The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its core workflow involves reading and analyzing arbitrary project files. Ingestion points: Target files and directories specified in the $ARGUMENTS variable. Boundary markers: No explicit delimiters are used to isolate untrusted file content from the agent's instructions. Capability inventory: The skill has the ability to read and write files, execute shell commands, and delegate tasks to other agents. Sanitization: No sanitization of the analyzed file content is mentioned before processing.
[COMMAND_EXECUTION]: The skill's reference materials in reference/knowledge-capture.md include instructions for using shell commands such as find and grep to perform project-wide searches and deduplication. This represents a risk if the target paths provided in arguments are maliciously crafted.
[DATA_EXFILTRATION]: The skill's analysis process covers sensitive areas of a codebase, including external interfaces and authentication setups. The templates in templates/interface-template.md explicitly prompt for documenting environment variables and authentication methods, which could lead to the exposure of sensitive architectural or credential-related information.

document