The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and analyze untrusted data from git diffs and repository files. If these files contain instructions designed to influence the agent's behavior, the agent might interpret them as instructions rather than data.
Ingestion Points: Data enters the agent's context through git diff commands and by reading the full content of changed files (e.g., Markdown, JS, HTML) as instructed in SKILL.md.
Boundary Markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings to help the agent distinguish between the skill's instructions and the content being reviewed.
Capability Inventory: The agent has the capability to execute shell commands like git and grep and read any file in the repository context.
Sanitization: There is no explicit sanitization or filtering of the content retrieved from the files before it is processed by the model.
Command Execution: The skill utilizes shell commands such as git diff and grep to perform its review tasks. This allows the agent to interact with the local file system and version control history as intended by the skill's design to identify and read changed content.

review-changes