do-task
10
总安装量
9
周安装量
#30554
全站排名
安装命令
npx skills add https://github.com/driangle/taskmd --skill do-task
Agent 安装分布
opencode
9
gemini-cli
9
github-copilot
9
codex
9
kimi-cli
9
cursor
9
Skill 文档
Do Task
Look up a task and start working on it.
Instructions
The user’s query is in $ARGUMENTS (a task ID like 077 or a task name/keyword).
- Look up the task: Run
taskmd get $ARGUMENTSto find the task- If not found, run
taskmd listto show available tasks and ask the user which one they meant
- If not found, run
- Read the task file with the
Readtool to get the full description, subtasks, and acceptance criteria - Mark the task as in-progress: Run
taskmd set <ID> --status in-progress - Start a worklog entry (if worklogs are enabled):
- Check
.taskmd.yamlforworklogs: false— if set, skip worklog steps - Otherwise, find or create the worklog file at
tasks/<group>/.worklogs/<ID>.md(ortasks/.worklogs/<ID>.mdfor root tasks) - Append a timestamped entry noting your approach and initial findings
- Check
- Do the task: Follow the task description and complete the work described
- Use
EnterPlanModefor non-trivial implementation tasks - Check off subtasks (
- [x]) in the task file as you complete them - Append worklog entries when you make key decisions, hit blockers, or complete significant subtasks
- Use
- Write a final worklog entry summarizing what was done, decisions made, and any open items
- Mark the task as done:
- Check
.taskmd.yamlforworkflow: pr-review— if set, use the PR-review workflow below - Solo workflow (default): Run
taskmd set <ID> --status completed --verify- The
--verifyflag will run any verification checks defined in the task before applying the status change - If verification fails, fix the issues and try again
- The
- PR-review workflow: Open a PR, then run
taskmd set <ID> --status in-review --add-pr <PR-URL>and stop
- Check
Worklog Format
Each worklog entry uses a timestamp heading followed by free-form notes:
## 2026-02-15T10:30:00Z
Started implementation of the search feature.
**Approach:** Using full-text search with the existing SQLite database
rather than adding Elasticsearch -- simpler and sufficient for our scale.
**Completed:**
- [x] Added search query parser
- [x] Created search index
**Next:** Add result ranking and write tests.