run-work
Execute a work unit end-to-end: sequence tasks by dependency, implement, test between tasks, commit, and track progress. Use to deliver a complete f…
它会碰到什么
这一栏是扫描器报的事实,不是结论。命中多不等于有毒(安全工具、规则库、示例脚本本来就会包含危险写法),命中少也不等于干净。它和你手上的凭据、文件、网络有什么关系,需要你自己看。
技能内容
> Invoked as /agiflow:run-work. In hosts without slash-prompts, this skill is triggered by matching intent and drives AgiFlow via its MCP tools.
Usage:
/agiflow:run-work <work-unit-slug-or-id>- Execute specific work unit/agiflow:run-work- List and select from available work units
Examples:
/agiflow:run-work DXX-WU-1(using slug)/agiflow:run-work 01K8FABMNEJG1XTA9JGHSNFV40(using ID)/agiflow:run-work(interactive selection)
Guardrails
- Favor straightforward, minimal implementations first and add complexity only when it is requested or clearly required.
- Keep changes tightly scoped to the requested outcome within the work unit scope.
- A work unit represents a cohesive feature/epic that can be completed in one Claude Code session.
If a work unit slug/id is provided, load it with get_work_unit; otherwise list available work units with list_work_units for selection.
AgiFlow Project Management Guidelines
Follow the shared AgiFlow project-management guidelines in [references/agiflow-agents.md](../../references/agiflow-agents.md) — agent assignment, the task status workflow and transitions, work-unit best practices, and the tags strategy apply to this workflow.
Task Status Workflow (per task)
Each task moves individually through: Todo → In Progress → Testing → Review
The work unit stays in_progress until all tasks reach Review or Done.
If any task hits Blocked, consider setting the work unit to blocked too.
IMPORTANT: Planning Status Guard
This skill ONLY executes tasks in "Todo" or later status. Tasks in "Planning" have NOT been groomed and are NOT ready for execution. Use backlog-grooming to promote Planning tasks to Todo first.
Steps
Track these steps as TODOs and complete them one by one.
1. Work Unit Selection & Loading
If work unit slug/id NOT provided:
- Use
list_work_unitsMCP tool to show available work units:
- Filter by
status: "in_progress"for active work, or work units with tasks in "Todo" - Do NOT pick up work units where all tasks are still in "Planning" status
- Display: slug, title, type, priority, task count, status
- Ask user to select which work unit to work on.
- Once user selects, proceed with the selected work unit slug/id.
If work unit slug/id IS provided: 4. Use get_work_unit MCP tool with the provided slug/id to retrieve:
- Work unit: title, description, goals, type, priority, status, dates, devInfo
- Tasks: All tasks associated with this work unit (already included in response)
- Note:
get_work_unitreturns tasks automatically - no separatelist_taskscall needed
- Review the work unit and all its tasks to understand:
- Complete feature scope and deliverables
- Task dependencies and execution order
- Acceptance criteria across all tasks
5b. PLANNING STATUS GUARD (MANDATORY):
- Check all tasks in the work unit
- If ANY task has status "Planning": REFUSE to execute
- Report: "Cannot execute: N task(s) are still in Planning status. Use backlog-grooming to promote them to Todo first."
- List the Planning tasks by title
- STOP execution — do not proceed
2. Start Work Unit Execution
- If work unit status is "planning":
- First verify ALL tasks are in "Todo" or later status (not "Planning")
- If verified, use
update_work_unitMCP tool to set status to "in_progress" - If any tasks are in "Planning", REFUSE and suggest
backlog-grooming
- Create a TODO list of all tasks in execution order (use TodoWrite tool).
- Document execution plan in work unit via
update_work_unitdevInfo:
devInfo: {
executionPlan: "Backend API → Frontend UI → Tests → Documentation",
sessionId: "<current-session-id>",
startedAt: "<timestamp>"
}
3. Execute Tasks Sequentially
- For each task in the work unit (in dependency order from the tasks array):
- Review task details: title, description, acceptance criteria, assignee
- Use
update_taskMCP tool to set status to "In Progress" - BEFORE editing any code: Use
architectMCPget_file_design_pattern(MANDATORY) - Implement the task following acceptance criteria
- AFTER editing code: Use
architectMCPreview_code_change(MANDATORY) - Update task
devInfowith implementation notes:
devInfo: {
filesChanged: ["path/to/file.ts:42"],
testResults: { passed: true, coverage: "85%" },
notes: "Implementation notes here"
}
- Mark acceptance criteria as checked via
update_task
Testing phase for each task:
- Use
update_taskto move status to "Testing" - Run unit tests, integration tests, type check and lint for affected code
- If tests PASS: proceed to move task to "Review"
- If tests FAIL:
- Increment
retryCountin devInfo - If
retryCount >= maxRetries(default: 2): move task to "Blocked", set work unit to "blocked", stop - If
retryCount < maxRetries: move task back to "Todo", document failure, continue to next task or stop session - Use
update_taskto set status to "Review" when tests pass
- Update your TODO list (mark task as completed)
- Between tasks:
- Commit changes with meaningful commit messages
- Run tests to ensure no regressions
- Update work unit progress in devInfo
4. Work Unit Progress Tracking
- The work unit tasks array automatically updates as tasks are completed.
