+2. List Prometheus plan files under `.omo/plans/`.
+3. If `plan-name` was provided, select the matching plan.
+4. If exactly one active or paused Boulder work exists for this session, resume it.
+5. If no active work exists and exactly one plan exists, select it.
+6. If multiple plans remain possible, ask one focused selection question.
+
+## Phase 2: Create or update Boulder state
+
+Write `.omo/boulder.json` before implementation starts. Session ids must be prefixed with `codex:` so the continuation hook can identify its own session.
+
+```json
+{
+  "schema_version": 2,
+  "active_work_id": "<work-id>",
+  "works": {
+    "<work-id>": {
+      "work_id": "<work-id>",
+      "active_plan": ".omo/plans/<plan-name>.md",
+      "plan_name": "<plan-name>",
+      "session_ids": ["codex:<session_id>"],
+      "status": "active",
+      "worktree_path": null
+    }
+  }
+}
+```
+
+If `--worktree` is set, verify the path with `git worktree list --porcelain` or create it with `git worktree add <path> <branch-or-HEAD>`, then store the absolute path as `worktree_path`. All edits, commands, tests, and evidence capture must run inside that worktree.
+
+## Phase 3: Execute the next checkbox
+
+1. Read the full selected plan.
+2. Find the first unchecked column-0 checkbox in `## TODOs` or `## Final Verification Wave`.
+3. Ignore nested checkboxes under acceptance criteria, evidence, and definition-of-done sections.
+4. Decompose that checkbox into atomic sub-tasks.
+5. Dispatch independent sub-tasks in parallel with `spawn_agent`; serialize only when one sub-task has a named dependency on another.
+
+Each sub-task message must include:
+
+1. Goal and exact files or directories in scope.
+2. When the task touches existing behavior: a baseline characterization test, written first, that asserts current observable behavior and passes on the unchanged code. Then the red test or failing reproduction for the new behavior before production changes. Pin the baseline as rigorously as the new test: exact inputs, exact observable, exact assertion.
+3. Implementation constraints from the plan and project rules.
+4. Automated verification commands to run.
+5. One Manual-QA channel, named with the exact tool and exact invocation (the literal `curl`, `send-keys`, `page.click`, payload, selectors, and the binary observable that decides PASS/FAIL), not "verify it works":
+   - HTTP call: `curl -i` against the live endpoint.
+   - tmux: a `tmux` session driven with `send-keys`, dumped via `capture-pane`.
+   - Browser use: use Chrome to drive the real page; if Chrome is not available, download and use agent-browser (https://github.com/vercel-labs/agent-browser).
+   - Computer use: OS-level GUI automation against the running desktop app when the surface is not a page.
+6. The adversarial classes that apply to this sub-task (from the 9 ultraqa classes) and how each is probed.
+7. Required artifact path and cleanup receipt.
+
+Apply ultraqa's 9 adversarial classes where relevant to each checkbox: malformed input, prompt injection, cancel/resume, stale state, dirty worktree, hung or long commands, flaky tests, misleading success output, repeated interruptions. A checkbox whose behavior is user-visible MUST probe every class that plausibly applies; record which classes were exercised and which were ruled not-applicable with a one-line reason.
+
+## Phase 4: Verify and record evidence
+
+For each checkbox, complete all five gates before marking it done:
+
+1. Plan reread: confirm the checkbox and acceptance criteria.
+2. Automated verification: run tests, typecheck, lint, build, or the plan-specific equivalent.
+3. Manual-QA channel: capture a real artifact, not a dry-run claim.
+4. Adversarial QA: exercise every applicable ultraqa class (malformed input, prompt injection, cancel/resume, stale state, dirty worktree, hung or long commands, flaky tests, misleading success output, repeated interruptions) and capture the observable result for each. "Tests pass" and a clean happy-path artifact are NOT sufficient when an adversarial class applies and was not probed.
+5. Cleanup: register every QA resource teardown as its own todo the moment it is spawned (QA scripts, tmux assets, browser / agent-browser sessions, PIDs, ports, containers, temp dirs), then execute each and capture the receipt. No QA asset is left running.
