The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the git command using the subprocess module in scripts/harness.py to identify project roots. The implementation uses structured arguments and environment scrubbing to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to persist project rules and context which the agent later retrieves and follows. This creates a vulnerability where an attacker who can influence the project files could inject instructions that the agent later adopts as valid guidance.
Ingestion points: Data ingestion occurs in scripts/harness.py through the write and handoff operations, which read content from specified files or standard input.
Boundary markers: There are no explicit delimiters used when the agent reads the stored notes to prevent the execution of embedded instructions.
Capability inventory: The skill provides capabilities for local file system manipulation (read, write, delete) and basic git status checks.
Sanitization: The knowledge_path function in the helper script validates filenames for the .md extension and ensures they do not contain directory traversal sequences like ...

harness-remember