Discover branches/review-pr<N>[-submit] worktrees separately from author
cleanup, fail closed on dirty trees / open PRs / active reviewer leases,
and remove only assessed-safe scratch paths via reconcile.
Add canonical_comment_validator and wire it into issue comments, PR review
bodies, reconcile post_comment, and structured comment helpers. Workflow-changing
comments without complete next-action state are rejected before any API call.
Includes unit/MCP integration tests, runbook docs, and a final-report rule
blocking false "comment posted" claims when validation failed.
Treat inventory mergeable/head as advisory. Add classification helper and
MCP tool so author sessions re-pin the live PR head before creating a
conflict-fix worktree, skip stale inventory false-negatives, and prove
classification in final reports.
Share review-workflow-load and review-decision-lock state across MCP
daemon process pools via a user-private durable store keyed by profile
identity (not host-global /tmp), with TTL and fail-closed profile checks.
Keep _fetch_pr_lease_comments_safe for gitea_assess_conflict_fix_push
only. Restore _list_pr_lease_comments to a single comments GET so merge
approval feedback and #485 non-list handling keep stable API call order.
Closes#519
Resolve PR via pulls API before fetching lease comments so
gitea_assess_conflict_fix_push fails closed with actionable reasons
instead of surfacing HTTP 500 when Gitea returns not-found errors.
Add regression tests and document assessment-failure handoff in work-issue.
Bare remote=prgs resolved to the hardcoded REMOTES default
Scaled-Tech-Consulting/Timesheet instead of the Gitea-Tools repository
the local git remote points at, causing false 404s and risking mutation
of a different repository.
Add remote_repo_guard.assess_remote_repo_match (pure) + formatter, wired
into gitea_mcp_server._resolve so every read/lookup/mutation tool fails
closed when the MCP-resolved org/repo disagrees with the local git remote
URL and the caller did not pass explicit org/repo. The guard is
best-effort: it skips when both org and repo are explicit, when the local
remote URL is unavailable, and is bypassed under pytest unless
GITEA_FORCE_REMOTE_REPO_CHECK is set.
Add tests/test_remote_repo_guard.py (pure-function cases + server wiring
proving a lookup tool cannot silently query a different repository), and
update author/reviewer/merger handoff templates to require explicit
remote/org/repo.
Closes#530
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
The second direct call after _verify_role_mutation_workspace caused
_clear_preflight_capability_state() to run twice (or between checks),
clearing valid capability/preflight state. This forced temp local
patches during #542/#517 merge/reconcile.
- Remove the redundant verify call (the _verify already performs
purity verification with correct worktree_path).
- Update test to reflect single invocation path and add explicit
test proving state preservation, single call, no dupe clear,
and no need for raw/temp fallbacks.
- Reconciler/controller flows stay fully MCP-native.
Fixes#548
resolve_task_capability("comment_issue") stamps _preflight_resolved_role
= "author" because comment_issue.required_role_kind = author.
_effective_workspace_role() prefers that stamp, so a genuine
prgs-reconciler session was classified as an author inside the #274
branch-only mutation guard and the #475 root checkout guard, blocking
gitea_create_issue_comment from the control checkout.
Fix keys the role exemptions off the actual active profile role as well
as the effective workspace role:
- Add _actual_profile_role(): derives the workspace role from the active
profile alone, never from _preflight_resolved_role.
- _enforce_branches_only_author_mutation: exempt when EITHER the
effective role OR the actual profile role is a non-author role.
- _enforce_root_checkout_guard: pass actual_role; assess_root_checkout_guard
now exempts a reconciler by resolved OR actual role.
Author blocking is preserved: an actual author profile classifies as
author in both signals, so it stays blocked on a contaminated control
checkout. Merger strictness stays keyed on the resolved task role so a
merger operating from its clean branches/ workspace under a non-merge
task is not over-tightened.
Regression tests (tests/test_issue_540_comment_role_poison.py) cover the
reconciler comment path, author-blocking path, reviewer/merger/reconciler
role classification under a poisoned author stamp, and the root guard
actual_role exemption.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Extend the review workflow load gate with pre-review command classification,
boundary violation tracking, and structured helper-result requirements in
final reports. Adds gitea_record_pre_review_command MCP tool.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Replace manual in-process _SESSION_LEASE seeding with an auditable
gitea_adopt_merger_pr_lease tool that posts durable adoption proof and
records sanctioned session provenance. Mutation gates now reject bare
record_session_lease calls without provenance from acquire/adopt/heartbeat.
Closes#536
Refuse merge-oriented reviewer-lease acquisition/adoption on an already
merged/closed PR, and add a read-first cleanup path for a lingering moot lease.
- reviewer_pr_lease.assess_acquire_lease: new pr_merged_or_closed flag; fail
closed with a post_merge_moot reason and no lease body when the PR already
merged/closed.
