sdk-package-publishing
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill implements strict operating rules that prevent the agent from handling actual secrets, passwords, or private keys, instead using placeholders and environment variables. It also prohibits the agent from executing irreversible operations (like publishing or unpublishing) directly, requiring them to be handed to the user for manual review and execution.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze project manifest files (e.g., package.json, pyproject.toml, pom.xml) and CI configuration files. While these ingestion points allow external data into the agent's context, the skill mitigates risks through explicit validation instructions (e.g., tag-equals-version guards) and by ensuring the agent only acts as a consultant rather than an autonomous executor.
- Ingestion points: manifest files (package.json, pyproject.toml, etc.) and CI configuration files.
- Capability inventory: The skill generates shell commands for package publication and registry management but does not execute them.
- Sanitization: Instructions prioritize dry-runs and human-in-the-loop review for all commands.
- Boundary markers: Uses placeholders for secrets and explicit command structures to delineate instructions.- [EXTERNAL_DOWNLOADS]: The skill recommends the use of official CI actions and tools from trusted organizations (e.g., GitHub Actions, PyPA, Rust-lang, RubyGems, Sonatype). All referenced external resources are well-known services within the established trust scope.
Audit Metadata