Proposal (rationale)

rungs worktrees answers one question: which worktrees hold uncommitted work on a branch that has already landed — “where work actually gets lost”, in its own words. F-057: when git status --porcelain fails in a worktree, the catch sets dirty = false, so the row prints merged · prunable, the label for the one state the command exists to warn about the opposite of. The runner’s own rule for an unreadable state is that it blocks; this command reports it as safe.

Decision

accepted — 2026-09-06 under WI-085. A failed read is an unknown with a reason, never a clean row.

Plan

Requirements

• A worktree row states clean, dirty or unknown; an unknown row carries the reason Git gave.
• The command’s output never prints prunable for an unknown row and says how many rows are unknown and why.
• Listing stays read-only; land’s holder checks and recovery behaviour are untouched.

Impacts

• src/concurrency.ts (WorktreeRow, worktrees()), src/cli.ts (cmdWorktrees), the existing regression in test/core.test.js.

Approach

Replace the boolean with a tri-state and keep the caught error’s first line as the reason. The regression removes a linked worktree’s directory from disk — Git still lists it, and status in a missing directory fails — so the failing read is real, not a fabricated result object.

Acceptance criteria / tests

1. A linked worktree whose directory was removed reports unknown with a reason containing the Git error, and the CLI line does not contain prunable.
2. The existing merged-and-dirty regression still passes with the new field names.
3. land tests are unchanged and pass.

Out of scope

• Repairing or pruning worktrees; the command reports and never removes.
• Any change to attribution or the compare-and-swap protocol.

Execution

Executed 2026-09-06 on feature/WI-089-truthful-worktree-state, cut from main bc8cc32 (WI-088 landed). WorktreeRow.dirty became state: 'clean' | 'dirty' | 'unknown' plus an optional reason (Git’s first line); the catch that set dirty = false now sets unknown and keeps the reason. cmdWorktrees prints an unknown row as status UNKNOWN with the reason beneath it, counts unknown rows as “could not be inspected” with the sentence “Unknown is not clean”, and computes the prunable count from state === 'clean' alone. land, preflight and session start do not read the row. No other consumer of the boolean existed (grep over src/ and test/).

Deviations. None.

Review

Against each acceptance criterion, 2026-09-06, Windows 11, Node v22.22.3:

1. Unknown with a reason. test/core.test.js “worktrees keeps a failed status read unknown…” adds a linked worktree, removes its directory (Git still lists it), and asserts state === 'unknown', a non-empty reason, and that the row has no dirty key. Through the CLI the row reads merged · status UNKNOWN feature/vanished, the reason line is printed, “1 worktree(s) could not be inspected” appears, and the word prunable does not.
2. Existing regression “worktrees reports merged-and-dirty…” passes with state in place of the boolean, and still asserts the worktree survives listing.
3. land untouched. No land, preflight or session start code changed; the full serial suite includes every land regression: node --test --test-concurrency=1 test/*.test.js after rebuild, 152 tests, 149 pass, 0 fail, 3 platform skips, 139 s.

node src/cli.ts check: 31 pass. Pending. The exact-SHA OS/Node matrix has not run: the branch is not pushed.