メニュー
Claim a taskwarrior task as in-flight (+ACTIVE) and write a coworker marker. Use when picking up a task from task-coordinate output before starting implementation.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/laurigates/claude-plugins --skill task-claimコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Install, configure and troubleshoot MCP servers. Use when adding/enabling servers, editing .mcp.json, fixing OAuth, or when a server runs stale code after an upstream fix.
FinOps snapshot — org billing, workflow stats, cache usage. Use when you want a high-level view of CI spending or workflow health before diving deeper.
GitHub Actions billing, workflow efficiency, and waste analysis at org or repo level. Use when investigating CI/CD costs, wasted runs, or optimizing triggers.
SKILL.md を表示中
SOC 職業分類に基づく
| name | task-claim |
| description | Claim a taskwarrior task as in-flight (+ACTIVE) and write a coworker marker. Use when picking up a task from task-coordinate output before starting implementation. |
| args | <task-id> [--no-coworker-marker] [--force] |
| allowed-tools | Bash(task *), Bash(git rev-parse *), Bash(git branch *), Bash(git config *), Bash(hostname *), Bash(jq *), Bash(bash *), Read, TodoWrite |
| argument-hint | task id (required) |
| created | "2026-05-09T00:00:00.000Z" |
| modified | "2026-05-22T00:00:00.000Z" |
| reviewed | "2026-05-22T00:00:00.000Z" |
Take ownership of a pending task. Stamps the task with this agent's identity and marks it +ACTIVE so concurrent agents see it is in flight and /taskwarrior:task-coordinate skips it on the next wave.
| Use this skill when... | Use a sibling skill instead when... |
|---|---|
Starting work on a task surfaced by /taskwarrior:task-coordinate | Filing a brand-new task — use /taskwarrior:task-add |
Resuming a task after a context break (re-stamps started time) | Pausing for handoff without closing — use /taskwarrior:task-release |
Forcing a stale claim onto yourself with --force | The task is done and ready to land — use /taskwarrior:task-done |
task --versionfind . -maxdepth 1 -name '.git' -print -quithostnametask _udasGit probes (git rev-parse --show-toplevel, git branch --show-current)
write to stderr in a no-git cwd, and stderr from a Context backtick
aborts the skill before its body runs. Git toplevel and current-branch
resolution are done in the body via the Bash tool (where
2>/dev/null and exit-code handling are tolerated).
Parse $ARGUMENTS:
$0 — task ID (required).--no-coworker-marker — skip the /git:coworker-check --claim step (use when the repo is intentionally shared and another agent will claim its own scope).--force — claim even when the task is already +ACTIVE (the previous claim is overwritten with an annotation noting the takeover).Execute this claim workflow:
The identity UDAs (agent / pid / host / branch / worktree) are part of
the canonical set in the shared script. Check for missing UDAs:
ensure-udas.shbash "${CLAUDE_SKILL_DIR}/../../scripts/ensure-udas.sh" --check
If it reports UDAS_MISSING greater than 0, confirm with the user (declarations
persist in ~/.taskrc) and install — the script is idempotent and installs the
full linkage + identity set in one pass:
bash "${CLAUDE_SKILL_DIR}/../../scripts/ensure-udas.sh"
task "$TASKID" export | jq '.[0] | {id, uuid, description, status, start, agent, pid, host, branch, worktree}'
Use export | jq — never task <id> info or task <id> list (both exit 1 on empty / closed tasks and cancel parallel siblings; see .claude/rules/parallel-safe-queries.md).
Capture the uuid field as $TASK_UUID — every mutating call from Step 3
onward addresses the task by $TASK_UUID, never $TASKID. $TASKID may be a
numeric ID that renumbers if another agent closes an unrelated task during the
coworker-check pause in Step 3; $TASK_UUID does not (see
.claude/rules/task-id-stability.md).
Read the JSON. Decide:
| Condition | Action |
|---|---|
status is not pending | Abort. Tasks must be pending to claim. |
start is set and --force not passed | Abort. Report current agent / pid / host / branch so the user can either --force or pick another task. |
start is set and --force passed | Annotate with takeover notice (Step 6) and proceed. |
start is unset | Proceed cleanly. |
Unless --no-coworker-marker was passed, write the session marker so destructive git ops in this clone (stash, reset, checkout --) trigger the cross-agent guard:
Use SlashCommand to invoke `/git:coworker-check --claim`.
The git-side marker (<git-dir>/.claude-session-<pid>) and the taskwarrior +ACTIVE claim are independent signals. Together they reinforce: a stash hook can read either one, and either is enough to pause a destructive op.
If the SlashCommand tool is unavailable in the current session, fall back to running the script directly:
bash "$(git rev-parse --show-toplevel)/git-plugin/skills/git-coworker-check/scripts/claim-session.sh" --project-dir "$(pwd)"
Capture once and reuse across the next two steps:
| Value | Source |
|---|---|
AGENT | claude-${CLAUDE_SESSION_ID:0:8} (short, stable for the session) |
PID | $$ (the bash subshell PID — sufficient to cross-link with the git marker since both are taken at the same moment) |
HOST | hostname |
BRANCH | git branch --show-current (may be empty on a detached HEAD — leave the UDA unset rather than writing the literal "HEAD") |
WORKTREE | git rev-parse --show-toplevel |
Two calls — start first so +ACTIVE is set even if the modify step fails:
task "$TASK_UUID" start
task "$TASK_UUID" modify \
agent:"$AGENT" \
pid:"$PID" \
host:"$HOST" \
branch:"$BRANCH" \
worktree:"$WORKTREE"
Omit branch: when BRANCH is empty (detached HEAD). Quote every value — worktree: paths frequently contain spaces.
task "$TASK_UUID" annotate "claimed by $AGENT (pid $PID) on $BRANCH from $HOST:$WORKTREE"
When --force is used over an existing claim, annotate the takeover separately first so the previous owner's identity is preserved in the audit trail:
task "$TASK_UUID" annotate "force-claim by $AGENT — previous: agent=$PREV_AGENT pid=$PREV_PID host=$PREV_HOST"
Print:
agent / pid / host / branch / worktree/taskwarrior:task-release <id> to hand off without closing/taskwarrior:task-done <id> once the work has landed| Context | Command |
|---|---|
| Active filter (parallel-safe) | task +ACTIVE export | jq '.[] | {id, agent, branch, host}' |
| Single-field DOM read | task _get "$TASKID".agent (exit 0 even when unset) |
| Cross-project active scan | task status:pending +ACTIVE export | jq |
| Stale-claim probe | task +ACTIVE start.before:now-4h export | jq |
| Skip empty-result failures | Always export | jq, never task active list |
| Field | Type | Source on claim |
|---|---|---|
agent | string | claude-${CLAUDE_SESSION_ID:0:8} |
pid | numeric | $$ at claim time |
host | string | hostname |
branch | string | git branch --show-current (omit on detached HEAD) |
worktree | string | git rev-parse --show-toplevel |
start (built-in) | date | Set automatically by task start; cleared by task stop |
| Tag (virtual) | Set by | Cleared by |
|---|---|---|
+ACTIVE | task start | task stop / task done |
/taskwarrior:task-coordinate — surfaces unblocked candidates to claim/taskwarrior:task-release — release without closing (handoff)/taskwarrior:task-done — close after landing (auto-releases the claim)/git:coworker-check — sister signal: process-scan + session marker.claude/rules/agent-coworker-detection.md — combined-signal rationale.claude/rules/parallel-safe-queries.md — export | jq idiom.claude/rules/task-id-stability.md — why this skill resolves $TASK_UUID once and mutates by UUID