The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses python3 -c and shell redirection (cat >>) to perform local file operations on ~/.scienceclaw/memory/findings.jsonl. These commands are used to append research records and perform periodic maintenance and recall of stored memory objects.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it stores and later retrieves data originating from external research tasks. Malicious instructions embedded in processed papers or API results could be persisted as 'strategies' and subsequently influence the agent's methodology in future sessions.
Ingestion points: External research findings, tool outputs, and hypothesis validation data processed during research recipes.
Boundary markers: Recalled memory records are re-injected into the agent's context without clear delimiters or instructions to ignore potential commands within the data.
Capability inventory: The agent has access to shell execution, file system modification, and various research tool APIs.
Sanitization: No sanitization or filtering logic is present to identify or remove instructions from the data before it is stored or recalled.

evolving-memory