The Agent Skills Directory

[DATA_EXPOSURE]: The script scripts/daily-meeting-update/claude_digest.py accesses the ~/.claude/projects directory to read session logs stored in JSONL format. This data includes full message history, git branch names, file paths, and bash command counts. While this is used for a 'Daily Meeting Update' feature, it results in the exposure of potentially sensitive development history and environment details to the agent's active context.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection via the digestion of previous conversation logs.
Ingestion points: scripts/daily-meeting-update/claude_digest.py reads and parses message content from local .jsonl files located in the ~/.claude/projects hierarchy.
Boundary markers: The script does not utilize delimiters or specific instructions to isolate the parsed text, increasing the risk that the agent may follow instructions embedded within the old logs.
Capability inventory: The script includes file system read access and JSON parsing; the overall skill context allows for standard agent operations.
Sanitization: No filtering or sanitization is performed on the text content extracted from the message logs before it is presented to the agent.

workplace-pro