Repository AuditWATCHUNKNOWN

AI Agent Runtime Review

Repository · Report VS-UNKNOWN-REPO-AI

Score / 100

Verdict

Monitor before public promotion

Mixed signals across activity, documentation or dependency surface. Review the evidence before relying on this repository.

Surface analyzed: Repository Audit
Scan source confidence: 74%
Methodology: Public sources · v0
Generated: May 25, 2026, 3:00 PM UTC

Upgrade path

Need a private evidence memo?

Turn this public artifact into an operator-reviewed Private Review memo. Paid review stays separate from the public score.

Request Private Review

Share this report

Ready

Public-source signals. Not a substitute for a formal audit. No buy/sell recommendation.

Elevated signal

Engineering controls

56/100

Elevated signal

Evidence depth

52/100

Elevated signal

Issue and PR flow

48/100

Elevated signal

Maintainer control

46/100

REPO

Repository identity

AI Agent Runtime Review

Generated May 25, 2026, 3:00 PM

Scanned GitHub repository

https://github.com/agent-stack/policy-runtime

VegaSec summary

This AI-agent runtime example is WATCH: public evidence shows policy and logging surfaces, but permissions, auditability and deployment claims still need operator review.

Repository freshness

Default branch state at scan time

Public-metadata triage on commit activity. This is not a code audit; it does not run against the repository's source, only against its public GitHub metadata.

Recent · pushed in the last 90 days

Default branch: main
Latest push: Apr 28, 2026, 9:00 AM(27 d ago)
Latest commit SHA: dd0a11c7e9a0
Commit timestamp: Apr 28, 2026, 8:45 AM

Re-submit the same repository URL on /scan to refresh the snapshot. Repositories with abandoned or stale activity are removed from Trust Index eligibility automatically — see docs/specs/trust-index.md.

Risk exposure

Risk exposure by category

Each row is a public-evidence dimension the audit scored. Higher values mean more signal worth investigating before relying on this repository.

Engineering controls
Elevated signal
56/100
Evidence depth
Elevated signal
52/100
Issue and PR flow
Elevated signal
48/100
Maintainer control
Elevated signal
46/100
Community surface
Moderate signal
44/100
Activity cadence
Moderate signal
40/100
Docs and security signal
Moderate signal
39/100

Repository hygiene checklist

Repository-specific checks for crypto codebases: freshness, engineering controls, tests, security policy, and dependency posture. This is metadata triage, not a code audit.

!
Default branch freshness
Recent pushes and commit timestamps captured at scan time.
!
CI / build workflow signal
Public workflow or build automation signals that indicate regression controls.
!
Automated test signal
Presence and depth of public test files or test runner evidence.
!
License and provenance
Public license and repository identity signals for reuse and operator review.
!
Security policy / disclosure
SECURITY.md, disclosure guidance, or explicit security contact evidence.
!
Dependency posture
Manifest and dependency-update signals that affect supply-chain hygiene.
!
Maintainer and community signals
Stars, forks, issue flow, and maintainer surface used as context, never as proof of safety.

✓ Pass! Caution✕ Fail— Unknown

Audit pack

Repository hygiene + security posture

Per-dimension reasoning over the public GitHub metadata captured at scan time. Hygiene markers are not a security audit; UNKNOWN entries are gaps for operator review.

Stalenessadequate
Evidence
- Latest push observed 27 days ago on main.
Continuous integrationunknown
Caveats
- CI workflow signal was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Automated testsunknown
Caveats
- Test file count was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Licenseunknown
Caveats
- License signal was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Security policyunknown
Caveats
- SECURITY.md signal was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Release hygieneunknown
Caveats
- Release tag signal was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Dependency postureunknown
Caveats
- Dependabot / dependency update bot signal was not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.
Maintainer signalsunknown
Caveats
- Maintainer signals (stars / forks / open issues) were not captured at scan time.
Next step
- Operator review required to gather the missing signal before this dimension can be scored.

Evidence

Evidence that explains the verdict

Each row is one public-data signal pulled from automated sources. Use it as a starting point for review, not as a final answer.

Permission boundaries are visible but need review
Audit-log behavior should be verified
Runtime policy claims require evidence

Baseline source confidence

Methodology baseline used by the verdict engine. Higher baseline reduces the penalty for ambiguous data; low baseline pushes the verdict toward UNKNOWN/WATCH. May differ from the scan source confidence shown at the top when the live scan supplied a fresher signal.

Badges

repository-auditagent-runtimepolicy-review-needed

How to read this report

Score, confidence, ranking — four moving parts, no hand-tuning.

Generated 2026-05-25T12:00:00.000-03:00 · Repository · REPO-AI

Score bands

0 – 19
AVOID-style triage
Multiple high or critical findings, low source confidence, or unverified context. Public action should pause until a Private Review or a re-scan with better evidence.
20 – 49
High caution
Notable findings or thin evidence. Worth a Private Review before relying on the artifact.
50 – 69
Watch · this report
Mixed signals: some evidence, some gaps. Re-scan after material changes to the target.
70 – 89
Clear surface (automated)
No critical or high findings via static analysis. Not a safety claim — human review still recommended for diligence work.
90 – 100
Strong surface (automated)
No critical or high findings, verified source, and high source confidence. Not a safety claim — human review still recommended for diligence work.

Score: Deterministic 0–100 derived from severity counts. Same formula across every report. Lower means more risk.
Confidence: Separate axis. Measures how much VegaSec trusts the input data. Low confidence pushes the verdict toward UNKNOWN; the two are never collapsed into one number.
Ranking / percentile: Only meaningful within the eligible Trust Index set. Shown on /trust-index when the eligible pool is large enough. Today the public index is a controlled preview.

Private Review is separate

A paid Private Review does not raise the public score on this report. Public score reflects public-source evidence only. Private Review adds operator attestation and a confidential memo, never a score bump.

Export and share

· Shareable public URL: copy from the report actions panel.
· Markdown export: every export carries the report id, target, verdict, generated timestamp, and the public-data disclaimer.
· PDF export: not in this MVP. Tracked as a follow-up — the Markdown export contains the same content and prints cleanly to PDF from a browser today.

Community artifact

Report ready to share, copy or export.

This link preserves the scanned target, route, confidence and Scan ID. Example report links are marked as public samples; live Token Audit and Repository Audit links require submitted context and fail closed to UNKNOWN when evidence is missing.

Ready

VegaSec is intelligence and triage. It is not financial advice, does not promise returns and does not replace manual due diligence.