rlaope/oh-my-hermes / .claude/agent-skills/ulw-work
ulw-work
Verified[omh] Accepted plan awaiting implementation. Aliases: ulw.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 14.6 KB |
| references/dependency-topology.md | markdown | 4.4 KB |
| references/details.md | markdown | 6.7 KB |
SKILL.md
View on GitHub---
name: "ulw-work"
description: "[omh] Accepted plan awaiting implementation. Aliases: ulw."
license: MIT
compatibility: "Requires the omh CLI on PATH (pip install oh-my-hermes)."
metadata:
hermes:
tags: [workflow, oh-my-hermes, execution]
category: execution
phase: parallel-delivery
role: handoff-guide
quality_tier: handoff-gated
---
# Ultrawork
This is an OMH `ultrawork` workflow skill, projected for Agent Skills hosts (Claude Code, Codex, Cursor, opencode, OpenClaw, pi).
## Why This Exists
Read `references/details.md` first.
## First Steps
Read `references/details.md` first.
## Do Not Use When
Read `references/details.md` first.
## Examples
Read `references/details.md` first.
## Completion Checklist
Read `references/details.md` first.
## Recovery Notes
- If lanes are non-disjoint, collapse to one owner or route back to the durable-checkpoint goal ledger before coding starts.
- If a worker does not ACK or return a result, keep that lane blocked/not_observed and expose the retry or reassignment action.
- If a worktree or shared-file conflict appears, pause parallel delivery and re-plan ownership before more edits.
- If a node fails, recover node-locally: the failure blocks only its dependents; read its error and retry first, amend the node definition when its prompt or contract is wrong, and steer a live lane instead of duplicating its owner - never rebuild the graph.
- Do not read a quiet or scheduled node as stalled; inspect returned output because a returned blocked response still completes the node and carries the blocker to report.
- [capability:coordinated_scope] If a coordinated worker has no ACK or result, mark that lane not_observed or blocked rather than infer progress.
- [capability:durable_checkpoint] If the goal ledger is stale or missing, inspect .omh/goals and ask which checkpoint to resume before continuing.
- [capability:durable_checkpoint] If a blocker checkpoint exists, keep the goal open and record the blocker plus the smallest unblock action.
## When to Use
Use when an accepted implementation plan can be split into independent, reviewable work lanes.
Strong routing signals: `ultrawork`, `$ultrawork`, `ulw`, `$ulw`; all of them in `references/details.md`.
## Catalog Metadata
Category: `execution`
Phase: `parallel-delivery`
Quality tier: `handoff-gated`
Reasoning demand: `heavy`
Quality bar:
- Do not start this engine as an automatic continuation of another skill's output: an accepted plan, a clarified brief, or a routing recommendation is planning evidence, not permission. Unless the user explicitly invoked this engine themselves, restate in one line what will start (engine, scope, selected executor) and wait for the user's explicit go-ahead first.
- Resolve the dependency_topology decision before any dispatch: work coupled by a shared invariant or inseparable edit boundary collapses to one owner; separable but ordered units get explicit acyclic dependency edges; independent units form the dependency-ready parallel frontier; no unit dispatches without scope, acceptance criteria, a verification command, and an owner route - load `references/dependency-topology.md` for the full discipline.
- When adapting an existing capability, read `../omh-verification-gate/references/adaptation-quality.md` before accepting a plan, dispatching work, or claiming completion; carry its criteria and comparisons into tasks and evidence.
- Attach acceptance criteria, verification commands, and review expectations to each lane.
- Keep dispatch, execution, review, CI, and merge status evidence separate.
- After final brief composition and before unattended coding handoff, explicitly run `omh handoff-risk-scan --brief-file <final-brief> --repo <workspace> --strict --json`. Route high_risk (exit 1) to existing confirmation or `omh-security-safety-review`; scan_error (exit 2) requires repaired input and a new scan. Clear never grants permission or bypasses metadata preflight, approval, or host policy.
- Write every lane or node prompt standalone with TASK, DELIVERABLE, SCOPE, VERIFY, and STOP WHEN in that order, exact paths and binary pass/fail observables, and one role per node; a dependency edge orders execution only and never substitutes upstream output.
- End every code-changing run with a verification fan-in that depends on all producer lanes, runs the repository's real test/build command, and reports captured binary pass/fail output; downstream consumers re-check upstream claims before trusting them.
