The Agent Skills Directory

[COMMAND_EXECUTION]: The file scripts/test_taste_memory.py utilizes the subprocess module to perform automated testing of the skill's internal CLI. These calls are confined to testing the skill's own logic within the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill manages a local JSON store that ingests external data, including design profiles and user feedback, which are subsequently returned to the agent's context. This ingestion creates a potential surface for indirect prompt injection.
Ingestion points: Data enters the system via the add command (which reads profile files) and the feedback command in scripts/reference_memory.py.
Boundary markers: The data is stored in structured JSON formats; however, the skill does not use specific delimiters or instructions to ensure the agent ignores malicious content embedded within user-provided strings like labels or notes.
Capability inventory: The skill performs local file read/write operations and utilizes subprocesses for internal testing.
Sanitization: Content is processed using standard JSON parsing, which maintains data structure but does not validate or sanitize the semantic content of the stored text fields.

reference-memory