Breaks the author worktree bootstrap deadlock (#892) so an author session starting from a clean control checkout can obtain a usable branches/ worktree through the sanctioned MCP path.
Root cause
assess_author_issue_bootstrap returned allowed/proven for a clean control checkout, but the shared predicate bootstrap_permits_control_checkout only accepted create_issue assessments (task_scope=create_issue_only, empty reasons, full base-tip field set). Author assessments never satisfied the predicate, so #274/#604 guards kept the ordinary control-checkout block on every gitea_bootstrap_author_issue_worktree call.
Fix
author_issue_bootstrap.assess_author_issue_bootstrap — emit create_issue-compatible proof fields (task_scope=author_issue_bootstrap, empty reasons when allowed, base_tips_verified, binding paths, strict tip proof).
create_issue_bootstrap.bootstrap_permits_control_checkout — accept task_scope=author_issue_bootstrap for author bootstrap tasks (still fail-closed for cross-scope smuggling).
Bootstrap used the documented #892 partial bypass (pre-existing clean worktree + lock) to land this fix; post-merge, clean control → gitea_bootstrap_author_issue_worktree is the primary path.
Risk
Low — narrow waiver remains fail-closed (clean base, tip proof, binding match, empty refusal reasons). Does not widen create_issue scope.
Align assess_author_issue_bootstrap with bootstrap_permits_control_checkout
so gitea_bootstrap_author_issue_worktree can create the first branches/
worktree without the lock↔worktree deadlock.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
STATE: reviewed-approved
WHO_IS_NEXT: merger
NEXT_ACTION: Merge PR #926 into master from an independent prgs-merger session, then advance the control checkout and restart the MCP fleet.
NEXT_PROMPT:
Invoke the canonical gitea-workflow skill first. Using only the sanctioned gitea-merger namespace with profile prgs-merger and role merger, merge PR #926 in Scaled-Tech-Consulting/Gitea-Tools at head 2066623986df9a05c0f24bb760d7c0abb5bf0d9e into master. Verify before merging: fleet is exactly five MCP servers under one Claude parent, parity green with mutation_safe true, PR #926 open and mergeable, head unchanged at 2066623986df9a05c0f24bb760d7c0abb5bf0d9e, and the sysadmin APPROVE review is visible at that exact head with approval_at_current_head true. Acquire the merger lease, merge, then confirm issue #892 closed and the branch state. Do not author, review, or approve.
WHAT_HAPPENED: Performed an independent review of PR #926 at head 2066623986 in the bound review worktree branches/review-pr926-issue-892-bootstrap. Read issue #892, the complete three-file change set, the surrounding guard source, and the new tests. Confirmed the #892 deadlock is present at the base revision and resolved at this head, with every fail-closed protection preserved.
WHY: The scope proof in create_issue_bootstrap.bootstrap_permits_control_checkout accepted only task_scope create_issue_only, while assess_author_issue_bootstrap emitted no task_scope key at all. The predicate therefore could never return True for an author bootstrap, even though the task gate above it already admitted the author task. This change completes that half-applied fix without widening any other proof obligation.
ISSUE: 892
HEAD_SHA: 2066623986
REVIEW_STATUS: APPROVE
MERGE_READY: yes
BLOCKERS: none
VALIDATION: Focused suites at head (test_issue_892_author_bootstrap_deadlock.py, test_create_issue_bootstrap.py, test_issue_757_bootstrap_guard_agreement.py, test_author_issue_bootstrap.py) = 105 passed, 31 subtests. Broad regression over 33 bootstrap/worktree/guard/preflight suites, identical set at both revisions: base 2b4e430 = 10 failed / 599 passed / 96 subtests; head 2066623 = 10 failed / 613 passed / 96 subtests. Delta is +14 passed, equal to the 14 new tests; zero new failures. The 10 failures share identical test IDs at both revisions, are pre-existing, and each passes when run alone.
LAST_UPDATED_BY: sysadmin / prgs-reviewer
Reviewer identity sysadmin on profile prgs-reviewer; PR author is jcwalker3. Distinct accounts, no self-review. Base branch master at 6a56260768ad2f69b773a8b7ab8d460ce821c914; merge base 2b4e43042a34f4e29617378ae79a7f5a3d312688.
Both production modules are byte-identical between the merge base and the current master tip, so this analysis holds against master and the merge is clean on the touched files. Nothing in the change touches role permissions, credential resolution, or process launching, and it adds no manual worktree or branch-creation bypass.
Deadlock present at base, resolved at head
A single scenario matrix was evaluated read-only in the review worktree against both revisions.
#
Scenario
base 2b4e430
head 2066623
A1
clean control checkout, author bootstrap
False
True
A2
create_issue bootstrap
True
True
D1
dirty control checkout
False
False
D2
non-base branch
False
False
D3
detached HEAD
False
False
D4
HEAD does not match live master tip
False
False
D5
remote tip unresolvable
False
False
D6
remote tip missing, no error
False
False
D7
local HEAD sha unknown
False
False
D8
workspace is not the canonical root
False
False
D9
under branches/ worktree
False
False
D10
unrelated task literal
False
False
D11
ordinary author mutation (commit_files)
False
False
S1
author scope presented for a create_issue task
False
False
S2
create_issue_only scope presented for an author task
False
False
S3
hand-built allowed=True over a dirty tree
False
False
S4
hand-built base_tips_verified over mismatched tips
False
False
Exactly one behavior delta, A1. Every denial and every cross-scope or hand-built assertion is refused identically at both revisions.
Fix assessment
bootstrap_permits_control_checkout now applies the scope proof per task branch rather than as a widened disjunction: a create_issue task still requires create_issue_only, an author bootstrap task requires author_issue_bootstrap, and a final else closes unknown tasks. Both directions of cross-scope reuse fail closed (S1, S2). Every other obligation is unchanged and still applies to both branches — bootstrap_path must be clean_canonical_control_checkout, reasons must be empty, dirty_files must be empty, under_branches must be exactly False, tip equality is re-derived rather than trusted from the assessment's own flag, and the assessment must bind to the exact workspace and canonical root.
Ordinary author mutations stay blocked from the control checkout: the task gate admits only bootstrap_author_issue_worktree and gitea_bootstrap_author_issue_worktree, so every other author task leaves the ordinary block in force (D11).
assess_author_issue_bootstrap now produces the full predicate-compatible field set through one constructor. proven and base_tips_verified are derived inside that constructor rather than accepted from callers, and task_scope is fixed to author_issue_bootstrap, so it cannot mint a create_issue-scoped assessment.
The allowed path is strictly tighter than before. The prior logic compared tips only when both were present, so a missing local HEAD or a missing remote tip left reasons empty and yielded allowed=True with no base-equivalence proof. The new logic fails closed on an unknown local HEAD, on a resolver error, and on a missing remote tip (D5, D6, D7), closing a pre-existing fail-open.
The under_branches result changed from allowed=True to not_applicable=True. This is operationally inert: when the workspace is a valid branches/ worktree the ordinary guard does not block and returns before the bootstrap assessment is consulted, and on the durable-block path the predicate refused it at both revisions, since the prior dict carried no under_branches key and the is not False check rejected the resulting None. The new shape simply states the intent correctly.
Two-guard agreement from #757 is preserved: anti_stomp_preflight.py:599 routes the same threaded assessment through the same predicate under the same task name, so the #274 and #604 guards cannot diverge. The waiver stays narrow — only the wrong-worktree verdict is waived, while root-checkout, repo, role, stale-runtime and lease checks are each still evaluated.
The #892 criterion that a refusal must name the missing precondition and the sanctioned call is met: format_create_issue_bootstrap_error at create_issue_bootstrap.py:322-330 surfaces exact_next_action, and author block assessments populate it through EXACT_NEXT_ACTION_AUTHOR_BOOTSTRAP.
Non-blocking findings
tests/test_issue_892_author_bootstrap_deadlock.py:151-211 — TestAuthorBootstrapPreflightIntegration.test_enforce_branches_only_allows_clean_control_for_bootstrap is described as exercising the real enforcer wiring and binds the server module as srv at line 156, but never calls srv._enforce_branches_only_author_mutation. The durable_block dict built at line 192 is passed to nothing, and the assertion at line 211 holds by construction. The branch at lines 200-208 reduces to the assertion already made at line 113. The net effect is a duplicate unit assertion under an integration name, and no test drives _enforce_branches_only_author_mutation with an author bootstrap task — test_issue_757_bootstrap_guard_agreement.py:349 reaches the real enforcer but only for create_issue tasks. A follow-up should cover that path end to end. Not a blocker here: the enforcer-to-predicate wiring is pre-existing and untouched by this change, and the predicate behavior is established above.
tests/test_issue_892_author_bootstrap_deadlock.py:19 — the unittest.mock binding is unused, as is the server-module binding at line 156.
create_issue_bootstrap.py:328 — author bootstrap refusals carry the prefix Create-issue bootstrap guard (#749): even when the task is the author bootstrap. The text is correct and actionable; only the guard named in the label is wrong.