- For each behavioral increment follow PIN -> RED -> GREEN -> SURFACE -> CLEAN: pin behavior a refactor could hide, capture the intended failing proof before implementation, make the smallest change, exercise the real user surface, and tear down every QA resource with a cleanup receipt; tests alone never prove completion.
- Keep one inspectable, append-only evidence ledger for the run using the available goal/runtime records: record the tier decision, dependency topology, todo transitions, command outputs, real-surface artifacts, and cleanup receipts when each occurs.
- For tests-first work, capture the new test's failing nonzero output before implementation and its passing zero output plus the full suite before done; never edit, delete, skip, or weaken a test to make it pass.
- [capability:coordinated_scope] Keep the current host as coordinator and narrator; delegate through the host's own subagent/task mechanism with explicit ownership, or run lanes sequentially when unavailable.
- [capability:delivery_boundary] Complete exactly one plan-to-PR delivery cycle, then stop with status, evidence gaps, or a next recommended workflow.
- [capability:delivery_boundary] Start a delivery cycle with codebase/source research and a ralplan-style decision record before implementation handoff.
- [capability:delivery_boundary] Run `omh-code-review` as a gate after implementation evidence exists; review preparation alone is not review evidence.
- [capability:delivery_boundary] End a delivery cycle with a PR-ready or PR-observed report that separates prepared, executed, reviewed, verified, CI, and PR evidence.
- [capability:delivery_boundary] Use the host's own subagent/task mechanism with explicit owner, acceptance criteria, verification commands, and a resumable ledger for durable checkpoints. Never imply hidden execution.
- When the user explicitly selects an external coding owner, use the host's authorized task mechanism, verify its actual capabilities, and capture its session/task handle. Missing delegation means sequential work or a named blocker, never fabricated dispatch.
- Name each lane owner and its actual host capabilities before dispatch; do not infer model or tool routing from inheritance.
- Choose the wait strategy before starting long-running work and bind it to a completion signal the host exposes, never to a status loop: a command that fits one tool call runs once in the foreground with a duration-sized timeout; a longer terminal command runs in the background with completion notification armed and no process-status polling; a delegated lane relies on its delivered result while the parent continues independent work or ends the turn; a CI, PR, deploy, file, port, log-line, or external-session condition uses the host's monitor when observed, else exactly ONE bounded watcher or adaptive backoff outside model turns. Record the handle and observation mode at dispatch; every armed wait needs a hard deadline, a cancellation path, and a fallback naming the missing capability. Each wait closes in one terminal state with bounded evidence; an unbounded idle or busy-wait is a defect and a lost notification times out. One decision-changing midpoint peek and any user-requested status check stay allowed; neither is the wait mechanism. Ladder and terminal states: shared rail.
- A mid-run user message is an interjection, not a stop: answer it briefly and, in the same reply, continue the run — re-read the phase todo when one is active and dispatch or advance the next pending step, or name the armed wait it is waiting on -- handle, bound completion signal, deadline -- instead of re-reading status. Only the user's explicit stop or cancel, or the engine's own completion gate, ends the run; when the interjection changes scope, say so and update the declared plan or todo instead of silently abandoning it. A mid-run message is the latest steering for the active task, not automatically a replacement objective: it replaces the objective when the user says so and steers the current one otherwise.
- A follow-up that needs new authority, materially expands the scope, or changes external state not already authorized is described first and started only on the user's approval; persistence never broadens the authorized scope. A refused escalation gets a safer alternative inside the boundary, or the authorization the boundary asks for — never a workaround or an indirect execution.
- The closing brief scales to the change: one or two sentences for a simple change. Lead with the result and validation; omit abandoned approaches unless they explain a tradeoff, and internal bookkeeping. Disclose requested-but-not-done work and material prepared/unmerged work; omit unrelated non-action footers. Required accounting stays outside this scaling. Order: result -> relevant evidence/limits -> observed accounting (when required) -> specific question (if useful). For a real unresolved user decision or a useful briefing discussion, offer one specific, content-linked question. Completed work needs no question; discussion grants no authority. Never estimate usage.
- Close with observed host accounting when available, or explicitly report run-summary not_available, before any specific next-action question. Never estimate token counts, elapsed time, or models used.
