
Repository Risk Simulator
Maps GitHub dependencies to OSV advisories and renders evidence-linked risk scenarios in an offline HTML report.
Baseline review at submission: an accuracy and overt-behavior check, not a full security audit, and it does not cover later changes to the repository. Learn more about our security review process.
For AI agents: this listing as JSON or Markdown. It is one of 148 listings in the full index.
Dependency triage usually stops at a list of advisories matched to package versions, which leaves the hardest question unanswered: does any of this matter for this application? This skill keeps that question visible. It pairs a GitHub-hosted SBOM and Google OSV advisory data with inspected source evidence, then models up to three hypothetical scenarios whose prerequisites you can toggle — while refusing to claim that any vulnerability is reachable or exploitable.
What it does
Given a GitHub repository URL, it produces a self-contained HTML report that works offline after generation:
- Inventory — starts GitHub’s asynchronous SPDX export and downloads the result. No local scanner, no Docker service, no MCP server, and no separate model API key.
- Advisory matching — queries Google OSV for packages with supported, exactly-versioned
purls (npm, PyPI, Maven, RubyGems, Go, crates.io, NuGet, Packagist, Hex, Pub), recording
matched,no_match,unsupported, anderrorstatuses separately so coverage gaps stay visible rather than reading as clean results. - Source evidence — slices exact numbered line ranges from manifests, lockfiles, imports, call sites, and configuration, recording the source commit and a SHA-256 for each excerpt.
- Assessment — marks each finding
unknown,potentially_applicable, orruled_out_under_evidence, with cited evidence and per-finding limitations. - Scenarios — up to three, each linked to a real finding, separating facts from assumptions and unknowns, stating its application evidence gap explicitly, and defining 1–8 prerequisite controls you can switch in the report.
Deliberate boundaries: it attempts no exploitation, includes no payloads or runnable exploit commands, never executes code from the analyzed repository, and never modifies it. Results are static assessments, not certified VEX.
How it works
Python 3.10+ standard-library helpers handle collection, matching, validation, and rendering;
the current Claude session does the reasoning over the downloaded evidence. The pipeline runs
preflight → collect → lookup → tree/source/evidence → init-analysis →
validate → render, and every intermediate artifact is preserved next to the report.
Three outbound HTTPS destinations are used and documented in the repository: api.github.com,
api.osv.dev, and GitHub’s SBOM download redirect host. Authorization is stripped on
cross-host redirects, signed download URLs are never stored, and an existing GITHUB_TOKEN or
GH_TOKEN is read from the environment only — never requested in chat or written into
artifacts.
The report embeds its own data, styles, and scripts, blocks network access with a Content
Security Policy, and renders untrusted text with textContent. Advisory URLs appear as text,
not as automatic requests. Scenario simulation is deterministic: any known contradiction blocks
the modeled path, otherwise any unknown yields unknown, otherwise prerequisites are satisfied.
A blocked result applies only to that scenario’s explicit prerequisites and rules out nothing
else.
A labeled synthetic demo (scripts/demo.py) and two test suites ship with the skill for
offline verification.