dependency-upgrader
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including package manifests, lockfiles, and vulnerability advisory databases which could potentially contain malicious instructions.\n
- Ingestion points: Dependency manifests (package.json, pom.xml, build.gradle), lockfiles (package-lock.json, yarn.lock), and output from auditing tools (npm audit, gh api, osv-scanner). Found in SKILL.md and referenced playbook files.\n
- Boundary markers: The skill does not define technical delimiters for ingested data, relying instead on natural language instructions for the agent to distinguish tool output from core instructions.\n
- Capability inventory: Shell command execution for dependency management and testing; file system write access for manifest updates; network access via CLI tools for registry and advisory lookups.\n
- Sanitization: Instructions emphasize manual verification, reviewing diffs before execution, using ignore-scripts where supported, and comparing findings against a baseline audit.\n- [COMMAND_EXECUTION]: Executes standard package manager and build system commands (npm, pnpm, yarn, bun, gradle, maven) and security CLI tools (gh, osv-scanner) to resolve dependency trees and run security scans.\n- [EXTERNAL_DOWNLOADS]: Fetches metadata, release notes, and security advisories from well-known services such as GitHub, NPM, Maven Central, and the OSV database. These actions are consistent with the skill's primary purpose and use official repositories.
Audit Metadata