+
+Append evidence to `.omo/start-work/ledger.jsonl` using one JSON object per line. Include at least `event`, `plan`, `task`, `session_id`, `commands`, `artifact`, `adversarial_classes`, and `cleanup` fields. `adversarial_classes` lists each probed class with its observable result and each ruled-out class with a one-line reason.
+
+## Phase 5: Mark progress
+
+Only after verification passes:
+
+1. Edit the plan checkbox from `- [ ]` to `- [x]`.
+2. Re-read the plan and confirm the remaining count decreased.
+3. Append a `task-completed` ledger entry.
+4. Continue with the next checkbox. Do not ask whether to continue.
+
+## Completion
+
+When all top-level checkboxes in `## TODOs` and `## Final Verification Wave` are complete:
+
+1. Run the plan's final verification commands.
+2. If worktree mode was used, sync `.omo/` state back to the main repo, merge or hand off exactly as requested, and remove the worktree only after successful merge or explicit handoff.
+3. Remove or mark the Boulder work as completed.
+4. Print an `ORCHESTRATION COMPLETE` block with the plan path, verification commands, artifacts, and cleanup receipts.
+
+## Hard rules
+
+- No production change before a failing test or reproduction exists, and no change to existing behavior before a baseline characterization test pins the current behavior and passes on the unchanged code.
+- No `--dry-run` as completion evidence.
+- No tests-only completion claim. A Manual-QA artifact is required.
+- No completion claim while an applicable ultraqa adversarial class was never probed. Each applicable class needs a captured observable result; each skipped class needs a one-line not-applicable reason in the ledger.
+- No unprefixed session ids in Boulder state. Codex sessions are always `codex:<session_id>`.
+- No stale-memory execution. The plan and ledger are the durable source of truth.

package/packages/omo-codex/plugin/skills/ulw-loop/SKILL.md

@@ -0,0 +1,222 @@
+---
+name: ulw-loop
+description: Goal-like loop that uses ultrawork mode to decompose work into systematic, evidence-bound steps.
+metadata:
+  short-description: Goal-like ultrawork loop for systematic decomposition
+---
+
+## Role
+Expert goal orchestration agent. You conduct; right-sized parallel subagents play. Plan multi-goal work that survives across turns and sessions, fan independent work out to workers, QA every result yourself, record only proven evidence.
+Use GPT-5.x style: outcome-first, evidence-bound, atomic decisions, no nested branching prose.
+
+## Goal
+Deliver every goal in `.omo/ulw-loop/goals.json` end-to-end.
+Prove EVERY success criterion with captured observable evidence from a real-usage scenario you actually ran (HTTP call / tmux / browser use / computer use — see the Manual-QA channels below).
+TESTS ALONE NEVER PROVE DONE. A green test suite is supporting evidence, not completion proof.
+Audit each pass, fail, block, steering change, and checkpoint in `.omo/ulw-loop/ledger.jsonl`.
+
+## Manual-QA channels (PICK ONE PER CRITERION — ACTUALLY RUN IT)
+For every criterion, build a real-usage scenario through ONE of these four channels and run it yourself before recording PASS. The full test suite being green is NEVER verification on its own.
+
+1. **HTTP call** — hit the live endpoint with `curl -i` (or a Playwright APIRequestContext); capture status line + headers + body.
+2. **tmux** — `tmux new-session -d -s ulw-qa-<criterion>`, drive with `send-keys`, dump via `tmux capture-pane -pS -E -`; transcript is the artifact.
+3. **Browser use** — use Chrome to drive the REAL page; if Chrome is not available, download and use agent-browser (https://github.com/vercel-labs/agent-browser). Capture action log + screenshot path. Never downgrade to a non-browser surface for a browser-facing criterion.
+4. **Computer use** — when the surface is a desktop/GUI app rather than a page, drive it via OS-level automation (a computer-use agent, AppleScript, xdotool, etc.) against the running app; capture action log + screenshot. Use this for any non-browser GUI criterion.
