The Agent Skills Directory

[SAFE]: The skill implements an advanced 'preflight canary' protocol that generates and verifies a high-entropy value in the conversation history to ensure that spawned subagents ('forks') have correctly inherited the session context before performing audits.
[SAFE]: User-provided configurations via ${user_config} placeholders are handled securely; the skill includes explicit logic to detect and ignore uninitialized template tokens to prevent them from being misinterpreted as skill names or commands.
[SAFE]: The skill explicitly documents the shared tool pool risks of subagent 'forks' and recommends a detection-based safety strategy, such as comparing the working tree state before and after execution, to identify and discard unauthorized writes.
[SAFE]: All external documentation references and links target official, well-known technical documentation (e.g., code.claude.com), posing no risk of malicious external downloads.

sweep-all