Add hard queue-target resolution wall before PR inventory or empty-queue claims #200

New Issue

2026-07-05T15:13:27-05:00

jcwalker3 commented

2026-07-05 15:13:27 -05:00

Problem

Reviewer agents can check the wrong repository queue and then report "no open PRs," even when the intended review backlog has open PRs. This creates a false stop condition and hides actionable PRs from the operator.

Observed failure

The agent repeatedly reported that Scaled-Tech-Consulting/mcp-control-plane had zero open PRs and concluded that there was nothing to review. However, the active review backlog is Scaled-Tech-Consulting/Gitea-Tools, which currently has six open PRs:

#195 — Add fail-closed PR inventory wall when list_prs returns an unexpected empty queue
#193 — Raise reviewer A-bar: capability, sweep, live-state, and role-boundary proofs
#192 — Add role-boundary proofs for reviewer queue workflows
#190 — Fix MCP registration and naming for Jenkins/GlitchTip servers and add discoverability coverage
#187 — Harden author-run reporting and mutation capability proof
#181 — Add fail-closed branch-identity proofs for author commit/push workflow

Required wall

Before any PR inventory, review selection, or empty-queue report, the workflow must resolve and lock the intended queue target.

The queue target must be derived from explicit operator context, including:

repo name mentioned by the user;
PR numbers/titles supplied by the user;
current working project context;
previous handoff target;
remote/owner/repo configured for the active task.

If these sources conflict, the agent must fail closed with a target-resolution blocker. It must not silently inventory a different repo.

Required behavior

If the operator provides PR numbers or PR titles, the agent must verify those PRs exist in the resolved repo before claiming the queue is empty.
If the agent inventories mcp-control-plane but the supplied PRs are in Gitea-Tools, the result must be target_repo_mismatch, not trusted_empty.
Empty-queue claims are forbidden unless the queue target has first been resolved and locked.
The final report must include the locked queue target: remote, owner, repo, source of target resolution, and whether user-provided PRs were reconciled.

Acceptance criteria

Add a test where the user supplies open Gitea-Tools PRs but the agent inventories mcp-control-plane; expected result: target_repo_mismatch.
Add a test where list_prs returns empty for the wrong repo; expected result: blocked, not "no open PRs."
Add a test where the user says "currently we have 6 open PRs" and provides PR numbers; expected result: the workflow must reconcile those PR numbers before stopping.
Add a test where repo context is ambiguous between mcp-control-plane and Gitea-Tools; expected result: fail closed and require target clarification or explicit operator target.
Empty-queue reports must include queue_target_lock.status == resolved.
pr_inventory_trust_gate must consume the queue target lock result; it cannot return trusted_empty if the repo target is unresolved or contradicted by operator context.

Suggested implementation

Add a queue_target_lock or resolve_pr_queue_target helper that runs before pr_inventory_trust_gate. The trust gate should refuse to evaluate empty inventory unless the target lock is resolved and consistent with operator-provided PR context.

#194 (false-empty investigation), PR #195 (pr_inventory_trust_gate helper), #196 (gate wired into live path), #198 (gate status required in reports). This issue adds the upstream target-resolution lock the gate must consume.
Context note for implementers: in the origin runs the operator prompt explicitly stated Repository: Scaled-Tech-Consulting/mcp-control-plane while the actionable backlog lived in Gitea-Tools — exactly the conflicting-sources case this wall must fail closed on (explicit directive vs. supplied PR context), rather than silently privileging either source.

