# Pick

> Rank the next ticket to pick up from the current repo's issue tracker. Reads issue_tracker and pick_buckets from repo-policy.json, shows a "Which bucket?" menu, and prints a top-5 list with copy-paste start lines. Read-only, terminal output only. Use when the user types /pick.

- Skill: `eric-lingren/pick` (Agent Skill)
- Install (CLI): `npx get-skill add eric-lingren/pick`
- Raw SKILL.md: https://api.skillmd.com/api/skills/eric-lingren/pick/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Eric-Lingren (https://skillmd.com/u/eric-lingren)
- Updated: 2026-11-19
- Page: https://skillmd.com/skills/eric-lingren/pick

---


# Pick

Thin orchestrator: resolve repo, choose a bucket, run a fetch script, render. Strictly read-only. No writes to GitHub or Linear, no report file. All logic lives in `~/.dotfiles/claude-code-shared/scripts/pick/`.

## Process

1. Resolve the repo and its config:

   ```bash
   bash ~/.dotfiles/claude-code-shared/scripts/pick/resolve-repo.sh
   ```

   If the output is not JSON (it reads `no issue_tracker configured for <repo>`), print it verbatim and stop. This is a clean exit, not an error.

2. Otherwise parse `{repo, issue_tracker, buckets}`. If `issue_tracker` is `linear`, use `pick-fetch-linear.sh` in step 4 instead of `pick-fetch-gh.sh` (same output shape; start line uses Linear's gitBranchName). Say "sprint", never "cycle", in anything shown to the user.

3. Choose the bucket. Free text after `/pick` (e.g. `/pick quick wins`) is mapped by plain-English match and skips the menu:

   ```bash
   python3 ~/.dotfiles/claude-code-shared/scripts/pick/map-bucket.py "<free text>" --repo "<repo>" <bucket> [<bucket> ...]
   ```

   Exit 0 prints the bucket to use. Exit 1 (no match) or bare `/pick`: ask "Which bucket?" with AskUserQuestion, one option per entry in `buckets` (max 4). Describe Linear buckets as: `my-current-sprint` "Tickets assigned to you in the current sprint", `my-next-sprint` "Tickets assigned to you in next sprint (same rules as my-current-sprint)", `unclaimed` "Unassigned tickets in the sprint pool".

3b. Focus (shared-pool buckets only): run `bash ~/.dotfiles/claude-code-shared/scripts/pick/focus-options.sh "<repo>" "<bucket>"`. If `ask` is false, skip. Otherwise ask "Focus?" with AskUserQuestion using `options` (projects from your current sprint, then "No focus"; the free-text "Other" is built in). Add `--focus "<choice>"` to both `rank-render.py` calls in step 4 (omit for "No focus"). Focus only boosts matching tickets to the top; it never filters and is never saved.

4. Fetch, score, and render:

   ```bash
   # issue_tracker github:
   bash ~/.dotfiles/claude-code-shared/scripts/pick/pick-fetch-gh.sh "<repo>" "<bucket>" > "$TMP/cands.json"
   python3 ~/.dotfiles/claude-code-shared/scripts/pick/blockers.py resolve "$TMP/cands.json" > "$TMP/r.json" && mv "$TMP/r.json" "$TMP/cands.json"  # blocker tiers: ready, stackable (wt <branch> <base>), blocked (hidden)
   # issue_tracker linear:
   bash ~/.dotfiles/claude-code-shared/scripts/pick/pick-fetch-linear.sh "<repo>" "<bucket>" > "$TMP/cands.json"
   python3 ~/.dotfiles/claude-code-shared/scripts/pick/rank-render.py "$TMP/cands.json" --scorer-input > "$TMP/scorer-in.json"
   ```

   The fetch script emits normalized candidate JSON (shape documented in its header, checked by `validate-candidates.py`). `--scorer-input` applies the cheap pre-sort (blocked hidden, label tier) and caps the list at the top 25.

   Spawn the `pick-scorer` agent with the contents of `scorer-in.json` as its only input. Save its JSON reply to `$TMP/scores.json` and check it:

   ```bash
   python3 ~/.dotfiles/claude-code-shared/scripts/pick/validate-scorer.py output "$TMP/scores.json" --input "$TMP/scorer-in.json"
   ```

   If the reply is invalid or the agent fails, render without scores (below). Then render:

   ```bash
   python3 ~/.dotfiles/claude-code-shared/scripts/pick/rank-render.py "$TMP/cands.json" --scores "$TMP/scores.json"
   ```

   This re-ranks by effort score, prints the top 5 in two sections (`ready`: no open blockers; `waiting on blockers`: blocker PR in review, not merged, with a `blocked by:` line), with a `why:` line per item and a start line `wt <branch>  ->  /grill-me <ID>` (`#N` for GitHub, Linear ID plus `gitBranchName` for Linear). Tickets from a different repo are kept and shown in an `other repos` section below the main list (up to 3, tagged with their repo). Unpointed tickets show `⚠ unpointed` and are never excluded. FE work is preferred: the scorer tags each ticket's `stack` (fe/be/full/infra) and `stack_pref.py` adds a penalty to backend/infra tickets in both sorts (title FE/BE heuristic when no scores). The stack shows at the end of each item line. A bucket with `"start": "grill-me"` (dotfiles `ideas`) prints `start: /grill-me <ID>` with no `wt` worktree, and its `seen-again` sort ranks tickets by their count of comments beginning `seen again:`. Without `--scores` the pre-sort order is used and no `why:` lines print.

   Linear only (required, before the first `rank-render.py` call): run `pick-fetch-linear-activity.sh > "$TMP/activity.json"` and add `--activity "$TMP/activity.json"` to every `rank-render.py` call (excludes tickets whose parent has a teammate active; annotates project overlap with 👥).

5. Print the script output to the terminal as-is. Do not edit, assign, label, or comment on any issue. The user starts a new session and grills the ticket manually.

<!-- learning-capture:start -->
Read and execute `~/.dotfiles/claude-code-shared/resources/learning-capture.md`.
This skill's slug is `pick`.
<!-- skill-done: pick -->
  - `/grill-me` — ticket chosen, start a new session and grill it
<!-- learning-capture:end -->

