TaskDock is an agent skill that plans and maintains a portable Desktop workspace for ongoing work. Your goal, decisions, current state, evidence, and results stay together so a fresh session can continue without reconstructing the entire chat.
任务坞：让每个复杂任务都有自己的桌面工作位置。
- Creates a task folder with a stable UUID, a human-readable entry point, current state, and a working plan.
- Plans categories from the actual work and adds folders when they have a use.
- Updates decisions, verification, and next actions as the task changes.
- Finds renamed or relocated task folders within specified search roots and repairs stale index entries.
- Detects conflicting copies and broken common Markdown links instead of silently guessing.
Use it for multi-step research, a launch, a design project, or continuing work that needs a durable record. Say:
用 TaskDock 给这个任务建桌面工作文件夹，规划怎么做，并在过程中维护进度。
找回昨天的任务文件夹，按它的状态继续做。
这个任务目录乱了，按工作逻辑重新整理并修好引用。
Simple answers and tiny edits do not need a workspace unless you ask for one.
npx skills add m1nga/taskdockThen ask your agent: Use $taskdock to organize this task and keep it ready to resume.
Requires Python 3.8 or later; no paid API is used by the filesystem helper. A skill
installation may need a fresh session for automatic discovery.
Commands below run from the installed skill directory. From another directory, use
the absolute path to its scripts/taskdock.py. Add --language en at initialization
for English task notes; Chinese is the CLI default.
TASK.json identifies the task. README.md explains what the folder handles. STATE.md tracks current facts, decisions, verified progress, and the next action. PLAN.md explains dependencies and why files are grouped that way. AGENTS.md lets an agent entering the folder find these records.
The task ID survives a move. Internal relative links survive a whole-folder move. The Desktop task index is a replaceable location hint, not the only copy of your work. A missing index can be rebuilt from task folders.
python3 scripts/taskdock.py init --title "Launch review" --goal "Ship a reviewed launch plan"
python3 scripts/taskdock.py locate --title "Launch review"
python3 scripts/taskdock.py check --path "/path/to/task-folder"
python3 scripts/taskdock.py resume --path "/path/to/task-folder"Use locate --id <task-id> --root <search-directory> for a task moved outside the default Desktop/Documents search. A folder copied twice keeps the same ID; the tool reports ambiguity so the intended working copy can be selected explicitly.
A folder becomes useful when it carries the reason for the work and the next decision, not merely a transcript. TaskDock starts small and expands its structure as deliverables and dependencies become real. It keeps existing repositories authoritative and task evidence separate from public packages.
It does not run while the agent is idle, guarantee discovery anywhere on a disk, move unrelated files, or replace backups. Scheduled work still requires an actual automation. The link check covers common inline Markdown links, not arbitrary HTML or Office document references.
Planning with Files informed the explicit recovery check and single-owner rule for shared state. TaskDock adapts those ideas to portable Desktop tasks and keeps updates tied to meaningful changes instead of a fixed tool-call counter. It does not copy another engine's hook support or claim that a plain skill installation enables background execution.
Thirteen executable tests cover moved folders, stale or missing indexes, duplicate identities, existing-file preservation, broken-link repair, bounded search, control-file symlinks, corrupt metadata, and same-name task isolation.
python3 scripts/test_taskdock.py -vThe tests establish filesystem behavior, not the quality of every future task plan.
Built by Ming. MIT licensed.