| name | work-help |
| description | This skill should be used when the user says "work help", "how does work manager work", "work usage", "work commands", "what can work do". Shows available commands, workflow, and tips.
|
work-help
Display the following usage guide to the user:
Work Manager — Usage Guide
Commands
| Command | What it does |
|---|
/work start | Begin new work — creates _notes/ with _summary.md and work notes |
/work status | Show current work status and progress |
/work recall | Re-orient: what was I doing, what's next? |
/work recall --deep | Full synthesis including all work notes |
/work recall <topic> | Show specific work note (e.g. work recall auth-flow) |
/work update | Log progress, capture work notes, review structure |
/work done | Mark work complete, check acceptance criteria |
/work pr | Create a PR from current work context |
/work install | Guided setup — plugin, QMD, mise, task scripts, worktrunk |
/work help | This guide |
Work Notes Hierarchy
repo-root/
_notes/
_summary.md # Index — compact overview, links to other _notes/ files
README.md # Work notes index and structure rules
auth-flow.md # Topic: how auth works
db-schema.md # Topic: database design decisions
perf-findings.md # Topic: performance research
...
Rules:
_notes/_summary.md = index only (plan, criteria, work notes links)
_notes/worklog.md = append-only progress log
_notes/*.md = one file per topic, under 100 lines each
- Topics split automatically when they grow too large
- Structure is reviewed on every update and session end
Multi-Repo Workspace
Work always spans multiple repos with different languages (Go, TypeScript, Rust, etc.).
The _notes/_summary.md tracks repos and their languages in a Repos table.
Repo list is mutable — use mise run task-append to add repos mid-work.
When repos change, update the table via /work update.
Phases & Agents
Each phase has a dedicated agent with restricted tools — mode enforcement is structural, not prompt-based.
research → plan → implement
↑ ↓
←────────←
| Phase | Agent | Tools | Cannot do |
|---|
| research | work-researcher | Read, Glob, Grep, Bash (r/o), Explore agents, Write (_notes/ only) | Edit source code |
| plan | work-planner | Read, Glob, Grep, Bash (r/o), Write (_notes/ only) | Edit source, spawn agents |
| implement | work-implementer | All tools | — |
Primary deliverables by phase:
- research:
_notes/research-*.md files (findings saved immediately)
- plan: updated
_notes/_summary.md plan + _notes/plan-*.md files (decisions saved immediately)
- implement: working code +
_notes/impl-*.md files (results documented)
Transition via /work update move to plan (or similar phrasing).
Workflow
- Start: checkout a branch, then
/work start — creates _notes/ with _summary.md, phase = research
- Research: explore, gather context — hooks auto-capture knowledge
- Plan:
/work update move to plan — build task list, write acceptance criteria, detail approach
- Implement:
/work update move to implement — write code, run tests
- Iterate: go back to research or plan when needed
- Add repos:
mise run task-append if you need more repos, then /work update to sync
- Check in:
/work recall to re-orient, /work recall <topic> for deep dive
- Finish:
/work done to verify criteria, then /work pr
How it works
- Router (
work-manager): reads phase from _notes/_summary.md, delegates to the right phase agent
- Phase agents: each has restricted tools baked into their definition — can't be overridden
- Hooks:
UserPromptSubmit detects new requirements, Stop logs progress as safety net
Tips
- Phase agents save findings as their PRIMARY output — not as a side effect
- Use
/work update to manually save findings or transition phases
_notes/ files are topic-based, not chronological — same topic accumulates in one file
- Edit any file directly if you need to fix something fast
/work recall --deep loads everything for full context (uses more tokens)