+
+Auxiliary surfaces (pure CLI stdout / DB state diff / parsed config dump) satisfy CLI- or data-shaped criteria but NEVER replace a channel scenario for user-facing behavior. `--dry-run`, printing the command, "should respond", and "looks correct" never count.
+
+## Delegation model (ATLAS-STYLE — YOU CONDUCT, WORKERS PLAY)
+You read, search, plan, integrate, and QA. You DELEGATE every code edit, test write, bug fix, and QA execution to a right-sized `spawn_agent` worker, then verify what comes back. Fan out independent tasks in PARALLEL in a single response; serialize only on a NAMED dependency (one task consumes another's output or edits the same file).
+
+Size each worker to the task — never spend `xhigh` on a one-liner, never send a race condition to a mini. Pass `model` + `reasoning_effort` per call (an override needs a non-full-history fork mode):
+
+| Task shape | agent_type | model | reasoning_effort |
+|---|---|---|---|
+| Trivial / mechanical (rename, move, obvious one-liner, config edit) | `worker` | `gpt-5.4-mini` | `low` |
+| Pure implementation against a clear spec (new function, endpoint, test from a named pattern) | `worker` | `gpt-5.3-codex` | `high` |
+| Deep debugging / race / perf / subtle cross-module reasoning | `worker` | `gpt-5.5` | `xhigh` |
+| QA execution (drive a channel, capture evidence) | `worker` | `gpt-5.3-codex` | `high` |
+| Read-only codebase search | `explorer` | role default | role default |
+| External library / docs research | `librarian` | role default | role default |
+| Final verification audit | `codex-ultrawork-reviewer` | role default | role default |
+
+Every worker message MUST carry: goal + exact files in scope; the baseline characterization test pinning current behavior when the task touches existing code, then the failing test / reproduction required before production code; constraints + project rules; the verification commands to run; the ONE Manual-QA channel and the exact evidence artifact to capture. Workers have NO interview context — be exhaustive, and forward accumulated learnings to every next worker. Do not use `list_agents` as a polling or status tool in long or high-context runs; it can replay large agent status and latest-message payloads. Track spawned agent names locally, use `wait_agent` for completion, send targeted followups only when needed, and `close_agent` after integrating each result.
+
+## Artifacts
+- `.omo/ulw-loop/brief.md`: original brief and durable constraints.
+- `.omo/ulw-loop/goals.json`: goals with embedded `successCriteria` per goal.
+- `.omo/ulw-loop/ledger.jsonl`: append-only audit trail.
+- Read artifacts before resuming, steering, or checkpointing.
+- Never invent state outside `.omo/ulw-loop` artifacts or `omo ulw-loop status --json`.
+
+## Bootstrap
+Do all three steps before execution. No edits, goal tools, or checkpointing before bootstrap completes.
+
+### 1. Create goals from the brief
+Resolve the CLI before the first command. If `omo` is absent from PATH, use the stable local installer bin or cached Codex component CLI. This is the same ulw-loop CLI, so PATH absence is not a blocker. If PATH is empty, the fallback uses shell builtins and absolute Node locations before reporting guidance, and records the failure in `.omo/ulw-loop/bootstrap-notepad.md`.
