The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands for its core automation features, specifically for project type detection and pull request analysis.
Evidence (references/auto-init.md): Uses find, grep, and directory traversal (..) to search for Compose-related strings in Gradle build files and Kotlin source code during session initialization.
Evidence (references/pr-review.md): Instructs the agent to use gh pr diff, gh pr view, and gh api to fetch data from GitHub. It also utilizes cat and find to inspect local project configuration files like .editorconfig and lint settings.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of external data from GitHub repositories to perform code reviews.
Evidence (references/pr-review.md): Fetches code diffs and full file contents from the GitHub API using the gh CLI. While these downloads target well-known services (GitHub), the content being retrieved is untrusted code from pull requests provided by the user.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its automated analysis of external, untrusted code.
Ingestion points: Pull request diffs and file contents fetched from remote GitHub repositories (references/pr-review.md).
Boundary markers: While the skill provides a structured workflow for reviews and emphasizes local reporting, it lacks explicit instructions for using delimiters to isolate untrusted content from the agent's internal instruction set.
Capability inventory: The agent can execute shell commands (gh, find, grep, cat) which provides a mechanism for malicious instructions embedded in a PR to potentially influence the environment if not handled carefully.
Sanitization: There is no evidence of explicit sanitization or filtering of the external code before it is processed by the AI agent.

jetpack-compose-expert