The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill captures user responses during multi-round interviews and synthesizes this data into a markdown artifact (explore.md). This process introduces a surface for indirect prompt injection, where malicious instructions provided by a user could be stored and later executed or interpreted by other agents or tools that process the generated PRD or project state.
Ingestion points: User chat responses collected during the interactive interview rounds defined in the workflow (SKILL.md).
Boundary markers: The skill instructions reference an external contract (../lcs-shared/contract.md) for artifact writing safety, although the specific delimiting techniques or guardrails are not implemented within the provided file content.
Capability inventory: The skill performs filesystem write operations to create artifacts in the .lcs/work-items/ directory and updates the .lcs/state.md file.
Sanitization: The instructions explicitly require the agent to "interpret faithfully" and "accept" custom answers without forcing them into predefined options, which indicates that user-provided text is passed through to the artifacts without rigorous filtering or sanitization.

lcs-explore