The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to load and follow instructions from a plan file, which serves as an entry point for external data. This creates a surface for indirect prompt injection if the plan file contains malicious instructions. The skill mitigates this through process-based controls. Ingestion points: The agent reads a plan file in the initial step of the workflow. Boundary markers: The skill does not define specific delimiters to separate plan instructions from system context. Capability inventory: The skill enables the agent to perform implementation tasks and run verification commands based on the plan's contents. Sanitization: The skill does not use automated sanitization; instead, it requires the agent to 'review critically' and consult a 'human partner' before proceeding with execution.
[SAFE]: The core logic of the skill focuses on safe execution practices, including batching work and stopping immediately if blockers or unclear instructions are encountered.

executing-plans