The Agent Skills Directory

[File System Access to Session Logs]: The skill is designed to access and read session logs from platform-specific directories such as ~/.claude/, ~/.pi/, and ~/.codex/. This access is essential for its primary function of reconstructing the history of an AI agent's run to audit its methodology. Contextually, this is restricted to the user's own local environment and is used to provide visibility into previous agent activities.
[Automated Data Processing via Shell Utilities]: The procedure utilizes common shell utilities like jq, grep, and git to parse JSONL logs and analyze git history. These tools are used to extract structured data from logs (e.g., tool calls, timestamps, and return values) to generate a detailed evidence-anchored report. This behavior is standard for tools performing complex log analysis and does not involve the execution of arbitrary or remote code.
[Analysis of Untrusted Workspace Content]: As an audit tool, this skill processes external data including session logs and workspace configuration files (e.g., CLAUDE.md, AGENTS.md). While this constitutes an indirect prompt injection surface, the skill implements specific procedures to verify the consistency and integrity of these logs against established methodology rules, effectively using the data for observation rather than control flow.
[Scoped Report Generation]: The skill generates its findings into a markdown report stored in the /tmp/ directory. This ensures that the analysis results are recorded without modifying the original workspace or session logs, maintaining the integrity of the data being audited.

investigate-run