The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its automated analysis of session transcripts.
Ingestion points: Session transcripts are loaded from ~/.codex/sessions/ and processed in scripts/run_insight_report.py.
Boundary markers: Transcripts are interpolated directly into prompts for the local LLM without explicit delimiters to separate user data from instructions.
Capability inventory: The skill can execute shell commands via subprocess.run (to interact with the Ollama API) and open local HTML files in a browser.
Sanitization: Session content is truncated to 8,000 characters but is not otherwise sanitized or escaped before being sent to the LLM.
[EXTERNAL_DOWNLOADS]: The skill's documentation and scripts suggest installing the Ollama service using an installation script fetched from ollama.com, which is a well-known service for local LLM inference.
[COMMAND_EXECUTION]: The scripts/run_insight_report.py script uses subprocess.run() to execute curl commands for communicating with the local Ollama API endpoint.

insight-report