The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The hook script hooks/rca-package.sh is programmed to automatically copy .claude/settings.json and .claude/settings.local.json from the project's hidden configuration directory into the public-facing RCA results folder. These files often contain sensitive agent configuration parameters, including API keys or local environment secrets.
[DATA_EXFILTRATION]: The hooks/rca-package.sh script actively searches for and copies the session's interaction transcript (claude_session.jsonl) into the final report directory. This behavior exposes the full history of the conversation, potentially including sensitive data, proprietary code, or project-specific details discussed with the agent.
[COMMAND_EXECUTION]: Multiple automation hooks, including hooks/rca-defect-coverage.sh and hooks/rca-depth-check.sh, execute complex Python logic via shell-invoked heredocs. This pattern of embedding and executing dynamic scripts within shell hooks can be a vector for code injection and complicates security auditing of the execution flow.
[DATA_EXFILTRATION]: The parallelization strategy described in references/parallelization.md instructs the agent to launch multiple subagents and aggregate their findings. This multi-agent architecture, combined with the automated packaging of all session artifacts, significantly increases the amount of diagnostic data and history persisted to the local file system in unencrypted form.

tao-analyze-changenet-rca