+```sh
+if command -v omo >/dev/null 2>&1; then
+  ULW_LOOP_CLI=omo
+else
+  CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
+  ULW_LOOP_CLI=
+  if [ -f "$CODEX_HOME/bin/omo" ] || [ -x "$CODEX_HOME/bin/omo" ]; then
+    ULW_LOOP_CLI="$CODEX_HOME/bin/omo"
+  else
+    for candidate in "$CODEX_HOME"/plugins/cache/sisyphuslabs/omo/*/components/ulw-loop/dist/cli.js; do
+      [ -f "$candidate" ] || continue
+      ULW_LOOP_CLI="$candidate"
+    done
+  fi
+
+  ULW_LOOP_NODE="$(command -v node 2>/dev/null || true)"
+  if [ -z "$ULW_LOOP_NODE" ]; then
+    for candidate in /opt/homebrew/bin/node /usr/local/bin/node /usr/bin/node; do
+      [ -x "$candidate" ] || continue
+      ULW_LOOP_NODE="$candidate"
+      break
+    done
+  fi
+
+  if [ -n "$ULW_LOOP_CLI" ] && [ -n "$ULW_LOOP_NODE" ]; then
+    omo() { "$ULW_LOOP_NODE" "$ULW_LOOP_CLI" "$@"; }
+  fi
+fi
+
+if [ -z "${ULW_LOOP_CLI:-}" ]; then
+  /bin/mkdir -p .omo/ulw-loop 2>/dev/null || mkdir -p .omo/ulw-loop 2>/dev/null || true
+  NOTE="${NOTE:-.omo/ulw-loop/bootstrap-notepad.md}"
+  printf '%s\n' "omo executable missing from PATH; cached ulw-loop CLI not found under ${CODEX_HOME:-$HOME/.codex}." >> "$NOTE" 2>/dev/null || true
+  printf '%s\n' "Install with bunx omo install --platform=codex or set CODEX_LOCAL_BIN_DIR to a PATH directory." >&2
+fi
+```
+If `ULW_LOOP_CLI` is empty, open the durable notepad first, record the missing CLI evidence, then surface the installer issue.
+
+Run one form:
+```sh
+omo ulw-loop create-goals --brief "<brief>" --json
+omo ulw-loop create-goals --brief-file <path> --json
+cat <brief> | omo ulw-loop create-goals --from-stdin --json
+```
+Write state through the CLI path. Do not hand-edit state files.
+
+### 2. Refine success criteria + a Prometheus-grade QA and parallelism plan per goal
+Gather context BEFORE planning — fire parallel `explorer` / `librarian` workers plus your own read-only tools; never plan blind.
+First survey the skills available in this system: read the description of every loosely-relevant skill, decide deliberately which ones this work will use, and prefer using as many genuinely-applicable skills as apply rather than working raw. Then size the scope: count distinct surfaces, files, and steps. For any non-trivial goal (2+ steps, multi-file, unclear scope, or an architecture decision) spawn the `plan` agent with the gathered context and let IT decide the wave ordering and parallel grouping; follow that order and grouping exactly and run the verification it specifies. Only a genuinely trivial single-step goal may skip the plan agent.
+Define pass/fail acceptance criteria before launching execution lanes. Include the command, artifact, or manual check that will prove success.
+Each goal MUST carry 3+ `successCriteria` covering happy path, edge, regression, and adversarial risk.
+For each criterion set, concretely and upfront: `id`, `scenario` (the exact tool — curl / tmux / playwright / computer-use — plus exact steps with specific inputs and a binary pass/fail), `expectedEvidence` (the exact artifact path, e.g. `.omo/ulw-loop/evidence/<goal>-<criterion>.<ext>`), adversarial classes, stop condition, and the Manual-QA channel (HTTP call / tmux / browser use / computer use) that will exercise it. Vague QA ("verify it works") is a rejected criterion — revise it before execution.
+Apply ultraqa classes where relevant: malformed input, repeated interruptions, prompt injection, cancel/resume, stale state, dirty worktree, hung or long commands, flaky tests, misleading success output.
+Use evidence verbs from the channel table (tmux transcript, curl status+body, browser screenshot, computer-use action log, CLI stdout, DB diff, parsed config dump) — not vibes.
+"Tests pass" is supporting signal, NEVER completion proof. Every criterion needs its own channel scenario, built fresh and exercised every time.
+
+**Plan for maximum parallelism.** Decompose each goal's criteria into atomic tasks (Implementation + its Test = ONE task, never split) and group them into dependency waves. Target 5–8 tasks per wave; <3 per wave (except the final wave) means under-splitting — extract shared prerequisites into Wave 1. For each task record its wave, what it blocks, what blocks it, the worker tier from the Delegation table, and its QA scenario + evidence path. Build a dependency matrix (Task | Depends on | Blocks | Can parallelize with) and name the critical path. Anything not on a real dependency edge MUST share a wave and dispatch together.
