Source: ce:review round 2 on feat/private-visibility-transition (adversarial, P3 conf 0.64)
Scope: scripts/reconcile-repos.ts — visibilityTransitionSucceeded counter in runIssueQueue
Problem
When issues.create returns an error but the alert was actually created on GitHub's side (e.g., network timeout AFTER the API processed the request, ambiguous 5xx, retry-induced double-write), the local counter increments failed even though the issue exists. Conversely, a successful API response could in theory be lost in transit.
Suggested fix (one of):
A: After an ambiguous create failure, check for the exact expected title via issues.listForRepo before incrementing failed. Reconciles the counter with reality at the cost of an extra API call per failure.
B: Document the counter semantics in JSDoc: "counts API acknowledgements, not confirmed-created alerts". Accept the imprecision.
Option B is fine for current scale (visibility transitions are rare). Option A becomes worthwhile if the counter is ever used for SLO/alerting.
Acceptance
- Counter semantics are unambiguous (either accurate or clearly documented).
Why not in the round-3 fix
P3 advisory. Round-3 was scoped to the new P1 + test gaps + shell-injection defense.
Source: ce:review round 2 on feat/private-visibility-transition (adversarial, P3 conf 0.64)
Scope:
scripts/reconcile-repos.ts—visibilityTransitionSucceededcounter inrunIssueQueueProblem
When
issues.createreturns an error but the alert was actually created on GitHub's side (e.g., network timeout AFTER the API processed the request, ambiguous 5xx, retry-induced double-write), the local counter incrementsfailedeven though the issue exists. Conversely, a successful API response could in theory be lost in transit.Suggested fix (one of):
A: After an ambiguous create failure, check for the exact expected title via
issues.listForRepobefore incrementingfailed. Reconciles the counter with reality at the cost of an extra API call per failure.B: Document the counter semantics in JSDoc: "counts API acknowledgements, not confirmed-created alerts". Accept the imprecision.
Option B is fine for current scale (visibility transitions are rare). Option A becomes worthwhile if the counter is ever used for SLO/alerting.
Acceptance
Why not in the round-3 fix
P3 advisory. Round-3 was scoped to the new P1 + test gaps + shell-injection defense.