Aurais Code Review Sentinel
Adversarial review · never silently approves
Paste a diff + context. Get a structured review: blockers, security, performance, maintainability, tests to add. Minimum 3 findings per review, always.
What it does
- Severity + category on every finding
- File + line hint cited for every issue
- Suggested direction, never a rewrite
- Security flags independent of what the PR claims to touch
- Tests-to-add list surfaced from the review
Adversarial by design
The bot's system prompt forbids silent approvals. Every review must surface at least 3 findings, even if they're small. If a diff is truly clean, you see the nits; if it's risky, you see the blockers. Every finding cites file + line hint; nothing hand-waved.
- car_id
- car-aurais-code-review-sentinel-c6d4421a0549
- agent_id
- aurais-code-review-sentinel@0.1.0
- org_id
- vorion-llc
- deployment_id
- dpl_4TWfrDX9SjZyvKEYtF8BWBnzY2WS
- tier
- T3 · BLACK_BOX
- trust_ceiling
- 600
- max_earnable_tier
- T4
- parent_hash
- (root — AgentAnchor A3I)
- context_hash
- sha256:f9e9be5554153d62a…
- capabilities
- 7 declared
- a3i_cert
- registry/code-review-sentinel →
7 capabilities
- · tool:adversarial_review
- · tool:security_scan
- · tool:suggest_tests
- · data:read:user:diff-transient
- · data:write:none
- · safety:never_silently_approves
- · api:post:api.anthropic.com
What “offline-attested” means: The CAR ID and capabilities are derived deterministically from the signed manifest and emitted in every proof chain — identity is verifiable. Live trust-score progression, lifecycle state transitions, and pre-action gating require the Vorion registry backend, which is in roadmap, not shipped. Tier 3 is the published operating envelope; the bot cannot silently exceed it.
Pricing
Free at BYOK · $12/mo hosted · Series of 50 (limited)