### Problem Reviewer agents can check the wrong repository queue and then report "no open PRs," even when the intended review backlog has open PRs. This creates a false stop condition and hides actionable PRs from the operator. ### Observed failure The agent repeatedly reported that `Scaled-Tech-Consulting/mcp-control-plane` had zero open PRs and concluded that there was nothing to review. However, the active review backlog is `Scaled-Tech-Consulting/Gitea-Tools`, which currently has six open PRs: * #195 — Add fail-closed PR inventory wall when list_prs returns an unexpected empty queue * #193 — Raise reviewer A-bar: capability, sweep, live-state, and role-boundary proofs * #192 — Add role-boundary proofs for reviewer queue workflows * #190 — Fix MCP registration and naming for Jenkins/GlitchTip servers and add discoverability coverage * #187 — Harden author-run reporting and mutation capability proof * #181 — Add fail-closed branch-identity proofs for author commit/push workflow ### Required wall Before any PR inventory, review selection, or empty-queue report, the workflow must resolve and lock the intended queue target. The queue target must be derived from explicit operator context, including: * repo name mentioned by the user; * PR numbers/titles supplied by the user; * current working project context; * previous handoff target; * remote/owner/repo configured for the active task. If these sources conflict, the agent must fail closed with a target-resolution blocker. It must not silently inventory a different repo. ### Required behavior * If the operator provides PR numbers or PR titles, the agent must verify those PRs exist in the resolved repo before claiming the queue is empty. * If the agent inventories `mcp-control-plane` but the supplied PRs are in `Gitea-Tools`, the result must be `target_repo_mismatch`, not `trusted_empty`. * Empty-queue claims are forbidden unless the queue target has first been resolved and locked. * The final report must include the locked queue target: remote, owner, repo, source of target resolution, and whether user-provided PRs were reconciled. ### Acceptance criteria * Add a test where the user supplies open Gitea-Tools PRs but the agent inventories mcp-control-plane; expected result: `target_repo_mismatch`. * Add a test where `list_prs` returns empty for the wrong repo; expected result: blocked, not "no open PRs." * Add a test where the user says "currently we have 6 open PRs" and provides PR numbers; expected result: the workflow must reconcile those PR numbers before stopping. * Add a test where repo context is ambiguous between `mcp-control-plane` and `Gitea-Tools`; expected result: fail closed and require target clarification or explicit operator target. * Empty-queue reports must include `queue_target_lock.status == resolved`. * `pr_inventory_trust_gate` must consume the queue target lock result; it cannot return `trusted_empty` if the repo target is unresolved or contradicted by operator context. ### Suggested implementation Add a `queue_target_lock` or `resolve_pr_queue_target` helper that runs before `pr_inventory_trust_gate`. The trust gate should refuse to evaluate empty inventory unless the target lock is resolved and consistent with operator-provided PR context. ### Related * #194 (false-empty investigation), PR #195 (`pr_inventory_trust_gate` helper), #196 (gate wired into live path), #198 (gate status required in reports). This issue adds the upstream target-resolution lock the gate must consume. * Context note for implementers: in the origin runs the operator prompt explicitly stated `Repository: Scaled-Tech-Consulting/mcp-control-plane` while the actionable backlog lived in `Gitea-Tools` — exactly the conflicting-sources case this wall must fail closed on (explicit directive vs. supplied PR context), rather than silently privileging either source.

jcwalker3 referenced this issue

2026-07-05 15:24:46 -05:00

Enforce exact-title and semantic duplicate stop before issue creation #207

jcwalker3 referenced this issue

2026-07-05 15:25:38 -05:00

Add hard queue-target resolution wall before PR inventory or empty-queue claims #202

jcwalker3 commented

2026-07-05 15:26:09 -05:00

Consolidation note: this issue is the canonical copy of an exact-title triple — #201 and #202 carry the identical title and body intent, filed by parallel author sessions within minutes (TOCTOU duplicate-search failure, now walled by #207). #202 has been closed as a duplicate by its authoring session; #201 still needs operator/triage consolidation into this issue.

Preserved deltas from the #202 copy for whoever implements this:

The queue_target_lock / resolve_pr_queue_target helper should build on review_proofs.resolve_repos_from_user_reference (added for #184 via PR #185), which already maps user repo references to canonical repos — the lock adds operator-PR reconciliation, conflict fail-close, and the lock-status output.
A target_repo_mismatch blocker report must also follow the #197 terminal-report purity rules.

Consolidation note: this issue is the canonical copy of an exact-title triple — #201 and #202 carry the identical title and body intent, filed by parallel author sessions within minutes (TOCTOU duplicate-search failure, now walled by #207). #202 has been closed as a duplicate by its authoring session; #201 still needs operator/triage consolidation into this issue. Preserved deltas from the #202 copy for whoever implements this: * The `queue_target_lock` / `resolve_pr_queue_target` helper should build on `review_proofs.resolve_repos_from_user_reference` (added for #184 via PR #185), which already maps user repo references to canonical repos — the lock adds operator-PR reconciliation, conflict fail-close, and the lock-status output. * A `target_repo_mismatch` blocker report must also follow the #197 terminal-report purity rules.

jcwalker3 commented

2026-07-05 15:33:39 -05:00

Additional observed failure (reviewer-side, new evidence):

A reviewer run authenticated correctly as sysadmin / prgs-reviewer and proved pagination for Scaled-Tech-Consulting/mcp-control-plane, returning 0 open PRs. However, the operator context stated that there were 4 open PRs in the active review backlog.

The agent treated the empty mcp-control-plane result as a clean stop instead of reconciling the contradiction. This is a target-resolution failure — and it shows the failure mode survives even with correct identity, correct capability, and a proven-complete single-repo inventory.

Correct behavior should be:

detect that operator context says open PRs exist;
verify which repo owns those PRs;
compare that repo to the resolved target repo;
return target_repo_mismatch if they differ;
refuse to report "no open PRs" or "nothing to review."

Important distinction:

The REST pagination proof and X-Total-Count: 0 only prove that the queried repo has zero open PRs. They do not prove that the intended queue is empty. Inventory-completeness proof (per-repo) and queue-target correctness are independent axes; pr_inventory_trust_gate covers the first, this wall covers the second, and trusted_empty requires both.

Required acceptance test (add to this issue's list):

Given operator context: "there are 4 open PRs"
And the agent inventories Scaled-Tech-Consulting/mcp-control-plane
And that repo returns 0 open PRs
But the active backlog is Scaled-Tech-Consulting/Gitea-Tools
Then the workflow must stop with target_repo_mismatch, not trusted_empty.