- [capability:single_owner_persistence] Do not enter a finish-until-done loop until scope, acceptance criteria, and verification commands are concrete.
- [capability:single_owner_persistence] For single-owner coding edits, prepare and track the selected runtime path instead of implying unobserved work happened or hiding execution inside chat narration.
- [capability:single_owner_persistence] Report single-owner completion only from observed execution and verification evidence, with remaining risks named.
- [capability:durable_checkpoint] Keep goal state durable, inspectable, and separate from chat narration in the metadata-only .omh/goals goal_ledger/v1.
- [capability:durable_checkpoint] Create a code-changing goal with `omh goal create --obligation slop-cleanup`; it completes only after a checkpoint cites one completed `omh-ai-slop-cleaner` run for it, recorded and cited with the commands that skill closes on (`--cleanup-outcome passed|skipped|reverted` and an outputs or verification summary the integrity check accepts; placeholders are refused). Without the flag the goal is not gated.
- [capability:durable_checkpoint] Checkpoint every success, blocker, and final quality gate with fresh evidence.
- [capability:durable_checkpoint] Reject completion with a summary-only goal_completion_gate/v1 result until required criteria, blockers, and explicitly linked runtime runs are satisfied.
- [capability:durable_checkpoint] Name the one element gating goal progress from the linked loop's loop_constraint_assessment/v1 before checkpointing the next step; load `ulw-loop/references/goal-constraint-discipline.md` for the method.
Required inputs:
Read `references/details.md` first.
Expected outputs:
- runtime handoff prompts or lane instructions
- status summary
- review/CI evidence requirements
- [capability:delivery_boundary] `durable_checkpoint` or selected executor/runtime handoff
Artifact expectations:
Read `references/details.md` first.
Safety rules:
- Do not run two concurrently runnable lanes with overlapping write scopes; a shared file requires an ordering edge or one owner.
- Keep orchestration/status and implementation ownership explicit in the current host; preserve prepared versus observed evidence boundaries.
- Record unobserved executor work as prepared_not_observed or not_observed.
- [capability:coordinated_scope] Use coordination lanes only when work is independent; if two lanes are not independent, collapse them under one owner or re-plan before dispatch.
- [capability:coordinated_scope] Keep shared-file edits under one owner; if integration reveals a shared-file conflict, stop lane fan-out and reassign ownership before continuing.
- [capability:coordinated_scope] Record unobserved delegation as not_observed; a delegation record exists only when separate participants are observed.
- [capability:delivery_boundary] Do not continue into a repeated feedback loop; recommend `ulw-loop` when the user wants ongoing cycles.
- [capability:delivery_boundary] Do not skip planning when the delivery request is broad, risky, or user-visible; a ralplan-style or reviewed plan names acceptance criteria, risks, and verification commands.
- [capability:delivery_boundary] Run docs sync only when behavior, setup, commands, examples, or public claims changed.
- [capability:delivery_boundary] Keep web research source-backed and permission-aware; do not run hidden network or LLM calls from OMH core.
## Runtime Evidence
Use the current host's own tools and subagent/task mechanism when available;
otherwise run the same lanes sequentially or name the unavailable capability.
A prepared plan, handoff, checklist, or skill installation is not execution,
review, CI, merge-readiness, or merge evidence. Record actual tool results, or
`not_observed` / `not_available`, in the record; never invent dispatch or host
accounting.
Treat supplied context as advisory, not proof of hidden memory reads or writes.
State scope, constraints, verification, and the stop condition before work.
Reply in the user's own words and the host's own voice: the host's persona owns
reply language, tone, speech level, and sentence endings, progress updates
included (where it sets no language, use the one the user wrote in), and OMH
shapes structure and content only; OMH's record terms
(surface, lane, wrapper, handoff, evidence boundary, not_observed) stay in
records and tool calls, never in the sentence the user reads unless they ask
about one; and when a real unresolved user decision or a useful briefing discussion
remains, offer one specific, content-linked question. Completed work needs no question;
omit unrelated non-action footers and generic offers to continue.
Supporting paths are relative to this skill directory; sibling skill paths are
relative to its parent. Resolve them from the host-provided skill base directory
(`{baseDir}` on hosts that provide it), never a hardcoded install location.
A named workflow not installed here is unavailable, not permission to emulate
its host-specific capabilities. Verify through the real surface before done.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |