The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies the .qa/ knowledge base as a potential vector for indirect prompt injection. It mitigates this by instructing the agent to treat entries as data rather than instructions, implementing a validator that flags command-like phrases (e.g., "ignore validation"), and requiring human review for modifications.
[CREDENTIALS_UNSAFE]: The memory_validate.py script includes dedicated regex scanners to detect and prevent the accidental inclusion of secrets (such as API keys, private keys, and JWTs) within the knowledge base files. No hardcoded credentials are present in the skill itself.
[DYNAMIC_EXECUTION]: The script uses __import__("sys").exit at the end of the file. This is a common and benign Python idiom used to import the system module locally for a clean exit, rather than dynamic execution of untrusted code.
[COMMAND_EXECUTION]: The skill utilizes Git commands (e.g., git rev-parse) and local Python execution to validate files. These operations are scoped to the local project directory and are necessary for the skill's functionality.

qa-memory