dot-tasks（点任务）

dot-tasks Skill

dot-tasks is a Python CLI (assumed installed) that tracks human/agent work in a repo-local .tasks/ directory.

Use this skill whenever a repository uses dot-tasks for task lifecycle tracking.

When To Use

• Use this skill when the user asks for task suggestions based on existing dot-tasks state.
• Use this skill when the user asks to begin or resume work on an existing task.
• Use this skill when the user asks for significant new work that should likely be tracked.

Plan Mode Deliverables

By default, when asked to work on a task, produce both artifacts in order:

1. Develop/refine the task specification and write it to spec.md.
2. Develop the implementation plan and write it to plan.md.

• plan.md must be based on a current, user-confirmed spec.md.
• Exceptions:

- If explicitly asked for spec-only work, produce/update only spec.md. - If resuming from a completed spec.md, validate it is still current, then produce/update plan.md.

• Artifact scope:

- spec.md: A precise, testable statement of what must be built, including scope, requirements, interfaces, and acceptance criteria. - plan.md: An ordered execution strategy for how to build it, including steps, dependencies, checkpoints, and risk mitigations.

Workflow 1: Suggest What To Work On Next

1. Discover candidate work:

- Run dot-tasks list todo --json (or dot-tasks list --json if status is not specified). - Inspect likely top candidates with dot-tasks view <task_name_or_id> --json.

1. Rank candidates using this rubric:

- Higher priority first (p1 before p2, etc). - Prefer unblocked tasks (dependency_health: ready) over blocked work. - Note spec readiness and effort when suggesting tasks to user.

1. Return the top few options with one-line rationale each.
2. If high-priority work is blocked, call that out explicitly and include:

- A short unblock path for the blocked item. - A suggested unblocked fallback task.

Workflow 2: Begin Or Resume Existing Task

1. Resolve the target task:

- If the user provides task_name/task_id, run dot-tasks view <task_name_or_id> --json. - If ambiguous, list likely matches and confirm the target with the user before binding.

1. Branch by task status:

- doing: resume by reading spec.md, plan.md, and recent activity.md, then continue from the latest checkpoint. - todo: run readiness checks before starting. - done: do not silently restart; ask whether to create a follow-up task or reopen scope explicitly.

1. Ensure spec is ready before planning implementation
2. If intent is unclear, ask directed open-ended questions in two stages to clarify intent and develop clear spec. Do not proceed on unstated assumptions when high-level intent is unclear.

Workflow 3: Significant New Work

1. Detect whether the request is substantial (multi-file, plan-heavy, or likely >=30 minutes).
2. If substantial, ask whether to create and bind a new dot-tasks task.
3. If the user agrees:

- Create task with dot-tasks create... (include summary and basic metadata). - Confirm the tracking target and bind work to that task.

1. If work is quick/simple, do not force task creation unless the user asks.

Shared Task Lifecycle Loop

For tracked task execution (regardless of how it was triggered), follow: create -> spec -> confirm -> start -> plan -> log-activity -> complete

• Follow Plan Mode Deliverables.
• Start active execution with dot-tasks start: sets status to doing and creates empty plan.md.
• If Plan Mode/intent work produced a finalized spec artifact, sync that Markdown to spec.md on the first execution turn after Plan Mode.
• If Plan Mode produced a finalized <proposed_plan>, sync that Markdown to plan.md on the first execution turn after Plan Mode, after spec confirmation.
• Keep plan.md current as implementation decisions become concrete.
• Log meaningful progress with dot-tasks log-activity --note.
• Log meaningful progress with dot-tasks log-activity --note.
• For major tasks, document decisions made during implementation in a decisions.md artifact; include key decision rationale and alternative choices considered. Add "# NOTE: " comments in code for significant decisions. For major decisions, pause and ask user for direction before proceeding. Report back to user at end of execution with a summary of decisions made.
• Use dot-tasks update for mid-flight metadata/scope/priority changes.
• Before dot-tasks complete, confirm acceptance criteria are satisfied.

Commands

# setup
dot-tasks init                                              # initialize .tasks/

# discover
dot-tasks list --json                                       # list tasks for matching
dot-tasks list [todo|doing|done] --json                    # narrow by status
dot-tasks view <task_name_or_id> --json                    # inspect one task
dot-tasks tags [todo|doing|done] --json                    # tag counts/triage

# lifecycle
dot-tasks create <task_name> --summary "..." --priority [p1|p2|p3|p4] --effort [s|m|l|xl] --tag <tag>
dot-tasks start <task_name_or_id>                          # move to doing + create plan.md
dot-tasks update <task_name_or_id> --priority p1 --effort m --tag backend
dot-tasks log-activity <task_name_or_id> --note "Progress note" [--actor agent]
dot-tasks complete <task_name_or_id>                       # move to done

# maintenance
dot-tasks rename <task_name_or_id> <new_task_name>         # rename task
dot-tasks delete <task_name_or_id>                         # soft-delete to trash

Guardrails

• Prefer dot-tasks commands over direct edits to task state files.
• Avoid silent auto-binding on fuzzy matches.
• Confirm task binding with the user before tracked execution.
• Apply ordering/default/exception behavior from Plan Mode Deliverables.
• Direct file edits are allowed for:

1. task.md for writing task summary/specs after dot-tasks create. 2. spec.md: A precise, testable statement of what must be built, including scope, requirements, interfaces, and acceptance criteria. 3. plan.md: An ordered execution strategy for how to build it, including steps, dependencies, checkpoints, and risk mitigations. 4. Other task-local artifacts (for example walkthrough.md, decisions.md, handoff.md) only when useful for scope and naturally produced during the session; confirm with the user before non-trivial additions, and do not manufacture extra artifacts for small/self-contained tasks.

• Do not rewrite activity.md history; append only.
• Respect dependency checks.

Data Contract

• Canonical metadata is in task.md frontmatter.
• Dependency references use task_id in metadata.
• Dependencies are displayed to humans as task_name (task_id).
• activity.md line format is YYYY-MM-DD HH:MM | actor | type | note.