Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Work

skills.sh GitHub License: MIT

Quality-first orchestration for coding agents.

Work turns a request into one exact goal, gives specialists complete bounded briefs, and keeps the parent responsible for integration and proof. It is a portable Agent Skill for Codex and other compatible coding agents.

Install

npx skills add xcaeser/work-skill --skill work --agent codex --global --yes

Install one focused mode by pointing the CLI at its folder:

npx skills add https://github.com/xcaeser/work-skill/tree/main/skills/work-plan \
  --agent codex --global --yes

Use

$work Fix the intermittent login failure and ship the smallest verified repair.
$work 2 sol xhigh Investigate the data-loss report and implement the durable fix.

If no team size is given, Work chooses the smallest useful team. Explicit model, effort, and count requests remain under the user's control.

Skills

Skill Purpose
$work Implement a task with an accountable parent lead
$work-init Start a project with a minimal verified setup
$work-plan Produce a detailed read-only implementation plan
$work-audit Audit code, diffs, or runtime flows
$work-clean Find smells, overengineering, and package opportunities
$work-quality Pass the Work golden rules into the conversation
$work-status Show the current goal, agents, evidence, and blockers
$work-help Show the available Work skills

Workflow

  1. Define one goal, ownership boundary, done criteria, and validation path.
  2. Launch the smallest useful team with complete executor briefs.
  3. Keep materially different approaches independent until their real strengths and gaps are visible; track the approach families explicitly.
  4. Require concrete changes, tests, lemmas, or other evidence—not status or optimism. Agents work until they finish or report a real blocker.
  5. Use adversarial checks, then have the parent integrate, verify, and report remaining gaps. Reopen a blocked route only with a genuinely new mechanism.

These principles are informed by OpenAI's proof prompt: diverse exploration, explicit route tracking, adversarial review, concrete evidence, and no premature return. Work applies that discipline to coding tasks; the PDF is not a Work API or model-pricing specification.

Quality bar

  • Software should not break, demand attention, or exceed its purpose.
  • Keep decisions human-owned and user-facing writing human-readable.
  • “Your work is done when it looks so simple that the consumer thinks they could have done it, which means they won’t appreciate how hard you worked.” — Visualize Value.
  • Prefer the smallest reliable change and remove obsolete compatibility paths.
  • Treat “done” as a claim that current evidence proves.

Engineering discipline

  • Choose the simplest implementation that fully meets current requirements.
  • Grow in working layers; never trade a working product for unfinished complexity.
  • Keep components modular and concerns clearly separated.
  • Check existing dependencies, documentation, and types before reimplementing functionality or adding a package.
  • Make architectural decisions for the long term, not as temporary stopgaps.

Validate

npx skills add . --list --full-depth
git diff --check

Each skill is a portable directory with SKILL.md frontmatter. The package has no runtime dependencies or bundled services.

License

MIT. See LICENSE.