McRitchie Studio

Agent Context

Give bin/agent-worktree remove a MERGE-VERIFIED --force escape hatch. PROBLEM: run_remove (bin/agent-worktree ~L1690) refuses on three blockers — dirty, 'branch content is not represented on <base>' (via cleanup_ready?), and not-git-registered. cleanup_ready?(record) = !dirty && (merged || equivalent_to_main) where :merged is git-ancestry — which SQUASH MERGES break. So worktrees whose PR was squash-merged into main under the old pre-release-branch flow are permanently un-removable (we have 8 such: PRs #76/#82/#83/#85/#86/#89 in mcritchie-studio, #159/#160 in turf-monster — all MERGED on GitHub but content-not-on-origin/release). There is NO override flag today. FIX: add --force to the 'remove' dispatch (parse like --yes at ~L2116: force = ARGV.delete('--force'); pass run_remove(app, task, yes:, force:)). In run_remove, --force SUPPRESSES ONLY the 'content is not represented on <base>' blocker — and ONLY when MERGE-VERIFIED: the branch has a MERGED PR. Verify via gh: the script already has github_repo_slug(dir) (returns owner/repo) and command_available?('gh'); use (a merged PR exists => verified). If --force AND a merged PR is found => clear that blocker, print '✓ <branch> PR #N merged — overriding content guard (--force)', proceed (still respect --yes for the actual teardown). If --force BUT no merged PR / gh unavailable / no github remote => DO NOT remove: keep the blocker + clear message ('--force requires a merged PR; none found for <branch>' or 'gh unavailable'). --force MUST NOT suppress the dirty or not-registered blockers (genuine data-loss / wrong-state — never force those). TESTABILITY (critical — mirror dor-check's DOR_CHECK_CHANGED_FILES injection): make the merged-PR check injectable so tests don't hit real gh — e.g. honor an env override like AGENT_WORKTREE_MERGED_PR (set => treat as merged; ''/unset => not merged) that short-circuits the gh call, OR extract a pure decision method force_clears_content_blocker?(record, force:, merged:) unit-tested directly. Update usage/help (L119) to document remove [--force] [--yes]. TESTS [unit]: the decision — force+merged clears content blocker; force+not-merged does NOT; force NEVER clears dirty or not-registered; no-force unchanged. [integration]: run_remove end-to-end (with the injection) over a temp worktree/record: force+merged removes; force+unmerged refuses; force+dirty refuses. Do NOT change cleanup_ready? semantics for the non-force path, or the reclaim/cleanup flows.