The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in the skill instructions or scripts.
[COMMAND_EXECUTION]: The script scripts/transcript-miner.py executes the ripgrep (rg) utility to search transcript files. The call uses subprocess.run with a list of arguments, which prevents shell injection by avoiding shell interpretation of inputs.
[DATA_EXFILTRATION]: While the skill processes sensitive local transcript data from .codex and .claude directories, it implements multiple layers of protection. The instructions mandate redaction of PII and secrets, and the helper script transcript-miner.py includes comprehensive regex patterns to automatically mask emails, API keys, tokens, and cryptographic addresses before they reach the agent context.
[PROMPT_INJECTION]: The skill instructions are focused on task analysis and do not contain directives to bypass safety guardrails, override system prompts, or ignore established rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes historical transcript data which could potentially contain malicious instructions from previous sessions. This risk is mitigated by the skill's design, which emphasizes summarization over direct quoting and utilizes a helper script for automated data sanitization and redaction.

agents-introspection