The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from several user-controlled or external markdown files to generate an execution contract.
Ingestion points: Reads from proposal.md, specs/, design.md, and tasks.md within the local workspace.
Boundary markers: The skill lacks explicit instructions to treat the content of these files as untrusted or to use delimiters when interpolating them into the final execution-contract.md.
Capability inventory: Includes filesystem writes (generating execution-contract.md) and state management via the ssf command-line tool (state set, state transition, state init, state rebuild).
Sanitization: No sanitization, validation, or escaping of the input document content is performed before processing or summarization.

contract-builder