P1 process: add duplicate-write idempotency guard for automation

[marcohost33-maker]

Problem

A repeated connector write during the 2026-07-14 knowledge-transfer pass created duplicate T2-governance issues #25 and #26. #26 was immediately closed as duplicate, but the event exposes a general idempotency gap for automated issue creation.

Required control

Before creating a new issue, automation should:

1. derive a deterministic idempotency key from repository, normalized title, scope/ref ID and operation type;
2. search open and recently closed issues for that key or exact normalized title;
3. reuse/comment on the existing issue when the intended scope matches;
4. create only when no equivalent record exists;
5. record the resulting issue number before any retry;
6. make retries read-after-write rather than blind replays.

Suggested metadata

Embed a hidden or visible marker such as:

IDEMPOTENCY_KEY: browser|T2-admission-policy|2026-07-14|create-issue

Metric

DuplicateWriteRate = duplicate side effects / attempted write operations

Hard target for deterministic automation: zero. Any duplicate must be closed or reconciled and retained as a negative result.

Acceptance

• issue/PR creation helper performs preflight search
• retries use read-after-write confirmation
• deterministic idempotency marker documented
• regression test simulates timeout after successful remote write
• negative-result entry retained

This issue is process hardening and does not block the browser runtime architecture decision.

[marcohost33-maker]

Triage 2026-07-17 — verdict: STILL-VALID · independent process hardening

The duplicate-write event (#25/#26, #26 closed as dup on 2026-07-14) is a real negative result; the idempotency-key + preflight-search + read-after-write-retry control is well-scoped and independent of the runtime architecture. Explicitly does not block the runtime decision.

Next action: implement the preflight/idempotency guard in the issue/PR creation helper + a regression test simulating "timeout after successful remote write"; retain the #26 duplicate as the negative-result entry. Off the critical path; low urgency during the CI/budget freeze (no CI to run the regression test now). Priority: P2. Kept open.