The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a customization feature that reads instructions from local .overlay.md files. This creates an attack surface where a malicious repository could influence the agent's behavior by providing these files, which the agent is instructed to treat as valid rules.
Ingestion points: The lib/load-overlays.sh script reads files matching *.overlay.md from the current project directory and the user's home directory.
Boundary markers: The script includes header markers (e.g., === commit rule... ===) when outputting the content to the agent, providing some separation but not full isolation.
Capability inventory: The agent can execute shell commands via git and bash, and can write to local files (feedback.md).
Sanitization: No sanitization or validation is performed on the content of the overlay files before they are interpreted by the agent.
[COMMAND_EXECUTION]: The skill relies on shell scripts and git commands to automate the commit process and manage configuration.
The skill executes bash "${SKILL}/lib/load-overlays.sh" to resolve rules.
The skill executes bash "${SKILL}/lib/record-feedback.sh" to prepare the logging environment.
The skill performs system information collection by running the hostname command, which is then recorded in a local log file.

commit