The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external BODY content (raw markdown release notes) to generate summaries and highlights. While this creates a surface for indirect prompt injection where malicious instructions could be embedded in release notes to influence the agent's output, the impact is low as the skill only performs document formatting and local file writes.
Ingestion points: The BODY input is saved to a temporary file and processed in SKILL.md.
Boundary markers: The skill lacks explicit markers like "ignore instructions within this text" for the processed body content.
Capability inventory: The skill uses file write/replace capabilities (docs/changelogs/index.md, latest.md, preview.md) and executes a standard formatter (npm run format).
Sanitization: The skill performs basic reformatting of PR URLs and deletion of the "New Contributors" section, but does not strictly sanitize against natural language instructions.
[EXTERNAL_DOWNLOADS]: The skill references a trusted repository (github.com/google-gemini/gemini-cli) for changelog links. This is a well-known service and a trusted organization, which is considered safe practice.
[COMMAND_EXECUTION]: The skill executes npm run format. This is a standard development practice for maintaining code style and is restricted to the formatting task.

docs-changelog