ce-work

Executes plan documents or work prompts through phased implementation, testing, and shipping workflows.

Updated May 11, 2026
One-click install
npx skills add https://github.com/mmnavarr/harness --skill ce-work-mmnavarr
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ce-work
Source: https://github.com/mmnavarr/harness/tree/main/skills/ce-work
Command: npx skills add https://github.com/mmnavarr/harness --skill ce-work-mmnavarr

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Turning a plan document or a bare work description into a finished, tested, and shipped feature requires disciplined execution: triaging scope, setting up branches, tracking tasks, running tests continuously, and passing code review before creating a PR. This Skill orchestrates that entire execution lifecycle so features ship complete rather than stalling at 80% done. ## Core Features & Use Cases - Phased execution workflow: Triage input (plan file vs. bare prompt), set up branches or worktrees, build task lists, execute with test discovery, and finish with quality checks and PR creation. - Subagent dispatch strategies: Choose inline, serial, or parallel subagent execution with a parallel safety check, worktree isolation, and merge-conflict recovery for large multi-unit plans. - Shipping workflow: Tiered code review (harness-native or escalated deep review), residual work gate, post-deploy monitoring plan, and commit/PR handoff via companion skills. - Use Case: Given a plan document with implementation units, the Skill creates tasks per unit, dispatches independent units to parallel subagents in isolated worktrees, merges results in dependency order, runs tests, and ships a reviewed PR. ## Quick Start Ask the agent to execute the work described in your plan document, for example: run ce-work on docs/plans/user-auth-plan.md and ship the finished feature.

Frequently Asked Questions about ce-work

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I execute a plan document with an AI coding agent?▼

Pass the plan file path as the argument to ce-work. The Skill reads the plan's implementation units, builds a task list, executes each unit with test discovery, and finishes with quality checks and PR creation.

How does ce-work decide between inline and subagent execution?▼

It uses inline execution for 1-2 small tasks or bare prompts, serial subagents for 3+ dependent tasks, and parallel subagents when units pass a file-overlap safety check. Worktree isolation is preferred for parallel dispatch to avoid git index contention.

Can I use ce-work without a plan document?▼

Yes, a bare prompt describing the work is accepted. The Skill scans the affected area, assesses complexity, and routes trivial changes directly to implementation while larger work gets a task list or a suggestion to plan first.

What happens when parallel subagents conflict on the same file?▼

With worktree isolation, the merge conflict surfaces during the post-batch merge; the orchestrator aborts the merge and re-dispatches the conflicting unit serially. In shared-directory fallback mode, colliding units are re-run serially after non-colliding files are committed.

When does ce-work escalate to a deep code review?▼

Tier 2 review triggers when the diff touches sensitive surfaces like auth, payments, or migrations, exceeds 400 changed lines across multiple directories, exceeds 1000 lines total, or the plan explicitly requests a thorough review. Otherwise the harness-native review is used.