None of these changes production behavior or the correctness of the fix.
Note for the merger
This branch trails master, but neither production module it edits has changed on master since the merge base, and Gitea reports mergeable: true. After the merge, advance the control checkout and restart the MCP fleet so the daemons load this fix before the author bootstrap path is exercised. Issue #931 remains unclaimed and can be re-allocated once that is done.
## Canonical PR State
STATE: reviewed-approved
WHO_IS_NEXT: merger
NEXT_ACTION: Merge PR #926 into master from an independent prgs-merger session, then advance the control checkout and restart the MCP fleet.
NEXT_PROMPT:
```text
Invoke the canonical gitea-workflow skill first. Using only the sanctioned gitea-merger namespace with profile prgs-merger and role merger, merge PR #926 in Scaled-Tech-Consulting/Gitea-Tools at head 2066623986df9a05c0f24bb760d7c0abb5bf0d9e into master. Verify before merging: fleet is exactly five MCP servers under one Claude parent, parity green with mutation_safe true, PR #926 open and mergeable, head unchanged at 2066623986df9a05c0f24bb760d7c0abb5bf0d9e, and the sysadmin APPROVE review is visible at that exact head with approval_at_current_head true. Acquire the merger lease, merge, then confirm issue #892 closed and the branch state. Do not author, review, or approve.
```
WHAT_HAPPENED: Performed an independent review of PR #926 at head 2066623986df9a05c0f24bb760d7c0abb5bf0d9e in the bound review worktree branches/review-pr926-issue-892-bootstrap. Read issue #892, the complete three-file change set, the surrounding guard source, and the new tests. Confirmed the #892 deadlock is present at the base revision and resolved at this head, with every fail-closed protection preserved.
WHY: The scope proof in create_issue_bootstrap.bootstrap_permits_control_checkout accepted only task_scope create_issue_only, while assess_author_issue_bootstrap emitted no task_scope key at all. The predicate therefore could never return True for an author bootstrap, even though the task gate above it already admitted the author task. This change completes that half-applied fix without widening any other proof obligation.
ISSUE: 892
HEAD_SHA: 2066623986df9a05c0f24bb760d7c0abb5bf0d9e
REVIEW_STATUS: APPROVE
MERGE_READY: yes
BLOCKERS: none
VALIDATION: Focused suites at head (test_issue_892_author_bootstrap_deadlock.py, test_create_issue_bootstrap.py, test_issue_757_bootstrap_guard_agreement.py, test_author_issue_bootstrap.py) = 105 passed, 31 subtests. Broad regression over 33 bootstrap/worktree/guard/preflight suites, identical set at both revisions: base 2b4e430 = 10 failed / 599 passed / 96 subtests; head 2066623 = 10 failed / 613 passed / 96 subtests. Delta is +14 passed, equal to the 14 new tests; zero new failures. The 10 failures share identical test IDs at both revisions, are pre-existing, and each passes when run alone.
LAST_UPDATED_BY: sysadmin / prgs-reviewer
NATIVE_REVIEW_PROOF: transport=native_mcp; entrypoint=mcp_server; token_fingerprint=cb971ff277593f1b
## Independence
Reviewer identity `sysadmin` on profile `prgs-reviewer`; PR author is `jcwalker3`. Distinct accounts, no self-review. Base branch `master` at `6a56260768ad2f69b773a8b7ab8d460ce821c914`; merge base `2b4e43042a34f4e29617378ae79a7f5a3d312688`.
## Scope
Exactly three files, merge base to head:
- `author_issue_bootstrap.py` (+157 / −46)
- `create_issue_bootstrap.py` (+18 / −6)
- `tests/test_issue_892_author_bootstrap_deadlock.py` (+215, new)
Both production modules are byte-identical between the merge base and the current `master` tip, so this analysis holds against `master` and the merge is clean on the touched files. Nothing in the change touches role permissions, credential resolution, or process launching, and it adds no manual worktree or branch-creation bypass.
