The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface through its memory persistence mechanism, where malicious instructions from a session can be recorded and later executed by the agent.\n
Ingestion points: The skill ingests data from the current session via the task_description and details fields processed by the calculate_task_attention and extract_key_details functions in attention_scorer.py.\n
Boundary markers: The markdown logs generated by the skill do not use delimiters or explicit 'ignore embedded instructions' warnings when documenting task details, which could lead to an agent accidentally obeying commands stored in the logs.\n
Capability inventory: The skill is explicitly designed to write to the filesystem (into the memory/ directory) and encourages the agent to read these logs in future sessions to maintain continuity.\n
Sanitization: The implementation in attention_scorer.py lacks any sanitization, filtering, or validation of the task descriptions or details before they are written to the persistent log files.

daily-log