fastify-workflows
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill guides the maintenance of workflows that ingest untrusted data from GitHub pull requests, creating a potential surface for indirect prompt injection.
- Ingestion points: Untrusted data enters the agent/workflow context via
github.event.pull_requestmetadata, including the contributor's login, repository name, and branch names inplugins-ci.ymlandreusable-release.yml. - Boundary markers: The skill explicitly instructs developers to use boundary markers by binding inputs to environment variables (
env:) rather than using direct interpolation (${{ inputs.* }}) in shell scripts, which reduces the risk of malicious input executing code. - Capability inventory: The workflows have significant capabilities, including the ability to write to the repository (
contents: write), publish to npm via OIDC (id-token: write), and execute arbitrary shell commands in the runner environment. - Sanitization: The skill mandates sanitization practices, such as validating the
BUMP_TYPEinput against a strict allowlist (patch, minor, major) and usingprintfto safely handle environment variables in shell commands. - [EXTERNAL_DOWNLOADS]: The skill documents the installation of the Yarn package manager from a remote source.
- Evidence:
references/ci-workflows.mddescribes installing Yarn viahttps://yarnpkg.com/install.shand executing it withbash. - Context: This is a standard installation method for Yarn 1.x targeting a well-known, official service domain (
yarnpkg.com).
Audit Metadata