## Deadlock present at base, resolved at head
A single scenario matrix was evaluated read-only in the review worktree against both revisions.
| # | Scenario | base `2b4e430` | head `2066623` |
|---|---|---|---|
| A1 | clean control checkout, author bootstrap | `False` | **`True`** |
| A2 | create_issue bootstrap | `True` | `True` |
| D1 | dirty control checkout | `False` | `False` |
| D2 | non-base branch | `False` | `False` |
| D3 | detached HEAD | `False` | `False` |
| D4 | HEAD does not match live master tip | `False` | `False` |
| D5 | remote tip unresolvable | `False` | `False` |
| D6 | remote tip missing, no error | `False` | `False` |
| D7 | local HEAD sha unknown | `False` | `False` |
| D8 | workspace is not the canonical root | `False` | `False` |
| D9 | under `branches/` worktree | `False` | `False` |
| D10 | unrelated task literal | `False` | `False` |
| D11 | ordinary author mutation (`commit_files`) | `False` | `False` |
| S1 | author scope presented for a create_issue task | `False` | `False` |
| S2 | create_issue_only scope presented for an author task | `False` | `False` |
| S3 | hand-built `allowed=True` over a dirty tree | `False` | `False` |
| S4 | hand-built `base_tips_verified` over mismatched tips | `False` | `False` |
Exactly one behavior delta, A1. Every denial and every cross-scope or hand-built assertion is refused identically at both revisions.
## Fix assessment
`bootstrap_permits_control_checkout` now applies the scope proof per task branch rather than as a widened disjunction: a create_issue task still requires `create_issue_only`, an author bootstrap task requires `author_issue_bootstrap`, and a final `else` closes unknown tasks. Both directions of cross-scope reuse fail closed (S1, S2). Every other obligation is unchanged and still applies to both branches — `bootstrap_path` must be `clean_canonical_control_checkout`, `reasons` must be empty, `dirty_files` must be empty, `under_branches` must be exactly `False`, tip equality is re-derived rather than trusted from the assessment's own flag, and the assessment must bind to the exact workspace and canonical root.
Ordinary author mutations stay blocked from the control checkout: the task gate admits only `bootstrap_author_issue_worktree` and `gitea_bootstrap_author_issue_worktree`, so every other author task leaves the ordinary block in force (D11).
`assess_author_issue_bootstrap` now produces the full predicate-compatible field set through one constructor. `proven` and `base_tips_verified` are derived inside that constructor rather than accepted from callers, and `task_scope` is fixed to `author_issue_bootstrap`, so it cannot mint a create_issue-scoped assessment.
The allowed path is strictly tighter than before. The prior logic compared tips only when both were present, so a missing local HEAD or a missing remote tip left `reasons` empty and yielded `allowed=True` with no base-equivalence proof. The new logic fails closed on an unknown local HEAD, on a resolver error, and on a missing remote tip (D5, D6, D7), closing a pre-existing fail-open.
The `under_branches` result changed from `allowed=True` to `not_applicable=True`. This is operationally inert: when the workspace is a valid `branches/` worktree the ordinary guard does not block and returns before the bootstrap assessment is consulted, and on the durable-block path the predicate refused it at both revisions, since the prior dict carried no `under_branches` key and the `is not False` check rejected the resulting `None`. The new shape simply states the intent correctly.
Two-guard agreement from #757 is preserved: `anti_stomp_preflight.py:599` routes the same threaded assessment through the same predicate under the same task name, so the #274 and #604 guards cannot diverge. The waiver stays narrow — only the wrong-worktree verdict is waived, while root-checkout, repo, role, stale-runtime and lease checks are each still evaluated.
The #892 criterion that a refusal must name the missing precondition and the sanctioned call is met: `format_create_issue_bootstrap_error` at `create_issue_bootstrap.py:322-330` surfaces `exact_next_action`, and author block assessments populate it through `EXACT_NEXT_ACTION_AUTHOR_BOOTSTRAP`.
## Non-blocking findings
1. `tests/test_issue_892_author_bootstrap_deadlock.py:151-211` — `TestAuthorBootstrapPreflightIntegration.test_enforce_branches_only_allows_clean_control_for_bootstrap` is described as exercising the real enforcer wiring and binds the server module as `srv` at line 156, but never calls `srv._enforce_branches_only_author_mutation`. The `durable_block` dict built at line 192 is passed to nothing, and the assertion at line 211 holds by construction. The branch at lines 200-208 reduces to the assertion already made at line 113. The net effect is a duplicate unit assertion under an integration name, and no test drives `_enforce_branches_only_author_mutation` with an author bootstrap task — `test_issue_757_bootstrap_guard_agreement.py:349` reaches the real enforcer but only for create_issue tasks. A follow-up should cover that path end to end. Not a blocker here: the enforcer-to-predicate wiring is pre-existing and untouched by this change, and the predicate behavior is established above.
