The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests board meeting transcripts and decision logs, which creates a potential surface for indirect prompt injection from untrusted meeting content.
Ingestion points: The scripts/decision_tracker.py tool reads meeting data from the local ~/.claude/decisions/approved/decisions.md file.
Boundary markers: The parser uses markdown headers (e.g., ## [YYYY-MM-DD]) and field labels (e.g., **Decision:**) to structure and delimit data.
Capability inventory: The Python reporting tool is restricted to local file reading and text output; it lacks network access, file-writing capabilities, or command execution logic.
Sanitization: While the script extracts specific fields via regular expressions, it does not perform explicit sanitization of the processed text for downstream LLM interaction.
[SAFE]: The skill follows best practices by storing its state in the agent's dedicated directory (~/.claude/decisions/). The included Python script uses only standard library modules and performs legitimate tracking and conflict detection logic without dangerous system calls.

decision-logger