The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands such as git rev-parse, ls, and mkdir to manage its internal file structure and session history. These commands are used for local environment discovery and are restricted to the repository's filesystem.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing and logging untrusted user input without utilizing boundary markers or sanitization.
Ingestion points: User answers to the reflection questions are captured and stored in the local memory directory.
Boundary markers: The instructions do not include explicit delimiters or warnings to ignore embedded instructions within user responses when generating logs or summaries.
Capability inventory: The agent has the ability to write files and execute directory listing commands, which could be influenced by malicious content if the logs were later processed as instructions.
Sanitization: User input is interpolated directly into markdown templates without filtering or escaping.

bampenpien