- Use
get_work_unitto check current state and task statuses - Track: How many tasks in Review or Done vs. total?
- Monitor: Are we on track for target date?
- Work unit stays
in_progressuntil all tasks reachRevieworDone
- Update work unit
devInfoas you progress viaupdate_work_unit:
devInfo: {
executionPlan: "...",
sessionId: "<session-id>",
startedAt: "<timestamp>",
progress: {
completedTasks: 3,
totalTasks: 8,
lastTaskCompleted: "Implement cart API",
currentTask: "Add cart UI component"
},
testResults: {
unitTests: "passing",
integrationTests: "passing",
coverage: "85%"
},
blockers: [] // or list any blockers encountered
}
5. Work Unit Completion
- When ALL tasks in work unit are in "Review" or "Done":
- Verify all tasks have status "Review" or "Done"
- Verify all acceptance criteria across all tasks are met
- Run full test suite for the feature
- Review all files changed (use git diff)
- Draft a PR description for the work unit (DO NOT create the PR yet - just draft the text):
- Title format:
[WORK-UNIT-SLUG] Work unit title(e.g.,[DXX-WU-1] Shopping cart feature) - Body should include:
- Work unit reference and summary
- List of tasks completed
- Summary of all changes across tasks
- Files modified (use
file.ts:42format) - Test results
- Use
update_work_unitto set status to "completed" and save draft PR text and commit message:
{
status: "completed",
completedAt: new Date(),
devInfo: {
...existing,
draftCommitMessage: "feat(cart): implement shopping cart feature\n\n- Add cart API endpoints\n- Add cart UI components\n- Add integration tests\n\nCloses: DXX-WU-1",
draftPr: {
title: "[DXX-WU-1] Shopping cart feature",
body: "## Summary\n\nImplemented shopping cart feature.\n\n## Tasks Completed\n\n- [DXX-1] Add cart API\n- [DXX-2] Add cart UI\n- [DXX-3] Add cart tests\n\n## Changes\n\n- Added cart endpoints\n- Added cart components\n- Added integration tests\n\n## Files Modified\n\n- src/api/cart.ts:42\n- src/components/Cart.tsx:15\n- tests/cart.test.ts:1\n\n## Test Results\n\nAll tests passing. Coverage: 85%"
},
finalNotes: "All tasks completed. Files changed: [...]. Tests passing.",
completedBy: "<member-id>",
totalDuration: "3.5 hours"
}
}
Note on draftCommitMessage: Write a conventional commit message that will be used for the final git commit. Format: type(scope): description with optional body listing changes.
- Create final summary comment documenting:
- Work unit scope and goals achieved
- All files created/modified (use
file.ts:42format) - Test coverage and results
- Any follow-up work needed
- Performance or architectural notes
6. Handle Blockers or Scope Changes
- If blocked on a task:
- Use
update_taskto set task status to "Blocked" - Use
update_work_unitto set status to "blocked" - Document blocker in work unit devInfo
- Use
create_task_commenton blocked task with details explaining what human intervention is needed - Consider creating follow-up tasks for blockers
- If scope changes during execution:
- Use
create_taskto add new tasks to work unit - Update work unit description/goals if needed
- Communicate scope change in work unit comments
Common Mistakes to Avoid
- ❌ Starting without reviewing all tasks in work unit (lack of context)
- ❌ Working on tasks in wrong order (missing dependencies)
- ❌ Skipping vibe-lint MCP validation (MANDATORY for every file)
- ❌ Not tracking work unit-level progress in devInfo
- ❌ Completing tasks in isolation without considering work unit goals
- ❌ Not running integration tests between tasks
- ❌ Moving task to "Review" without first going through "Testing"
- ❌ Marking work unit complete when tasks are still in "In Progress" or "Testing"
- ❌ Not documenting scope changes or blockers
- ❌ Forgetting to commit between tasks (losing incremental progress)
- ❌ Not updating work unit status when blocked
- ❌ Setting work unit to "completed" before all tasks reach "Review" or "Done"
- ❌ Executing tasks that are still in "Planning" status (must be promoted to "Todo" via
backlog-groomingfirst)
想直接用这个技能?
本站把开放许可(MIT / Apache 等)的技能按仓库打包整理到网盘,点一下转存到你自己的网盘,不用一个个从 GitHub 拉。许可未声明的技能只给原始仓库链接,不打包。
它属于哪个仓库
plugins/AgiFlow/ai-plugin/skills/run-work/SKILL.md