The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill processes session logs that may contain untrusted data or adversarial instructions from previous turns or external tool outputs.
Ingestion points: The skill reads JSONL files from ~/.claude/projects/ (SKILL.md).
Boundary markers: Absent. There are no instructions or delimiters defined to prevent the agent from following instructions embedded within the logs during analysis.
Capability inventory: The skill uses Bash, Grep, and Read tools to search, filter, and display the content of these logs.
Sanitization: Absent. The logs are processed using raw shell commands (grep, sed) and the results are presented directly to the agent context.
[DATA_EXFILTRATION]: Sensitive Data Exposure. The skill is specifically designed to access and analyze ~/.claude/projects/, which contains a complete history of the user's interactions with the AI, including source code, environment details, and potentially sensitive data exchanged during sessions. While no external exfiltration was detected, the access to this directory represents a high-privilege read operation.

session-log-analyzer