The Agent Skills Directory

[SAFE]: The skill implements strong defensive measures against indirect prompt injection. It explicitly instructs the agent multiple times (e.g., in the Preconditions and Subagent-Per-Task sections) to treat content read from external files like plan documents, design specifications, and source code as data only and to disregard any embedded instructions.
[SAFE]: Data safety protocols are integrated into the trace emission logic. The skill includes specific regular expressions to detect and redact various types of credentials, including AWS keys, GitHub tokens, and OpenAI/Stripe API keys, before they are recorded in YAML logs.
[SAFE]: The skill utilizes standard development tools (Git commands) and operates within a controlled local file system scope (docs/plans, docs/designs) for its orchestration logic. No suspicious network operations or privilege escalation attempts were detected.

executing-plans