The Agent Skills Directory

[SAFE]: The skill implements a parallel, blind execution model via a built-in workflow mechanism. This ensures that individual personas (seats) operate in isolation without cross-seat input, which mitigates the risk of coordinated prompt injection attacks.
[SAFE]: A 'Security force-floor' is defined, which automatically activates a security-focused persona whenever the input contains signals related to authentication, secrets, payments, or trust boundaries. This is a defensive design pattern.
[SAFE]: No evidence of unauthorized data exfiltration, network operations, or hardcoded credentials was found. The mention of local file paths (e.g., ~/.claude/branching-flow.md) refers to the maintainer's external manual process and does not involve automated file access by the skill.
[SAFE]: The 'disable-model-invocation: true' configuration in the frontmatter follows the principle of least privilege by preventing other skills from invoking this workflow unexpectedly.

council