+Record manual QA notes when behavior is user-visible.
+Revise any criterion that lacks observable `expectedEvidence` or a named channel before execution.
+
+### 3. Inspect state
+Run `omo ulw-loop status --json`.
+Read pending goals, criteria IDs, current ledger head, blockers, and aggregate Codex objective.
+
+## Execution Loop
+Loop per goal. Cap at 5 cycles per goal. Cap identical same-criterion failures at 3.
+
+### Acquire Next Goal
+1. Run `omo ulw-loop complete-goals --json` and read the handoff, including criteria.
+2. Call `get_goal` and inspect active Codex state.
+3. Apply this table exactly:
+
+| get_goal result | action |
+|-----------------|--------|
+| no active goal | Call `create_goal` with the handoff payload. |
+| same aggregate objective active | Continue the current ulw-loop story. |
+| different goal active | STOP. Checkpoint blocked and surface the conflict. |
+4. If retrying failed work, run `omo ulw-loop complete-goals --retry-failed --json`.
+5. Never create a second Codex goal for the same aggregate objective.
+
+### Per-Criterion Cycle
+1. PLAN: read `criterion.scenario`, `criterion.expectedEvidence`, prior ledger entries, and safety bounds. Identify which tasks in the current wave are independent.
+2. Register atomic todos: `path: <action> for <criterion> - verify by <check>`.
+3. DELEGATE-IN-PARALLEL: dispatch every independent task in the wave at once via right-sized `spawn_agent` workers (Delegation table). Each worker does strict TDD on its task: when the task touches EXISTING behavior, PIN it FIRST — write a characterization test that asserts the current observable behavior and PASSES on the unchanged code, so any later regression fails loudly. Then RED (the new failing assertion must fail for the RIGHT reason — no syntax/import error), then the SMALLEST GREEN change; a GREEN needing >~20 lines means the test was too coarse — instruct a split. The baseline-pin scenario must be as rigorous and specific as the new-behavior scenario: exact inputs, exact observable, exact assertion. Serialize only on a NAMED dependency.
+4. INTEGRATE + CRITICAL SELF-QA (EVERY WORKER RETURN): do NOT trust the worker's report. Read the diff yourself, re-run its tests, and run LSP diagnostics on the changed files. Treat "done" as a claim to disprove. If the diff drifts, the test is hollow, or evidence is missing, RESPAWN the worker with the specific failure context. Forward every finding/learning to subsequent workers.
+5. EXECUTE-AS-SCENARIO: ACTUALLY run the Manual-QA channel scenario the criterion named (HTTP call / tmux / browser use / computer use — see the channel table above). Run it yourself for the orchestrator check; for heavier flows dispatch a dedicated QA worker (`worker`, `gpt-5.3-codex`, `high`) whose ONLY job is to drive the channel and write the artifact to the named evidence path. The unit suite being green is NEVER substitute. If the scenario FAILS, respawn the implementing worker with the captured failure — do not hand-patch around it.
+6. CAPTURE: collect the observable artifact path: transcript, stdout, screenshot, assertion, status+body, diff, or parsed dump. No artifact written at the evidence path — not done; record BLOCKED and respawn QA.
+7. CLEAN (PAIRED, NEVER SKIP): tear down every runtime artifact step 5 spawned BEFORE recording — server PIDs (`kill`, verify `kill -0` fails), `tmux` sessions (`tmux kill-session -t ulw-qa-<criterion>`; confirm `tmux ls`), browser / Playwright contexts (`.close()`), containers (`docker rm -f`), bound ports (`lsof -i :<port>` empty), temp sockets / files / dirs (`rm -rf` the `mktemp` paths), QA-only env vars, AND `close_agent` on every finished worker. Register each teardown as its own todo the moment the QA spawns the resource (scripts, tmux assets, browsers / agent-browser sessions, PIDs, ports) so none is forgotten. Embed a one-line cleanup receipt in the evidence string, e.g. `cleanup: killed 12345; tmux kill-session ulw-qa-foo; rm -rf /tmp/ulw.aB12cD; close_agent w-3`. Missing receipt → record BLOCKED, not PASS.
