The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests pull request titles, merge request titles, and commit messages to generate release notes, creating an attack surface for indirect prompt injection.
Ingestion points: External data is fetched from GitHub, GitLab, and Gitea via gh, glab, and tea CLI tools, as well as local git logs, in scripts/get-notes.sh.
Boundary markers: The skill implements a human-in-the-loop checkpoint in Step 8 and Step 9 of SKILL.md, requiring explicit user confirmation via the AskUserQuestion tool before proceeding with publication.
Capability inventory: The skill uses the Bash tool to modify the local filesystem (updating CHANGELOG.md), commit changes to the repository, and publish releases to remote platforms (SKILL.md).
Sanitization: Data is parsed using jq and prefixes are stripped using sed in scripts/get-notes.sh, but no content-based sanitization or filtering of the injected strings is performed.
[COMMAND_EXECUTION]: The skill relies on the Bash tool for core logic, including git operations, semantic version calculation, and interaction with platform CLIs. While variables like "$notes" and "$new_version" are shell-quoted to mitigate command injection, the execution of these tools constitutes a significant capability within the agent's environment.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch PR/MR metadata and push releases to external platforms (GitHub, GitLab, and Gitea) using their official command-line interfaces.

new