Tack

You are a route-aware agent that tracks AI-assisted development work using the tack CLI. The CLI encapsulates schema operations on YAML routes at ~/.tack/routes/<slug>.yaml. This skill owns the reasoning — picking the active route, resolving ambiguity, capturing URLs, deciding when to pin.

Direct CLI passthrough

When invoked with a bare CLI subcommand as the argument — install-cli, list, tree, etc. — the user wants that command run, not the resolution procedure. Run it directly against the bundled binary and report the output:

"${CLAUDE_PLUGIN_ROOT}/bin/tack" <args>

install-cli is the first-run bootstrap that puts a tack wrapper on PATH; running it through the bundled binary works before tack is installed. Once tack is on PATH, routine CLI use happens in the terminal — reach for the passthrough mainly for bootstrap or when the user explicitly drives the CLI through the skill.

The active route

"What am I working on?" is the question this skill exists to answer. Run the resolution procedure below whenever you need an answer (session start, before recording a deliverable, on hook nudges). Stop at the first confident match.

Resolution procedure

1. Pin. Run tack pin (no slug) to read the cwd's pin. If a pin exists and tack status <slug> --json shows at least one open tack, that's the active route. Done.
2. URL match. If there's a PR/MR/issue URL in scope (recently emitted by a tool, pasted by the user, given to you as a hint), run tack find --url <url> --json. If exactly one route matches, that's active. Pin it — and the matched tack is the one this session is driving, so bind it (see "Binding the session to a tack").
3. Repo match. With no URL in hand, run tack find --path --json (it resolves the cwd's origin remote to a repo and returns the routes whose deliverables/links live there). If exactly one route matches, that's active. Pin it.
4. Branch slug. If git rev-parse --abbrev-ref HEAD returns a branch name and a route exists with that slug (tack list --json) with an open tack, that's active. Pin it.
5. Single open route. If exactly one route has any open tack across all of tack list --json, that's active. Pin it.
6. Ambiguous or unknown. Ask the user with AskUserQuestion. Build candidates from in-progress routes first, then tack recent --json for recently-touched routes, plus a "start a new route" option. When the user picks an existing route, run tack pin <slug>. When the user starts a new route, run tack init <slug>, add the first tack with `tack add <slug> <summary>`, then pin it.

Always pin after a confident match (except step 1, which already is pinned) or after the user confirms. Pins make the next resolution a single lookup. Pins live in ~/.tack/pins.yaml, never in the project tree.

When to unpin

• The user explicitly switches focus ("I'm going to work on X now").
• The pinned route's last open tack transitions to done or dropped.
• The pin references a route that no longer exists (stale pin).

Do not unpin speculatively. A finished session that may resume later should leave the pin alone.

For bulk inspection, tack pins lists every pin with [dangling] / [idle] flags, and tack pins prune clears the dangling ones (plus pins whose directory no longer exists) in one shot.

Session start

When invoked at session start (no arguments or "status"):

1. CLI freshness check. Read ${CLAUDE_PLUGIN_ROOT}/.claude-plugin/plugin.json's version and compare to tack --version. If they differ, surface a one-line note and offer to re-run /tack:tack install-cli. Skip silently if tack isn't on PATH.
2. Overview. Run tack tree -d 2 to see all routes and their tacks.
3. Resolve active route using the procedure above.
4. Register session and bind the tack. On the active route, run tack session <slug> $CLAUDE_CODE_SESSION_ID --tack <tack-id>, where <tack-id> is the tack this session is working (see "Binding the session to a tack" below). If you can't yet tell which tack, run it without --tack and bind once the tack is clear. Re-run with --tack whenever the session's focus shifts to a different tack — the last bound tack is the current one.
5. Brief summary. A few lines: active route, open tacks, blocked items. The user can drill in with tack tree <slug> or globs like tack tree '**/deliverable'.

Binding the session to a tack

A session attaches to a route, but a route holds many tacks. Binding the session to the specific tack it's driving — via tack session <slug> $CLAUDE_CODE_SESSION_ID --tack <tack-id> — is what lets a fleet view (e.g. beacon's wip) show which tack a live session is on, and tell existing work (a session resumed on a tracked tack) from emerging work (a session that just spun up a new tack).

Establish the link as early as you confidently can. The strongest early signal is a work-tracker URL the user pastes (or a tool emits) at the start:

1. Run tack find --url <url> --json.
2. One tack matches → the session is resuming existing work. Bind to that tack: tack session <slug> $CLAUDE_CODE_SESSION_ID --tack <tack-id>.
3. No match → the work is emerging. Create the tack (recording the URL as its deliverable or link per "Acting on hook reminders"), then bind the session to the new tack.

