- Home
- Skills
- Testing & Debugging
- evidence first pr risk auditor
More screenshots
Works with the AI Agents you already use
evidence first pr risk auditor
Read-only pull request and Git diff risk reviews with traceable evidence, confidence labels, targeted regression checks and an optional offline Python preflight. No external API keys required.
$10
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
evidence first pr risk auditor
Example session with this skill installed
Review the attached synthetic Git patch with Evidence-First PR Risk Auditor. Keep the review strictly read-only.
Run the bundled local Python preflight if available. Distinguish observed code changes from unproven security impacts, cite changed file and line locations, and propose targeted regression tests.
State exactly which checks were executed and which were only recommended.
Input: examples/synthetic-auth-timeout.patch, a synthetic fixture containing changes to src/auth.py and src/api.ts.
- Read your context and instructions
- Compiled the evidence first pr
Pull Request Risk Review — Synthetic Demonstration
Verdict: REVIEW REQUIRED
Potential authentication boundary regression detected. The supplied patch does not provide enough context to confirm an exploitable security vulnerability.
Scope
The review covers examples/synthetic-auth-timeout.patch, containing two fictional changed files.
Executed checks
- Offline deterministic Python preflight
- Unit test suite: 12/12 tests passed
- Two heuristic risk candidates reported
- No target repository files modified
Findings
| Priority | Confidence | Location | Finding |
|---|---|---|---|
| High | INVESTIGATE | src/auth.py, old line 8 | Authentication guard removed |
| Medium | INVESTIGATE | src/api.ts, new line 4 | Request timeout behavior requires investigation |
Authentication boundary
The patch removes a condition checking user.is_authenticated and introduces a check based on user.id.
This could weaken the authentication boundary if a user identifier remains available after a session is invalidated. However, route middleware and session handling are not visible in the supplied patch.
Recommended verification: Inspect the complete authorization flow and add regression tests covering anonymous users, expired sessions and valid authenticated sessions.
Network reliability
A request using explicit timeout handling is replaced with fetch(url).
The patch does not show whether timeout or cancellation handling exists elsewhere.
Recommended verification: Inspect caller-level cancellation behavior and add tests for slow, interrupted and failed requests.
Evidence boundary
The preflight was executed on a synthetic Git patch, and the deterministic unit tests passed.
No production application was tested. No runtime exploit was attempted. No external network requests were made.
Final assessment
Both findings require additional context and focused regression tests before a definitive merge decision.
Evidence-first reviewing means documenting uncertainty rather than presenting assumptions as confirmed defects.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Evidence-First PR Risk Auditor
Review code changes with evidence, not guesswork.
Evidence-First PR Risk Auditor is a structured, read-only review workflow for Git diffs and pull requests. It helps developers identify change-related risks, produce auditable findings, and define targeted regression checks.
Capabilities
- Analyze supplied code changes for potential authentication, authorization, validation, network, persistence, and reliability regressions.
- Reference actual file paths and old/new diff line numbers.
- Distinguish confirmed defects from likely risks and unresolved hypotheses.
- Recommend minimal remediation, verification steps, and focused regression tests.
- Run an optional, offline Python preflight that produces deterministic risk candidates.
- Operate within explicitly authorized source-code boundaries without modifying repositories.
Demonstrated engineering foundation
The original Agentic PR Risk Auditor research project used a deterministic evidence collector combined with a locally executed qwen2.5-coder:7b model.
On its locked 24-case benchmark, the selected workflow achieved:
- Precision: 1.000
- Recall: 0.889
- F1: 0.941
- Exact-case accuracy: 0.917
The baseline achieved F1 0.750. The experiment also documented two false negatives and rejected an additional verifier stage after it produced worse results.
These results belong to the original model-powered research implementation, not to the portable skill package.
The portable v1.1.0 release separately passed 12 deterministic unit tests and a reproducible synthetic diff demonstration.
What's included
A complete SKILL.md, risk taxonomy, report template, Python preflight, reproducible example, unit tests, test results, and evidence-provenance documentation.
Requirements
A compatible AI agent able to read SKILL.md and a user-provided Git diff. Python 3 is optional.
The preflight requires no paid API, external service, credentials, or network connection.
Limitations
This skill is not a penetration-testing tool or security certification. Heuristic findings require contextual verification, and false positives or false negatives remain possible. It does not automatically edit files, commit changes, deploy applications, or access credentials.
Source project: https://github.com/kostasuser01gr/agentic-pr-risk-auditor-submission
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.
Static analysis of supplied code changes only. Heuristic findings can include false positives and false negatives. Full application context and independent testing remain necessary before confirming security defects.
The optional Python preflight identifies candidate patterns, not proven vulnerabilities.
The original research project's 24-case benchmark and F1 score of 0.941 do not represent an accuracy measurement of the portable v1.1.0 helper.
The skill does not automatically modify repositories, access credentials, deploy applications or perform active security testing.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 17 days ago
- Accounts
Apple Music MCP
Permissions
File Scopes
This skill performs read-only analysis of explicitly provided or authorized code changes. File access is limited to the review scope approved by the user. Terminal access is optional and used exclusively to run the local Python preflight. The skill does not modify source files, create commits, deploy applications, access credentials or transmit code to external services through its bundled helper. Findings are evidence-backed and include explicit uncertainty labels. No security guarantees or unverified execution results are claimed.