The Agent Skills Directory

[SAFE]: The skill is designed to read and analyze Claude Code session logs stored in ~/.claude/projects/. These logs contain sensitive data including conversation history, file contents, and internal reasoning. Accessing this data is documented as the core functionality of the skill, and the provided cc-session script performs these operations locally without external data transmission.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes past session logs which may contain untrusted data from previous interactions.
Ingestion points: Session logs in JSONL format are read from ~/.claude/projects/ using the cc-session script or manual jq recipes.
Boundary markers: No explicit instruction delimiters or 'ignore' directives are used when the agent processes the extracted log content.
Capability inventory: The agent has access to Bash, Read, Grep, and Glob tools for file access and command execution.
Sanitization: The skill extracts and displays text from logs without sanitizing or escaping natural language content that could contain embedded instructions.

introspect