The Agent Skills Directory

[PROMPT_INJECTION]: The 'Auto Mode Detection' section contains explicit instructions to bypass safety guidelines and human oversight. It directs the agent to 'Skip all confirmation prompts' and 'merge passing branches without pausing' when specific keywords such as 'yolo mode', 'bypass permissions', or 'auto-accept' are detected. This mirrors 'Developer Mode' or 'DAN' style injections intended to override operational constraints.
[PRIVILEGE_ESCALATION]: By instructing the agent to bypass human-in-the-loop confirmations for critical operations (e.g., merging code and pushing to remote branches), the skill effectively escalates the agent's autonomous capabilities beyond safe defaults, reducing the user's ability to review and intercept potentially harmful modifications.
[COMMAND_EXECUTION]: The skill utilizes powerful shell commands, including git merge, git push, and git worktree add. When these tools are used in conjunction with the 'Auto Mode' bypass, the risk of unauthorized or automated destructive changes to source code repositories is significantly increased.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (source code) through an automated refactoring pipeline, creating a vulnerability surface where malicious code could influence the agent's behavior.
Ingestion points: Phase 1 ('Scan & Detect') reads target files specified by the user or found in the workspace.
Boundary markers: None identified; there are no instructions to the agent to treat file contents as untrusted or to ignore embedded instructions within comments or strings.
Capability inventory: The skill possesses file-write capabilities, shell command execution (git), and the ability to spawn subagents with delegated prompts.
Sanitization: No sanitization, escaping, or filtering of the ingested file content is described before the data is passed to subagents or used to drive the refactoring logic.

refactor