Pickup Task

Optional input: which task or plan to pick up (a task ID, a plan name, or a description).

This skill executes work that's already been planned. Use scripts/tome.py (tome help) to manage status changes — don't hand-execute git or status moves. tome is on PATH in Bash; if not found, fall back to python "$TOME_PLUGIN_ROOT/scripts/tome.py" . It resolves which vault via --vault / walking up from cwd / VAULT_ROOT; bare paths like wiki/SCHEMA.md are relative to the vault root, not the plugin root.

The user is always happy to answer questions. If intent or scope is unclear — which task, how far to take it, an ambiguity in the plan — ask.

1. Prep on Haiku (gate). Steps 2–3 are mechanical — a git pull, priming, and a task

lookup — and don't need a strong model. If you are not running as Haiku, stop and ask whether to switch to Haiku for the prep (the user may decline and stay on their current model — either is fine; you can't switch your own model, only they can). Already Haiku, or once they've answered: continue. The tier that matters for the actual work is a separate decision, made at step 4 once the task is known.

1. Prime yourself on the vault. Run tome sync to pull, then tome prime --full

(skip if already primed this session) — prints SCHEMA.md and the index in one shot.

1. Locate the task and its plan. If the user named a task or plan, find it. If they

named a milestone instead (an id like m-0 or its title — the epic layer, see wiki/SCHEMA.md), resume it rather than picking one task in isolation: run tome task task list --milestone --plain to list its open children (backlog.md already orders each status group by ordinal), then pick the next one — an In Progress child first (someone left it mid-flight), else the highest-priority To Do child, ties broken by list order. Confirm the pick with the user if it's not clear-cut. Selection data comes entirely from that list output; there's no separate tome command for this. With no name given at all, check the board (tome task task list --plain) and the project's live plans (plans/, not plans/archive/) and confirm which one they mean.

1. Check the agent-tier label. With the task located, read its agent: label

(tier: haiku — accepts either, resolving the other half if linked (a plan without a task, or a task without a plan, is normal). Sets the plan active, moves the task to In Progress (-a @me), logs work-started, syncs, then prints the task text and the full plan body as your working context — read it in full, follow its [[wikilinks]]`, and if you have concerns stop and discuss with the user before continuing.

1. Do the task work. Execute the plan in the relevant code repo (not the vault),

following that repo's CLAUDE.md. Stick to the plan's scope; if you hit a fork it doesn't resolve, ask rather than guess. Verify per the plan's verification section and report results honestly.

1. Commit, then present the work. Commit by default. Show the user what changed and the

verification results. The code repo is separate from the vault and follows its own commit conventions — check its CLAUDE.md. Push only after the user approves.

1. Close out the tracking. Once the work has landed: tome done --summary "..."

citing the shipping commit (--as superseded/--as abandoned instead of the default done, if that's how it landed). This archives the plan (moves it to plans/archive/, regenerates its project hub and the index), checks the task's acceptance criteria, closes and completes it with the summary, re-points its --ref at the archived path, logs done, and syncs — one command, no deferred sweep. Umbrella plans: if the task is one phase of a milestone plan shared by sibling phase tasks, tome done closes only that task and leaves the plan active (it prints how many open siblings kept it alive); the plan archives automatically when the last sibling closes. Don't try to close the shared plan slug while phases remain open — it's refused unless --force.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: chicken-noodle-chris
• Source: chicken-noodle-chris/tome
• License: MIT

Install and usage instructions live in the source repository linked above.