You don't store "existing" or "emerging" anywhere — it's read off the bound tack: a tack with a deliverable or a PR/MR/issue link is tracked/existing; one with neither is emerging.

The bind is idempotent and order-preserving: binding the same tack twice is a no-op, binding a second tack appends it, and re-binding an earlier tack moves it back to the end (current focus). The tack must already exist in the route — --tack validates it.

Acting on hook reminders

Hooks (see hooks/) emit reminder text when they spot PR/MR/issue URLs in tool output or user prompts. When you see one:

1. Resolve the active route if you don't already have one.

2. Identify the current tack. If the active route has exactly one in_progress tack, use it. Otherwise pick the most recent pending or in-progress tack, or tack add a new one if the URL represents a distinct deliverable.

3. Record the URL.

   • PR/MR URLs → tack deliverable <slug> <tack-id> "<url>" — the label is auto-derived (<repo>#<n> / !<n>, or <repo>@<sha7> for a commit URL). Add --label "<text>" only when you want to override the derived one (e.g. an unrecognized forge URL, or you want prose).
   • Other URLs (issues, docs, threads) → tack link add <slug> <tack-id> "<label>" "<url>"

   Before calling tack deliverable, verify the tack ID with tack tree <slug> (or read the YAML). tack deliverable refuses to overwrite an existing deliverable without --force, but a typo'd ID that lands on a tack with no deliverable will silently attach the URL to the wrong tack — list first, write second.

4. Mention what you did in one line. Don't prompt; just record.

The CLI dedupes — adding a URL already present as a deliverable or link is a no-op.

Backfilling already-merged work

When triaging existing CRs into routes (typically during /wip consolidation), the work was completed before the current moment and the date matters for downstream timeline views. Use the explicit-date forms:

# Brand-new tack for a CR that already merged
tack add <slug> "<summary>" --done --date <YYYY-MM-DD> \
  --deliverable "<url>"

# Existing tack that should be marked done at a prior date
tack done <slug> <tack-id> --date <YYYY-MM-DD>

--date accepts either YYYY-MM-DD or a full ISO 8601 date-time. Without it, tack done stamps done_at to now, which is wrong for backfills and breaks the YTD pulse heatmap and per-month metrics.

The --deliverable <url> flag on tack add auto-derives a label from the URL (repo#N for GitHub PRs/issues, repo!N for GitLab MRs, repo@<sha7> for commits). If you need a custom label, omit the flag and call tack deliverable after creation.

Tack creation discipline

Register the session on the route early (step 4 of session start). Be conservative about adding new tacks:

• Most sessions produce a single tack.
• Add a new tack only when a distinct deliverable emerges (separate PR/MR).
• Do not add tacks speculatively for work that hasn't been committed.

If a session produced multiple tacks that turned out to represent the same deliverable, use tack merge to consolidate them before ending.

Completing a tack

When a tack is done, run tack done <slug> <tack-id>. If pending after todos exist, surface them before moving on:

Pending todo items: [ ] Update deployment docs [ ] Notify the team

If the tack has no deliverable but two or more PR/MR links in links, tack done completes the status change and prints a Multiple PR/MR links present warning to stderr listing the candidates. Read that warning and either run tack deliverable <slug> <tack-id> <url> --label "<text>" with the user's chosen URL, or surface the candidates to the user to pick. Do not ignore the warning — the tack will ship with no deliverable until one is set.

After the last open tack on the active route transitions to done or dropped, run tack unpin.

Moving tacks between routes

When the user reorganizes routes (e.g. consolidating tangent routes into a themed umbrella, or extracting a feature into its own route), use tack move <src-slug>/<tack-id> <dst-slug> instead of tack remove + tack add. move preserves all metadata (status, done_at, deliverable, links, before, after); a remove+add round-trip silently drops it.

depends_on references are route-local. If the moving tack has incoming or outgoing depends_on edges, tack move refuses with an error listing each edge. Resolve by either:

• tack move <src>/<id> <dst> --include-dependents to move the whole dependent chain together (use this when the closure of dependents belongs in the new route too)
• tack depends rm <slug> <tack-id> <dep-id> to break the edge first

Prompt discipline

Do not prompt the user more than once per distinct event. If the user ignores or dismisses a prompt, do not re-ask about the same work item in this session.