Additional observed failure (reviewer-side, new evidence): A reviewer run authenticated correctly as `sysadmin / prgs-reviewer` and proved pagination for `Scaled-Tech-Consulting/mcp-control-plane`, returning `0` open PRs. However, the operator context stated that there were 4 open PRs in the active review backlog. The agent treated the empty `mcp-control-plane` result as a clean stop instead of reconciling the contradiction. This is a target-resolution failure — and it shows the failure mode survives even with correct identity, correct capability, and a proven-complete single-repo inventory. Correct behavior should be: * detect that operator context says open PRs exist; * verify which repo owns those PRs; * compare that repo to the resolved target repo; * return `target_repo_mismatch` if they differ; * refuse to report "no open PRs" or "nothing to review." Important distinction: The REST pagination proof and `X-Total-Count: 0` only prove that the queried repo has zero open PRs. They do not prove that the intended queue is empty. Inventory-completeness proof (per-repo) and queue-target correctness are independent axes; `pr_inventory_trust_gate` covers the first, this wall covers the second, and `trusted_empty` requires both. Required acceptance test (add to this issue's list): Given operator context: "there are 4 open PRs" And the agent inventories `Scaled-Tech-Consulting/mcp-control-plane` And that repo returns `0` open PRs But the active backlog is `Scaled-Tech-Consulting/Gitea-Tools` Then the workflow must stop with `target_repo_mismatch`, not `trusted_empty`.

jcwalker3 referenced this issue

2026-07-05 15:35:19 -05:00

Add hard queue-target resolution wall before PR inventory or empty-queue claims #201

sysadmin commented

2026-07-05 16:10:07 -05:00

Additional observed failure:

A reviewer run again resolved and inventoried Scaled-Tech-Consulting/mcp-control-plane, proved that repo had zero open PRs, and stopped with trusted_empty.

This did not reconcile against the active operator context, where the review backlog is in Scaled-Tech-Consulting/Gitea-Tools.

The run proves only that mcp-control-plane has no open PRs. It does not prove the intended queue is empty.

Required behavior:

Before inventory, resolve the queue target from operator context.
If the target repo conflicts with the active backlog context, stop with target_repo_mismatch.
pr_inventory_trust_gate must not return trusted_empty if queue target resolution is unresolved or contradicted.

Additional observed failure: A reviewer run again resolved and inventoried `Scaled-Tech-Consulting/mcp-control-plane`, proved that repo had zero open PRs, and stopped with `trusted_empty`. This did not reconcile against the active operator context, where the review backlog is in `Scaled-Tech-Consulting/Gitea-Tools`. The run proves only that `mcp-control-plane` has no open PRs. It does not prove the intended queue is empty. Required behavior: * Before inventory, resolve the queue target from operator context. * If the target repo conflicts with the active backlog context, stop with `target_repo_mismatch`. * `pr_inventory_trust_gate` must not return `trusted_empty` if queue target resolution is unresolved or contradicted.

jcwalker3 added the status:in-progress label 2026-07-06 10:28:39 -05:00

jcwalker3 referenced this issue from a commit

2026-07-06 10:30:07 -05:00

feat(reviewer): add queue-target lock before PR inventory trust gate (#200)

jcwalker3 referenced a pull request that will close this issue

2026-07-06 10:30:14 -05:00

feat(reviewer): add queue-target lock before PR inventory trust gate (closes #200) #240

jcwalker3 commented

2026-07-06 10:30:15 -05:00

Claimed and implemented on feat/issue-200-queue-target-lock.

Approach: reconcile_queue_target locks the intended queue repo from operator context / supplied PR backlog before list_prs. pr_inventory_trust_gate returns target_repo_mismatch (not trusted_empty) when the inventoried repo does not own the operator-supplied PRs. Wired into assess_reviewer_queue_inventory.

Tests: 7 new unit tests covering mismatch, ambiguity, conflicting sources, PR-number reconciliation, final-report fields, and canonical inventory integration. All 11 queue-target + inventory tests pass.

PR opened for review.

Claimed and implemented on `feat/issue-200-queue-target-lock`. **Approach:** `reconcile_queue_target` locks the intended queue repo from operator context / supplied PR backlog before `list_prs`. `pr_inventory_trust_gate` returns `target_repo_mismatch` (not `trusted_empty`) when the inventoried repo does not own the operator-supplied PRs. Wired into `assess_reviewer_queue_inventory`. **Tests:** 7 new unit tests covering mismatch, ambiguity, conflicting sources, PR-number reconciliation, final-report fields, and canonical inventory integration. All 11 queue-target + inventory tests pass. PR opened for review.

sysadmin referenced this issue

2026-07-06 11:21:02 -05:00

feat(reviewer): add queue-target lock before PR inventory trust gate (closes #200) #240

jcwalker3 referenced this issue from a commit

2026-07-06 11:29:38 -05:00

feat(reviewer): add queue-target lock before PR inventory trust gate (#200)

sysadmin referenced a pull request that will close this issue

2026-07-06 11:29:50 -05:00

feat(reviewer): add queue-target lock before PR inventory trust gate (closes #200) #240

sysadmin closed this issue

2026-07-06 11:29:56 -05:00

sysadmin referenced this issue from a commit