The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive data stored in the ~/.claude/projects/ directory. These logs include full conversation transcripts, file snapshots, and internal assistant reasoning blocks. While the data is processed locally, it represents exposure of highly sensitive information to the agent.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze session logs which are a repository of all past tool outputs, user messages, and external data fetched during a session. This creates a surface for indirect prompt injection if the logs contain malicious instructions from previously visited websites or processed files.\n
Ingestion points: Reads from ~/.claude/projects/*.jsonl files via the cc-session script and jq recipes.\n
Boundary markers: None identified; the instructions do not specify markers to distinguish untrusted log content from analysis instructions.\n
Capability inventory: Uses Bash, Read, Grep, and Glob tools. The skill also instructs the agent to suggest modifications to the user's permissions.allow configuration based on usage history, which could lead to reduced security oversight.\n
Sanitization: No validation or sanitization of log content is performed before processing.\n- [COMMAND_EXECUTION]: The scripts/cc-session script executes various bash, jq, grep, and rg commands. It dynamically constructs jq filters and passes user-provided search patterns to shell commands, which is a standard but noteworthy capability for a log analysis tool.

introspect