Stats
Actions
Tags
From claude-tasks
Use whenever a task should be added to the claude-tasks queue — either because the user said "add a task / track this / remember to do X / put this on the list", OR because you (Claude), in the middle of other work, discovered something that should become a tracked task: a follow-up, a deferred fix, a TODO you left, a security issue, an improvement, a "we should also…". This is how Claude self-captures work into the queue from any session. Fill the brief out as meaningfully as you can from what you already know.
How this skill is triggered — by the user, by Claude, or both
Slash command
/claude-tasks:queue-taskThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Add a well-formed **brief** to the task queue. A brief is one markdown file rich enough
Add a well-formed brief to the task queue. A brief is one markdown file rich enough
that a future agent can execute it with zero prior context. Add one from anywhere with
the bundled add_task.py.
Add a brief whenever you would otherwise leave a loose end:
Do NOT capture: things you're doing right now, pure conversation, or duplicates of briefs that already exist (check first — see Dedup).
Run the script. Fill out as much as you meaningfully can from context — don't leave fields blank that you can infer:
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/add_task.py" \
--title "<scannable one-liner, lead with the noun>" \
--goal "<the end state, not 'do X'>" \
--context "<everything a zero-context agent needs: repos, paths, links, why>" \
--criteria "<checkable condition>" --criteria "<another>" \
--type todo|project|research|decision \
--importance 1|2|3|4 \
--autonomy full|needs-input|blocked \
--effort xs|s|m|l|xl --due YYYY-MM-DD \
--domain "<one of the domains in tasks.toml>" \
--source claude --commit
--ready: add it if you can ALSO fill out goal + criteria + context well enough to
execute. Then it lands in ready/. Otherwise (default) it lands in inbox/ for later
enrichment (the enrich skill turns inbox items into ready briefs).--autonomy full ONLY if a future agent could complete it end-to-end with zero
user input. Be honest — over-tagging full makes the queue lie about what's delegable.--commit writes, commits, and pushes. Use it unless you have a reason to batch.The script finds the queue via $CLAUDE_TASKS_DIR → nearest .tasks/ → ~/tasks, or
pass --root <path> explicitly. If it reports "No task queue", run /tasks-init first.
Before adding, check for an existing brief on the same thing:
grep -rl "<keyword>" "$CLAUDE_TASKS_DIR"/{inbox,ready,in-progress,done} 2>/dev/null
If one exists, sharpen it (add a note to its body) instead of duplicating. Fewer, richer briefs beat many thin ones.
--source claude so a review can show what Claude self-generated.importance. Default 3 unless it's genuinely urgent/strategic.inbox/ and let enrichment
handle it. Capturing badly beats losing the idea.Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub kmmwren/claude-tasks --plugin claude-tasks