- reviewer_pr_lease.assess_post_merge_moot_lease: pure assessment. Only treats a
lease as moot when the PR is merged/closed; never proposes touching an active
lease on an open PR; provides a terminal phase:released (blocker:
post-merge-moot) body to neutralise the moot lease append-only; idempotent
once the released marker is the newest entry.
- gitea_acquire_reviewer_pr_lease: fetch live PR state, pass the flag, surface
post_merge_moot on refusal (no comment posted).
- gitea_cleanup_post_merge_moot_lease: new read-first tool. Reports merged
state, merge_commit_sha, linked-issue closure, lease-moot, cleanup
performed/skipped, and no_merge_or_adoption. apply=True posts the released
marker only when merged/closed with an active lease; refuses on open PRs.
- tests/test_post_merge_moot_lease.py: 10 tests covering acquire refusal on
merged PR (no post), safe/idempotent cleanup, and open-PR foreign-lease
protection.
No relaxation of #407 reviewer-lease or #16 gated-merge gates for open PRs.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Move @mcp.tool() from internal _list_open_pulls helper onto
gitea_lock_issue so author sessions can discover and call the
issue lock required by gitea_create_pr.
Add regression tests for MCP registration and create_pr lock flow.
Namespace-scoped workspace resolution prevents GITEA_AUTHOR_WORKTREE from
poisoning reviewer, merger, and reconciler purity checks. Each role uses
its own worktree env var with actionable binding-source errors and safe
reconnect guidance. Preserves #274/#475 author and control-checkout guards.
Read-only gitea_whoami calls no longer clear a valid capability proof when
whoami was already verified clean. Capability is consumed on mutation (one
resolve per mutation), and verify_preflight_purity rejects task mismatches.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Stale worktrees accumulate under branches/ when runs stop early, race a
sibling session, hit a validation failure, or lose cwd state, making later
workflow decisions harder and riskier.
Changes:
- worktree_cleanup_audit.py — ownership metadata, safety-first classifier
(active_open_pr, active_issue_work, dirty_local_worktree,
clean_stale_removable, detached_review_leftover, unsafe_unknown), TTL
expiry, per-worktree removal proof, and success-completion cleanup plan.
Only clean_stale_removable and detached_review_leftover are removable;
dirty/PR/leased/protected/unknown worktrees are never auto-deleted.
- gitea_audit_worktree_cleanup — read-only MCP tool that classifies every
branches/ entry, fetches live open-PR heads (fails closed if unavailable),
and returns counts, removable candidates, and git worktree list proof.
- work-issue.md 22A + review-merge-pr.md 28 — cleanup/TTL workflow rules.
- tests/test_worktree_cleanup_audit.py — 30 cases covering all nine
acceptance scenarios plus inference, metadata, TTL, and report accuracy.
Validation:
/Users/jasonwalker/Development/Gitea-Tools/venv/bin/python \
-m unittest tests.test_worktree_cleanup_audit tests.test_merged_cleanup_reconcile -q
38 passed.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Introduce root_checkout_guard helper and enforce it in verify_preflight_purity
so author, reviewer, and merger mutations fail closed when the stable control
checkout is on the wrong branch, detached, dirty, or diverged from prgs/master.
Isolated branches/... worktrees and reconciler paths remain allowed.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Normal author work is unchanged: gitea_lock_issue still requires the worktree
to be base-equivalent to master/main/dev, and gitea_create_pr still targets a
normal base branch. A *stacked* PR (based on another unmerged PR's branch) is a
new explicit, proof-backed path — it never bypasses the issue lock.
- stacked_pr_support.py: pure policy. Approve a non-master base only when it is
declared AND owned by a live OPEN PR whose number matches; reject arbitrary,
mismatched, or stale (merged/closed) branches; require the PR body to document
the stack (base branch, stacked-on PR, merge ordering).
- issue_lock_worktree.py: read_worktree_git_state / _find_matching_base_ref gain
an opt-in extra_bases arg so an approved stacked base can anchor
base-equivalence in addition to master/main/dev (empty by default = unchanged).
- gitea_mcp_server.py:
- gitea_lock_issue gains stacked_base_branch + stacked_base_pr. When declared,
it verifies the open PR owns the branch, anchors base-equivalence to it, and
records approved_stacked_base = {branch, pr_number, verified_open} on the lock.
- gitea_create_pr validates a non-master base against the lock's approved
stacked base, re-checks the dependency PR is still open, and enforces the
stacked-PR body fields. Master-based path untouched.
- Docs: llm-workflow-runbooks.md and work-issue.md document when stacked PRs are
allowed and the required PR-body wording.
Tests: test_stacked_pr_support.py (18 policy cases), stacked base-equivalence in
test_issue_lock_worktree.py, approved_stacked_base persistence round-trip in
test_issue_lock_store.py. Existing lock/create_pr regressions still pass.
The #482 -> #479/#478 case motivates this: create_pr previously could not open a
stacked PR because the lock required a master-equivalent worktree.
Closes#484.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
sysadmin