+8. RECORD exactly one result:
+   - PASS: `omo ulw-loop record-evidence --goal-id <id> --criterion-id <id> --status pass --evidence "<observable> | <cleanup receipt>" --json`
+   - FAIL: `omo ulw-loop record-evidence --goal-id <id> --criterion-id <id> --status fail --evidence "<observable> | <cleanup receipt>" --notes "<diagnosis>" --json`
+   - BLOCKED: `omo ulw-loop record-evidence --goal-id <id> --criterion-id <id> --status blocked --evidence "<observable>" --notes "<safety/blocker/leftover-state>" --json`
+9. If actual does not match expected, diagnose, respawn the right-sized worker with the failure context to fix minimally, and rerun the SAME criterion (including a fresh cleanup).
+10. After 3 same-criterion failures, exit the goal with diagnosis.
+11. After 5 cycles on one goal without all criteria passing, checkpoint failed.
+12. Continue only when the next pending criterion has a concrete `expectedEvidence` target.
+
+### Goal Completion
+1. Confirm every criterion is `pass` with `omo ulw-loop criteria --goal-id <id> --json`.
+2. Call `get_goal` for a fresh snapshot.
+3. Run `omo ulw-loop checkpoint --goal-id <id> --status complete --evidence "<criteria evidence summary>" --codex-goal-json <snapshot> --json`.
+4. If blocked or failed, checkpoint with `--status blocked` or `--status failed` and include diagnosis evidence.
+5. If this is the final goal, run the final quality gate first and pass `--quality-gate-json`.
+
+## Final Quality Gate
+Trigger only when one goal remains and all its criteria are passing.
+1. Run targeted verification for changed behavior.
+2. Run `ai-slop-cleaner` on changed files. If no relevant edits exist, record a passed no-op cleaner report.
+3. Rerun verification after cleanup.
+4. Run `$code-review`.
+5. Clean review means `codeReview.recommendation == "APPROVE"` and `codeReview.architectStatus == "CLEAR"`.
+6. If review is non-clean, run `omo ulw-loop record-review-blockers --goal-id <id> --title "<...>" --objective "<...>" --evidence "<review findings>" --codex-goal-json <snapshot> --json`.
+7. If clean, checkpoint final completion:
+```sh
+omo ulw-loop checkpoint --goal-id <id> --status complete --evidence "<e2e evidence + manual QA notes>" --codex-goal-json <snapshot> --quality-gate-json <json-or-path> --json
+```
+`--quality-gate-json` shape:
+```json
+{
+  "aiSlopCleaner": { "status": "passed", "evidence": "cleaner report" },
+  "verification": { "status": "passed", "commands": ["npm test"], "evidence": "post-cleaner verification" },
+  "codeReview": { "recommendation": "APPROVE", "architectStatus": "CLEAR", "evidence": "review synthesis" },
+  "criteriaCoverage": { "totalCriteria": N, "passCount": N, "adversarialClassesCovered": ["malformed_input", "..."] }
+}
+```
+
+## Dynamic Steering
+Use steering only for structured evidence-backed mutation. Reject natural-language steering requests.
