From work-manager
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.
How this skill is triggered — by the user, by Claude, or both
Slash command
/work-manager:work-helpThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Display the following usage guide to the user:
Display the following usage guide to the user:
| 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 |
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 eachWork 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.
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:
_notes/research-*.md files (findings saved immediately)_notes/_summary.md plan + _notes/plan-*.md files (decisions saved immediately)_notes/impl-*.md files (results documented)Transition via /work update move to plan (or similar phrasing).
/work start — creates _notes/ with _summary.md, phase = research/work update move to plan — build task list, write acceptance criteria, detail approach/work update move to implement — write code, run testsmise run task-append if you need more repos, then /work update to sync/work recall to re-orient, /work recall <topic> for deep dive/work done to verify criteria, then /work prwork-manager): reads phase from _notes/_summary.md, delegates to the right phase agentUserPromptSubmit detects new requirements, Stop logs progress as safety net/work update to manually save findings or transition phases_notes/ files are topic-based, not chronological — same topic accumulates in one file/work recall --deep loads everything for full context (uses more tokens)npx claudepluginhub popoffvg/claude-plugin-work-manager --plugin work-managerManages tasks using native Claude Code subagent tools (TaskCreate, TaskUpdate, TaskList, TaskGet). Tracks TODOs, checkpoints progress, and resumes work across sessions.
Guides task management for complex multi-step projects: create fine-grained work items with dependencies, track progress using claude-reliability CLI commands like 'work next' and 'work create'.
Auto-analyzes project state including tasks, source code, git conflicts, and agents to recommend 1-2 optimal skills. Use on 'what's next?' questions or /workflow trigger.