The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation describes the use of a local Python script (scripts/decision_tracker.py) to perform administrative tasks such as searching decisions, checking for conflicts, and identifying overdue action items. This script is part of the skill's internal logic and is executed via standard command-line interfaces.
[PROMPT_INJECTION]: The skill handles indirect prompt injection surfaces by processing board meeting transcripts. Evidence Chain: 1. Ingestion points: Raw transcripts stored in memory/board-meetings/ are read during the synthesis phase. 2. Boundary markers: Data is structured using markdown headers but lacks explicit instruction-filtering delimiters. 3. Capability inventory: The skill interacts with the local file system and internal Python scripts. 4. Sanitization: A mandatory human-in-the-loop (founder approval) step filters content before it reaches the long-term Layer 2 memory, serving as a primary defense against adversarial content.
[DATA_EXPOSURE]: The skill handles sensitive business data, including board minutes and strategic rationales. This information is stored locally in markdown files within the memory/ directory. No network-based exfiltration or unauthorized external communication patterns were detected.

decision-logger