refactor(review): improve specialist causal analysis

[cv]

Outcome

Improve PR Review Advisor specialist investigation quality using independently selected prompt variants from three model-backed historical-PR A/B rounds.

Reason

The clean benchmark showed that broad prompts could attribute inherited defects to a PR, miss incomplete immutable-pin migrations, and conflate producer or rejection tests with positive consumer proof.

Changes

• Require changed-file, parent-state, consumer, and contract causality.
• Narrow architecture ownership and add migration-completion and reduction-simplification specialists.
• Adopt independently selected delivery, documentation, and verification prompts.
• Preserve tracked internal symlinks in local snapshots while removing escaping links; direct patching could not preserve this security boundary and valid transitions together.
• Protect snapshot behavior with internal-link and escaping-to-internal retarget regression coverage.

Verification

• CLI and plugin builds — passed.
• npm run typecheck:cli — passed.
• npm run checks:repository — passed.
• Focused integration tests — 47/47 passed.
• Diff whitespace check — passed.
• Independent documentation review — PASS.
• Three A/B rounds on fix(policy): reject untrusted private hosts #10659, test(e2e): select messaging and Shields runtime proofs #10536, and feat(openshell): add external gateway status #10618 with azure/openai/gpt-5.6-terra — completed.
• No secrets, API keys, or credentials are present in the diff.

Review notes

This draft is stacked on #10813. Output length and finding count were not selection criteria; prompts were selected independently for factuality, recall, causality, calibration, and remedies.

---

Signed-off-by: Carlos Villela cvillela@nvidia.com

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4b7fb494-cb72-42ec-a5fe-e3fd6a40ff96

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-code-quality]

Code Coverage Overview

Languages: TypeScript

TypeScript / code-coverage/plugin

The overall line coverage in commit 6066995 in the feat/advisor-causal-... branch remains at 96%, unchanged from commit 19bb986 in the refactor/pr-advisor-... branch.

TypeScript / code-coverage/cli

The overall line coverage in commit 6066995 in the feat/advisor-causal-... branch remains at 83%, unchanged from commit 19bb986 in the refactor/pr-advisor-... branch.

Show a line coverage summary of the most impacted files.

File	refactor/pr-advisor-... 19bb986	feat/advisor-causal-... 6066995	+/-
src/lib/onboard...aging-config.ts	90%	86%	-4%
src/lib/onboard...host-forward.ts	84%	81%	-3%
src/lib/policy/index.ts	74%	74%	0%
src/lib/onboard...dge-provider.ts	91%	91%	0%
src/lib/sandbox...rce-identity.ts	82%	82%	0%
src/lib/onboard...uild-context.ts	74%	75%	+1%

---

_{Updated September 02, 2026 18:00 UTC}

[github-actions]

PR Review Advisor finished for commit 6066995. Include the Advisor findings in the complete PR feedback collection. Verify and group valid findings before repair.

All previous runs