[P0] Break the author worktree bootstrap deadlock #892

New Issue

2026-07-24T21:36:47-05:00

jcwalker3 commented

2026-07-24 21:36:47 -05:00

Parent epic: #887
Priority: P0

Problem

gitea_bootstrap_author_issue_worktree raises Durable author worktree resolution blocked (#618): workspace is the stable control checkout on every clean issue pick, with and without an explicit worktree_path, in dry-run and live, reproduced across at least five daemon generations and on clean picks #662–#666. The wall is inside the tool, not in the capability gate — the gate returns allowed and the tool then refuses.

Every documented substitute is also closed:

gitea_bootstrap_author_issue_worktree -> RuntimeError (#618).
scripts/worktree-start -> No session issue lock is bound.
gitea_lock_issue -> #274 worktree path does not exist.
git worktree add -> denied by the harness.

Doors 2 and 3 form a literal cycle: the script demands a bound lock, and locking demands an existing worktree. That cycle is independent of #618.

A partial bypass exists and should inform the fix, not substitute for it: gitea_lock_issue against an already existing, clean, master-equivalent branches/* worktree succeeds, because lock_issue is not walled the way bootstrap is. That only works when such a worktree happens to exist.

Prerequisites

This issue is strictly upstream of the atomic author finalization child — a branch that cannot be bootstrapped cannot be finalized.

Acceptance criteria

An author session starting from the stable control checkout can obtain a usable issue worktree through a sanctioned MCP path, with no shell git worktree add and no pre-existing branches/* directory.
The lock<->worktree cycle is broken: at least one of the two operations can bootstrap the other's precondition.
The bootstrap refusal, where it remains correct, reports which precondition is missing and the exact sanctioned call that satisfies it — not a bare RuntimeError.
Regression: from a clean control checkout with an empty branches/, a full allocate -> worktree -> lock sequence completes.
Regression: the #274 worktree path does not exist branch is distinguished from "does not belong", since the two need different recovery.

Duplicate verdict

NOT a duplicate. Related to #618 / #850 (referenced by the error text itself) and #274. Filed because the four-door shape — every documented recovery path also closed — is not captured by any single open issue.

Parent epic: #887 Priority: P0 ## Problem `gitea_bootstrap_author_issue_worktree` raises `Durable author worktree resolution blocked (#618): workspace is the stable control checkout` on every clean issue pick, with and without an explicit `worktree_path`, in dry-run and live, reproduced across at least five daemon generations and on clean picks #662–#666. The wall is inside the tool, not in the capability gate — the gate returns allowed and the tool then refuses. Every documented substitute is also closed: 1. `gitea_bootstrap_author_issue_worktree` -> `RuntimeError` (#618). 2. `scripts/worktree-start` -> `No session issue lock is bound`. 3. `gitea_lock_issue` -> #274 `worktree path does not exist`. 4. `git worktree add` -> denied by the harness. Doors 2 and 3 form a literal cycle: the script demands a bound lock, and locking demands an existing worktree. That cycle is independent of #618. A partial bypass exists and should inform the fix, not substitute for it: `gitea_lock_issue` against an *already existing*, clean, master-equivalent `branches/*` worktree succeeds, because `lock_issue` is not walled the way bootstrap is. That only works when such a worktree happens to exist. ## Prerequisites This issue is strictly upstream of the atomic author finalization child — a branch that cannot be bootstrapped cannot be finalized. ## Acceptance criteria - An author session starting from the stable control checkout can obtain a usable issue worktree through a sanctioned MCP path, with no shell `git worktree add` and no pre-existing `branches/*` directory. - The lock<->worktree cycle is broken: at least one of the two operations can bootstrap the other's precondition. - The bootstrap refusal, where it remains correct, reports which precondition is missing and the exact sanctioned call that satisfies it — not a bare `RuntimeError`. - Regression: from a clean control checkout with an empty `branches/`, a full allocate -> worktree -> lock sequence completes. - Regression: the #274 `worktree path does not exist` branch is distinguished from "does not belong", since the two need different recovery. ## Duplicate verdict NOT a duplicate. Related to #618 / #850 (referenced by the error text itself) and #274. Filed because the *four-door* shape — every documented recovery path also closed — is not captured by any single open issue.

jcwalker3 added the status:ready workflow-hardening safety stale-runtime type:bug labels 2026-07-24 21:36:47 -05:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Scaled-Tech-Consulting/Gitea-Tools#892