The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted changelog data from external package registries and GitHub to generate PR summaries. This creates an attack surface where malicious input could influence the LLM's output. Ingestion points: SKILL.md (Step 5) and references/changelog-patterns.md describe fetching data from registry.npmjs.org, pypi.org, and api.github.com. Boundary markers: There are no specific delimiters or protective instructions used when sending the external changelog text to the Gemini API for summarization. Capability inventory: The resulting summary is written to a file and used as the body for gh pr create and git commit commands in SKILL.md (Step 7). Sanitization: The skill does not sanitize or filter the external content before processing it with the LLM.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to well-known registries and services including registry.npmjs.org, pypi.org, and api.github.com. It retrieves repository data, changelog files from raw.githubusercontent.com, and interacts with generativelanguage.googleapis.com for content generation.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools and local scripting. It runs package management commands (npm, pip, cargo, go, bundle), manages branches and commits using git, and uses gh (GitHub CLI) for creating pull requests. It employs python3 -c to parse JSON data from tool outputs and API responses; while this pattern was flagged as potential remote code execution by automated scans, the analysis confirms it is used for data processing of JSON content rather than executing remote scripts.

dependency-update-bot