+
+| Kind | When to use | Required fields |
+|------|-------------|-----------------|
+| add_subgoal | Real blocker found; new story required | `--title`, `--objective`, `--evidence`, `--rationale` |
+| split_subgoal | Story too large; needs decomposition | `--goal-id`, `--children` JSON, `--evidence`, `--rationale` |
+| reorder_pending | Discovered dependency order | `--order` JSON array of ids, `--evidence`, `--rationale` |
+| revise_pending_wording | Title/objective ambiguous | `--goal-id`, `--title?`, `--objective?`, `--evidence`, `--rationale` |
+| revise_criterion | Criterion lacks observable PASS evidence | `--goal-id`, `--criterion-id`, `--scenario?`, `--expected-evidence?`, `--evidence`, `--rationale` |
+| annotate_ledger | Audit-only note | `--evidence`, `--rationale` |
+| mark_blocked_superseded | Old story replaced by new evidence | `--goal-id`, `--replacements?`, `--evidence`, `--rationale` |
+
+Command form: `omo ulw-loop steer --kind <kind> [<kind-specific-fields>] --evidence "<...>" --rationale "<...>" --json`.
+Structured prompt directives accepted: `OMO_ULW_LOOP_STEER: { ... }`, `omo.ulw-loop.steer: {...}`, `omo ulw-loop steer: {...}`.
+
+## Constraints
+1. NEVER call `update_goal` mid-aggregate; only on final story after the quality gate passes.
+2. NEVER call `create_goal` when `get_goal` shows a different active goal.
+3. NEVER mark `criterion.status == "pass"` without captured observable evidence in `record-evidence`.
+4. NEVER bypass the criteria gate at checkpoint; all criteria must be `pass` before `--status complete`.
+5. Baseline build/lint/typecheck/test commands are necessary evidence, NOT SUFFICIENT completion proof. Criteria coverage with observable evidence is the gate.
+6. Treat `.omo/ulw-loop/ledger.jsonl` as the durable audit trail; checkpoint after every success or failure.
+7. Per-story Codex goal mode is opt-in only with `--codex-goal-mode per-story`; default is aggregate.
+8. Structured steering directives mutate state through validation; normal prose does not.
+9. Evidence MUST be observable from the real surface: tmux transcript, curl status+body, browser/Playwright assertion, CLI stdout, DB state diff, parsed config dump.
+10. Apply ultraqa's 9 adversarial classes where relevant per goal: malformed input, prompt injection, cancel/resume, stale state, dirty worktree, hung commands, flaky tests, misleading success output, repeated interruptions.
+11. After completing an aggregate ulw-loop run, clear the Codex goal manually with `/goal clear` before starting another in the same session.
+12. The shell command emits a model-facing handoff; only the Codex agent calls `get_goal`, `create_goal`, or `update_goal` tools.
+13. NEVER record `--status pass` while a QA-spawned process, `tmux` session, browser context, bound port, container, or temp file / dir is still alive, or while any worker is still open. The evidence string MUST include the cleanup receipt. Leftover runtime state = BLOCKED, not PASS.
+14. DELEGATE all code edits, test writes, fixes, and QA execution to right-sized `spawn_agent` workers (Delegation table); you read, search, plan, integrate, and QA. NEVER record `--status pass` from a worker's self-report — only from evidence you re-verified yourself. Dispatch independent tasks in parallel; serialize only on a NAMED dependency.
+
+## Stop Rules
+- All goals complete plus all criteria `pass` plus final quality gate clean: DONE.
+- 3x same criterion failure: checkpoint failed, surface diagnosis.
+- 5 cycles on one goal without all-pass: checkpoint failed, surface.
+- Safety boundary such as destructive command, secret exfiltration, or production write: block and surface a safe substitute.
+- Codex `get_goal` reports a different active goal: checkpoint blocker, stop, surface.
+- Leftover state from QA (live process, `tmux` session, browser context, bound port, temp dir): NOT pass. Clean up, append the receipt, then continue.
+- User issues `/cancel`: release in-progress state cleanly and do not auto-resume.

package/packages/omo-codex/plugin/skills/ulw-loop/agents/openai.yaml

@@ -0,0 +1,6 @@
+interface:
+  display_name: "ulw loop"
+  short_description: "Goal-like ultrawork loop for systematic decomposition"
+  search_terms:
+    - "ulw-loop"
+  default_prompt: "Use $ulw-loop to break this work into a systematic ultrawork loop with evidence-backed checkpoints."