The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/analyze.py) to perform usage data analysis and prompt structure evaluation. This execution is limited to local file system operations and results in a generated report.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from historical session logs and system prompt content. While this presents an attack surface where past malicious inputs could theoretically be processed, the risk is minimal as the script only performs statistical counting and structural analysis.
Ingestion points: Raw conversation logs located in ~/.pi/agent/sessions/ and system prompt text dumped via the extension.
Boundary markers: None explicitly defined in the data ingestion phase.
Capability inventory: Limited to local file reads, regex parsing, and generating a Markdown text report.
Sanitization: Content is parsed for tool/skill names and counted; no part of the ingested data is treated as executable code.
[DATA_EXPOSURE]: The skill accesses sensitive local files, including conversation history and system prompt configurations, to provide diagnostics. All analyzed data remains within the user's local ~/.pi/agent/insight/ directory and is not transmitted over the network.

pi-insight