2. `tests/test_issue_892_author_bootstrap_deadlock.py:19` — the `unittest.mock` binding is unused, as is the server-module binding at line 156.
3. `create_issue_bootstrap.py:328` — author bootstrap refusals carry the prefix `Create-issue bootstrap guard (#749):` even when the task is the author bootstrap. The text is correct and actionable; only the guard named in the label is wrong.
None of these changes production behavior or the correctness of the fix.
## Note for the merger
This branch trails `master`, but neither production module it edits has changed on `master` since the merge base, and Gitea reports `mergeable: true`. After the merge, advance the control checkout and restart the MCP fleet so the daemons load this fix before the author bootstrap path is exercised. Issue #931 remains unclaimed and can be re-allocated once that is done.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
sysadmin
Summary
Breaks the author worktree bootstrap deadlock (#892) so an author session starting from a clean control checkout can obtain a usable
branches/worktree through the sanctioned MCP path.Root cause
assess_author_issue_bootstrapreturnedallowed/provenfor a clean control checkout, but the shared predicatebootstrap_permits_control_checkoutonly accepted create_issue assessments (task_scope=create_issue_only, emptyreasons, full base-tip field set). Author assessments never satisfied the predicate, so#274/#604guards kept the ordinary control-checkout block on everygitea_bootstrap_author_issue_worktreecall.Fix
author_issue_bootstrap.assess_author_issue_bootstrap— emit create_issue-compatible proof fields (task_scope=author_issue_bootstrap, emptyreasonswhen allowed,base_tips_verified, binding paths, strict tip proof).create_issue_bootstrap.bootstrap_permits_control_checkout— accepttask_scope=author_issue_bootstrapfor author bootstrap tasks (still fail-closed for cross-scope smuggling).tests/test_issue_892_author_bootstrap_deadlock.py.Files
author_issue_bootstrap.pycreate_issue_bootstrap.pytests/test_issue_892_author_bootstrap_deadlock.pyValidation
Worktree / proof
branches/fix-issue-892-author-bootstrap-deadlockfix/issue-892-author-bootstrap-deadlock2066623986df9a05c0f24bb760d7c0abb5bf0d9ejcwalker3/prgs-authorgitea_bootstrap_author_issue_worktreeis the primary path.Risk
Low — narrow waiver remains fail-closed (clean base, tip proof, binding match, empty refusal reasons). Does not widen create_issue scope.
Closes #892
Handoff
WHO_IS_NEXT: reviewer — independent review of PR; do not self-merge.
repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-reviewer
session_id: 45657-6b1b84aba2e2
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: claimed
candidate_head:
2066623986target_branch: master
target_branch_sha:
2b4e43042alast_activity: 2026-07-25T23:32:03Z
expires_at: 2026-07-25T23:42:03Z
blocker: none
repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-reviewer
session_id: 52497-a4b43ab643a5
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: claimed
candidate_head:
2066623986target_branch: master
target_branch_sha:
6a56260768last_activity: 2026-07-26T08:05:18Z
expires_at: 2026-07-26T08:15:18Z
blocker: none
repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-reviewer
session_id: 52497-df4ee80a7322
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: claimed
candidate_head:
2066623986target_branch: master
target_branch_sha:
6a56260768last_activity: 2026-07-26T08:16:07Z
expires_at: 2026-07-26T08:26:07Z
blocker: none
Canonical PR State
STATE: reviewed-approved
WHO_IS_NEXT: merger
NEXT_ACTION: Merge PR #926 into master from an independent prgs-merger session, then advance the control checkout and restart the MCP fleet.
NEXT_PROMPT:
WHAT_HAPPENED: Performed an independent review of PR #926 at head
2066623986in the bound review worktree branches/review-pr926-issue-892-bootstrap. Read issue #892, the complete three-file change set, the surrounding guard source, and the new tests. Confirmed the #892 deadlock is present at the base revision and resolved at this head, with every fail-closed protection preserved.WHY: The scope proof in create_issue_bootstrap.bootstrap_permits_control_checkout accepted only task_scope create_issue_only, while assess_author_issue_bootstrap emitted no task_scope key at all. The predicate therefore could never return True for an author bootstrap, even though the task gate above it already admitted the author task. This change completes that half-applied fix without widening any other proof obligation.