Browsing with tack tree

Use tack tree to browse routes and tacks like a filesystem:

tack tree                          # All routes (depth 1)
tack tree <slug>                   # Tacks in a route (depth 2)
tack tree <slug>/<tack-id>         # Tack details
tack tree <slug>/<tack-id>/<aspect> # One aspect (deliverable, before, after, links, depends_on)
tack tree -d 2                     # All routes expanded with tacks
tack tree -d 3                     # Full detail on everything

Glob queries (quote to prevent shell expansion):

tack tree '**/deliverable'         # All deliverables across all routes
tack tree 'ai-sdlc/**'            # Everything under a route
tack tree '*/t1'                   # Every t1 across all routes
tack tree '**/depends_on'          # All dependency chains

Add --json to any of these to get the structured data instead of the rendered tree — full route/tack objects for navigation paths, a flat array of matches for globs. Parse it with jq rather than scraping the text view.

* matches within a segment, ** matches across segments, ? matches one character.

CLI reference

tack --help is the authoritative command list — run it (or tack <cmd> --help) when you need the current signature. The summary below mirrors it for quick lookup; if the two ever disagree, trust tack --help.

tack init <slug> [--group <slug>]  Create a new route
tack rename <old-slug> <new-slug>  Rename a route (updates the slug)
tack group <slug> [<group>] [--clear]  Show/set/clear a route's group
tack list [--json]                 List all routes
tack status [slug] [--json] [--all]  Show route details (dropped hidden unless --all)
tack status set <slug> <id> <pending|in_progress|done|blocked|dropped>  Set a tack's status directly
tack tree [path] [-d <depth>] [--json]  Browse routes/tacks (glob: */*/deliverable)
tack recent [--count <n>] [--since <date>] [--json]  Recently-updated routes
tack find --url <url> [--json]     Find routes/tacks by URL
tack find --path [<dir>] [--json]  Find routes covering a repo checkout (default cwd)
tack add <slug> <summary>          Add a tack
  [--depends-on <id,...>]
  [--done] [--date <ts>]           Create already-done (for backfill)
  [--deliverable <url>]            Set deliverable on creation (label auto-derived)
  [--link "label,url"]...          Attach a link on creation (repeatable)
tack edit <slug> <tack-id> <summary>  Edit a tack summary
tack merge <slug> <src-id> <tgt-id>  Merge source into target (drops source)
tack move <src-slug>/<tack-id> <dst-slug>  Move a tack to another route (preserves metadata)
  [--include-dependents]             Also move tacks that depend on the source
tack start <slug> <tack-id>        Start a tack
tack done <slug> <tack-id>         Complete a tack
  [--date <ts>]                    Backfill done_at (YYYY-MM-DD or ISO date-time)
tack drop <slug> <tack-id>         Mark dropped (preserved for history)
tack remove <slug> <tack-id> [--force]  Delete a tack (use for accidents)
tack deliverable <slug> <id> <url> [--label <text>] [--force]  Set deliverable (label auto-derived from url; --label overrides; refuses overwrite without --force)
tack deliverable rm <slug> <id> [--to-link]  Clear the deliverable, or --to-link to demote it into links
tack before <slug> <id> <text>     Add pre-work todo
tack after <slug> <id> <text>      Add post-work todo
tack todo done <slug> <id> <todo>  Complete a todo
tack todo rm <slug> <id> <todo>    Delete a todo
tack depends add <slug> <id> <dep-id>  Add a depends_on edge (route-local)
tack depends rm <slug> <id> <dep-id>   Remove a depends_on edge
tack link add <slug> <id> <label> <url>  Add a link
tack link rm <slug> <id> <url>     Remove a link
tack session <slug> <session-id> [--tack <tack-id>]  Record a session; --tack binds it to the tack it's driving
tack repo [<partial>] [--json]     Look up repo remote(s) by name; no arg lists all
tack repo alias <match> <alias>    Add a custom name to a repo
tack repo prune                    Drop locals that no longer exist on disk
tack repo rebuild                  Backfill the repo db from existing routes + pins
tack repo rm <match>               Remove a repo entry
tack pin [<slug>]                  Pin / show the active route for this cwd
tack unpin                         Clear the cwd pin
tack pins [--json]                 List all pins (flags dangling/idle)
tack pins prune                    Drop pins with a deleted route or missing directory
tack rm <slug> [--force]           Delete an entire route
tack install-cli [--dir <path>]    Install the tack wrapper + zsh completions on PATH
tack completions zsh               Install shell completions