The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read iteration artifacts and use their content to update project-level documentation, which presents a surface for instructions embedded in those artifacts to influence the agent.
Ingestion points: The skill reads data from .agents/flow/it_{iteration}_PRD.json, .agents/flow/it_{iteration}_refactor-report.md, .agents/flow/it_{iteration}_progress.json, and .agents/state.json (SKILL.md).
Boundary markers: None. The instructions do not specify the use of delimiters or warnings to ignore instructions found within the iteration artifacts.
Capability inventory: The skill has the capability to read project files and write to core documentation files including .agents/PROJECT_CONTEXT.md, ROADMAP.md, AGENTS.md, and README.md (SKILL.md).
Sanitization: None. There is no evidence of content validation or sanitization before the information from artifacts is integrated into the documentation updates.
Human Review: The skill includes a mandatory "Plan and present changes" step, requiring user confirmation before any modifications are applied, which serves as a significant control against automated injection attacks.

approve-prototype