ISSUE: 892
HEAD_SHA:
2066623986REVIEW_STATUS: APPROVE
MERGE_READY: yes
BLOCKERS: none
VALIDATION: Focused suites at head (test_issue_892_author_bootstrap_deadlock.py, test_create_issue_bootstrap.py, test_issue_757_bootstrap_guard_agreement.py, test_author_issue_bootstrap.py) = 105 passed, 31 subtests. Broad regression over 33 bootstrap/worktree/guard/preflight suites, identical set at both revisions: base
2b4e430= 10 failed / 599 passed / 96 subtests; head2066623= 10 failed / 613 passed / 96 subtests. Delta is +14 passed, equal to the 14 new tests; zero new failures. The 10 failures share identical test IDs at both revisions, are pre-existing, and each passes when run alone.LAST_UPDATED_BY: sysadmin / prgs-reviewer
NATIVE_REVIEW_PROOF: transport=native_mcp; entrypoint=mcp_server; token_fingerprint=cb971ff277593f1b
Independence
Reviewer identity
sysadminon profileprgs-reviewer; PR author isjcwalker3. Distinct accounts, no self-review. Base branchmasterat6a56260768ad2f69b773a8b7ab8d460ce821c914; merge base2b4e43042a34f4e29617378ae79a7f5a3d312688.Scope
Exactly three files, merge base to head:
author_issue_bootstrap.py(+157 / −46)create_issue_bootstrap.py(+18 / −6)tests/test_issue_892_author_bootstrap_deadlock.py(+215, new)Both production modules are byte-identical between the merge base and the current
mastertip, so this analysis holds againstmasterand the merge is clean on the touched files. Nothing in the change touches role permissions, credential resolution, or process launching, and it adds no manual worktree or branch-creation bypass.Deadlock present at base, resolved at head
A single scenario matrix was evaluated read-only in the review worktree against both revisions.
2b4e4302066623FalseTrueTrueTrueFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalseFalsebranches/worktreeFalseFalseFalseFalsecommit_files)FalseFalseFalseFalseFalseFalseallowed=Trueover a dirty treeFalseFalsebase_tips_verifiedover mismatched tipsFalseFalseExactly one behavior delta, A1. Every denial and every cross-scope or hand-built assertion is refused identically at both revisions.
Fix assessment
bootstrap_permits_control_checkoutnow applies the scope proof per task branch rather than as a widened disjunction: a create_issue task still requirescreate_issue_only, an author bootstrap task requiresauthor_issue_bootstrap, and a finalelsecloses unknown tasks. Both directions of cross-scope reuse fail closed (S1, S2). Every other obligation is unchanged and still applies to both branches —bootstrap_pathmust beclean_canonical_control_checkout,reasonsmust be empty,dirty_filesmust be empty,under_branchesmust be exactlyFalse, tip equality is re-derived rather than trusted from the assessment's own flag, and the assessment must bind to the exact workspace and canonical root.Ordinary author mutations stay blocked from the control checkout: the task gate admits only
bootstrap_author_issue_worktreeandgitea_bootstrap_author_issue_worktree, so every other author task leaves the ordinary block in force (D11).assess_author_issue_bootstrapnow produces the full predicate-compatible field set through one constructor.provenandbase_tips_verifiedare derived inside that constructor rather than accepted from callers, andtask_scopeis fixed toauthor_issue_bootstrap, so it cannot mint a create_issue-scoped assessment.The allowed path is strictly tighter than before. The prior logic compared tips only when both were present, so a missing local HEAD or a missing remote tip left
reasonsempty and yieldedallowed=Truewith no base-equivalence proof. The new logic fails closed on an unknown local HEAD, on a resolver error, and on a missing remote tip (D5, D6, D7), closing a pre-existing fail-open.The
under_branchesresult changed fromallowed=Truetonot_applicable=True. This is operationally inert: when the workspace is a validbranches/worktree the ordinary guard does not block and returns before the bootstrap assessment is consulted, and on the durable-block path the predicate refused it at both revisions, since the prior dict carried nounder_brancheskey and theis not Falsecheck rejected the resultingNone. The new shape simply states the intent correctly.Two-guard agreement from #757 is preserved:
anti_stomp_preflight.py:599routes the same threaded assessment through the same predicate under the same task name, so the #274 and #604 guards cannot diverge. The waiver stays narrow — only the wrong-worktree verdict is waived, while root-checkout, repo, role, stale-runtime and lease checks are each still evaluated.The #892 criterion that a refusal must name the missing precondition and the sanctioned call is met:
format_create_issue_bootstrap_erroratcreate_issue_bootstrap.py:322-330surfacesexact_next_action, and author block assessments populate it throughEXACT_NEXT_ACTION_AUTHOR_BOOTSTRAP.Non-blocking findings
tests/test_issue_892_author_bootstrap_deadlock.py:151-211—TestAuthorBootstrapPreflightIntegration.test_enforce_branches_only_allows_clean_control_for_bootstrapis described as exercising the real enforcer wiring and binds the server module assrvat line 156, but never callssrv._enforce_branches_only_author_mutation. Thedurable_blockdict built at line 192 is passed to nothing, and the assertion at line 211 holds by construction. The branch at lines 200-208 reduces to the assertion already made at line 113. The net effect is a duplicate unit assertion under an integration name, and no test drives_enforce_branches_only_author_mutationwith an author bootstrap task —test_issue_757_bootstrap_guard_agreement.py:349reaches the real enforcer but only for create_issue tasks. A follow-up should cover that path end to end. Not a blocker here: the enforcer-to-predicate wiring is pre-existing and untouched by this change, and the predicate behavior is established above.tests/test_issue_892_author_bootstrap_deadlock.py:19— theunittest.mockbinding is unused, as is the server-module binding at line 156.create_issue_bootstrap.py:328— author bootstrap refusals carry the prefixCreate-issue bootstrap guard (#749):even when the task is the author bootstrap. The text is correct and actionable; only the guard named in the label is wrong.None of these changes production behavior or the correctness of the fix.
Note for the merger
This branch trails
master, but neither production module it edits has changed onmastersince the merge base, and Gitea reportsmergeable: true. After the merge, advance the control checkout and restart the MCP fleet so the daemons load this fix before the author bootstrap path is exercised. Issue #931 remains unclaimed and can be re-allocated once that is done.repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-reviewer
session_id: 52497-df4ee80a7322
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: released
candidate_head:
2066623986target_branch: master
target_branch_sha:
6a56260768last_activity: 2026-07-26T08:19:56Z
expires_at: 2026-07-26T08:29:56Z
blocker: manual-release
repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-merger
session_id: 52586-294892e2f1d7
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: claimed
candidate_head:
2066623986target_branch: master
target_branch_sha:
6a56260768last_activity: 2026-07-26T08:24:54Z
expires_at: 2026-07-26T08:34:54Z
blocker: none
Stale #332 review-decision lock cleanup (#594)
Status: APPLIED
sysadminprgs-merger2026-07-26T08:25:36.220396+00:00approveon PR fix(bootstrap): allow author worktree bootstrap from clean control checkout (Closes #892) (#926)closed(merged=True)8c1d22a658ecb5715e3f4db6eb7ebb408830c1011prgs-reviewerManual deletion of session-state files is not the workflow.
This path only clears a lock when the referenced PR is merged/closed.
finalized_at: 2026-07-26T08:26:31Z
finalized_by_identity: sysadmin
finalized_by_profile: prgs-merger
finalized_by_session_id: 52586-294892e2f1d7
finalization_outcome: released
finalization_reason: PR #926 merged successfully at 2066623986df9a05c0f24bb760d7c0abb5bf0d9e; merge commit
8c1d22a658finalized_lease_comment_id: 17362
repo: Scaled-Tech-Consulting/Gitea-Tools
pr: #926
issue: #892
reviewer_identity: sysadmin
profile: prgs-merger
session_id: 52586-294892e2f1d7
worktree: /Users/jasonwalker/Development/Gitea-Tools/branches/review-pr926-issue-892-bootstrap
phase: released
candidate_head:
2066623986target_branch: master
target_branch_sha:
6a56260768last_activity: 2026-07-26T08:26:31Z
expires_at: 2026-07-26T08:36:31Z
blocker: PR #926 merged successfully at 2066623986df9a05c0f24bb760d7c0abb5bf0d9e; merge commit
8c1d22a658