The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external implementation plans.
Ingestion points: The skill instructions in SKILL.md explicitly direct the agent to read and follow an external plan file.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the plan content as data only or to ignore embedded instructions within that plan.
Capability inventory: The agent is instructed to 'Follow each step exactly', which could involve file operations, tool usage, or command execution defined within the untrusted plan file.
Sanitization: No validation or sanitization mechanisms are defined for the content of the plan files being processed.
[SAFE]: The skill follows standard operational procedures for task management and includes appropriate safety stops, such as requiring user consent before modifying main branches and instructing the agent to stop if verifications fail.

executing-plans