The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements a proactive 'Scrub sensitive content' step in Phase 3. It specifically targets and redacts patterns for common credentials (e.g., sk-, ghp_, bearer tokens) and home directory paths using regex-based identification before reporting. This prevents the accidental leakage of secrets that might be present in git history or environment artifacts.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests external data (git logs, plan files, and session artifacts) which could potentially contain malicious instructions, the risk is neutralized by platform-level tool restrictions. The skill is limited to read-only tools (Read, Grep, Glob), and the instructions explicitly forbid executing remediation steps, ensuring the agent remains a passive diagnostic observer.
[COMMAND_EXECUTION]: The skill utilizes git commands and filesystem searches (find) for evidence gathering. These are used strictly for data retrieval (git log, git status, git worktree list) and do not involve executing user-supplied strings as shell commands. The environment configuration (allowed-tools) further restricts these capabilities to safe, read-only operations.

forensics