fix(tracker): scan candidates independent of --limit; validate --status column (#66, #67)

[pgebheim]

Closes #66. Closes #67.

Two related bugs in scripts/rig-tracker.sh verb_select, both surfacing as a silent empty result that reads like "no work" when there is work / the config is wrong.

#66 — --limit truncated candidates before the intersection

verb_select fetched both sources — the per-label issue queries and the board column — with the result --limit, then intersected. So --limit meant "consider N candidates", not "return N results", and next (hardcoded --limit 1) pulled one issue per shape label and one board item; their intersection is empty unless that single board item happens to be the dispatchable one. On a real board, next returned [] with dispatchable work sitting in Todo.

Fix: decouple scan depth from result count. Pull SCAN_DEPTH (default 500, RIG_TRACKER_SCAN-overridable) rows from each source, intersect, then return the first <limit>.

#67 — select --status <name> silently [] for a non-existent column

The status was used only as a client-side jq equality filter, so a column name not on the board (a typo, or drift after a Projects column rename) matched zero rows → [], exit 0 — indistinguishable from "no work in that column". set-status already validates option names; select didn't.

Fix: assert_valid_status checks the name against the field's real options (same source set-status uses) and dies otherwise. A valid-but-empty column still returns [] / exit 0 — the two cases are now distinguishable.

Tests

scripts/rig-tracker.test.ts: the mock gh now honors --limit and answers project field-list, and the board returns a non-Todo item first so the truncation is observable. Added:

• rig-tracker: next returns [] with a board configured — limit truncates candidates before the column intersection #66 regression — next finds the dispatchable item even though it isn't first on the board.
• rig-tracker: select --status silently returns [] for a column that doesn't exist on the board #67 — an unknown column errors (exit ≠ 0, "not a column"); a valid-but-empty column (In Review) is a clean [] / exit 0.

Verified these 3 assertions fail on the pre-fix script and pass after (13/13 green with the fix).

🤖 Generated with Claude Code