fix: gate AI slop advisory to confirmed contributors

[JSONbored]

Motivation

• The AI slop advisory could be invoked for any PR author when slopGateMode != off and slopAiAdvisory was enabled, allowing untrusted PRs to consume shared Workers AI budget.
• The budget estimate undercounted worst-case model usage because the runner may perform retries and a fallback model, amplifying actual calls versus the recorded estimate.

Description

• Require a confirmedContributor check in runAiSlopForAdvisory and no-op when it is false or the advisory has no headSha so unconfirmed PR authors cannot trigger the AI call (src/queue/processors.ts).
• Move contributor confirmation earlier in the PR processing flow and pass confirmedContributor through to the slop advisory helper so the gating is enforced before any AI invocation (src/queue/processors.ts).
• Increase the neuron-budget estimate to account for the worst-case retry/fallback call count by adding SLOP_AI_MAX_MODEL_CALLS = 6 and using it in estimateNeurons (src/services/ai-slop.ts).
• Add unit coverage that verifies unconfirmed contributors do not invoke the AI runner and update existing tests to pass the new confirmedContributor parameter (test/unit/ai-slop.test.ts).

Testing

• Ran type checking with npm run typecheck and it completed successfully.
• Ran the AI slop unit tests with npx vitest run test/unit/ai-slop.test.ts and the suite passed (28 tests, all passed).

---

Codex Task

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#727 is no longer open.	No action.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[ghost]

gittensory · advisory review

Reviewed 3 changed file(s) — two independent AI reviewers.

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

Reviewer A · gpt-oss-120b — recommends ✅ merge
The PR adds a confirmation check so AI slop advisory only runs for verified contributors, introduces a constant to bound model calls, and updates tests accordingly. The changes are well‑scoped, compile‑time safe, and include appropriate test coverage.

Suggestions

• Consider documenting why the official contributor detection is only performed when gateEnabled is true, to avoid future confusion.
• If runAiSlopForAdvisory could be called from other places in the future, make confirmedContributor an optional parameter with a default of false to keep the API flexible.

Worth double-checking

• If there are other call sites for runAiSlopForAdvisory that were not updated, compilation will fail.
• Skipping the contributor check when gateEnabled is false may unintentionally suppress AI advisory for confirmed contributors.

Reviewer B · nemotron-3-120b-a12b — recommends ✅ merge
The PR gates the AI slop advisory to confirmed contributors to prevent untrusted PRs from consuming the shared Workers AI budget. It adds a confirmedContributor check in runAiSlopForAdvisory, updates the neuron budget estimation to account for up to 6 model calls, and includes a new test verifying the advisory is skipped for unconfirmed contributors. The change aligns with the project's resource protection goals and maintains the public/private boundary.

Worth double-checking

• The neuron budget estimation now uses a fixed multiplier of 6 (SLOP_AI_MAX_MODEL_CALLS); verify that runGittensoryAiSlopAdvisory does not exceed 6 actual AI calls to ensure accurate budget tracking.
• Ensure the moved official detection logic in maybePublishPrPublicSurface correctly handles edge cases (e.g., missing author, disabled gate) after the refactor.

[JSONbored]

Superseded by #737. Your full-gate insight (no AI spend at all for unconfirmed authors, matching runAiReviewForAdvisory) was the right call and has been adopted in #737 — but on the current post-BYOK base, so it preserves the BYOK integration from #728 (this branch predated it and would have dropped settings/BYOK from runAiSlopForAdvisory).

One deliberate difference: I did not carry over the estimateNeurons bump from 1 → SLOP_AI_MAX_MODEL_CALLS=6. The number is accurate as a worst case (2 models × 3 retries), but the AI review path estimates by logical free-call count rather than worst-case retries, so a 6× reservation would be inconsistent with that convention and would over-reserve the daily neuron budget (premature quota_exceeded). Keeping the estimate at 1 logical call matches runAiReviewForAdvisory. Thanks for catching the vector.