The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from pull request diffs and external documentation (release notes, changelogs, migration guides) to generate a summary report. This creates an attack surface where malicious instructions could be embedded in these external sources to influence the agent's report or the subsequent actions of the commander agent.\n
Ingestion points: Pull request diffs, lock files, and external documentation URLs (release notes, GitHub changelogs, official documentation).\n
Boundary markers: The skill does not define explicit boundary markers or delimiters when interpolating untrusted data into its context or the resulting report.\n
Capability inventory: The agent can post comments to GitHub pull requests and assign labels. Its output is consumed by a commander agent that makes merging decisions.\n
Sanitization: The skill instructs the agent to ignore the pull request summary description and prioritize code and official documentation, which mitigates simple injection in descriptions but does not sanitize content from documentation or code comments.\n- [COMMAND_EXECUTION]: The skill instructions include the use of the grep command to inspect package-lock.json files for specific package versions. This is a standard, low-risk use of a local tool for data extraction from specific files.\n- [EXTERNAL_DOWNLOADS]: The skill fetches and reads content from external URLs including official documentation, release notes, and migration guides. While these are external sources, they are essential for the skill's primary function of auditing dependency updates and generally target well-known package registries or project websites.

dependency-update-report