The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a protocol where subagents ingest and execute instructions from a markdown-based contract, creating a potential surface for indirect prompt injection if the source data is not trusted.
Ingestion points: The DELEGATION CONTRACT template in SKILL.md defines fields like Scope, Input Interface, and Constraints that are populated at runtime and processed by subagents.
Boundary markers: The template includes a mandatory Out of Scope section which acts as a logical boundary to restrict subagent actions.
Capability inventory: The contract guidelines recommend executing verification commands using npx (e.g., tsc, eslint, vitest).
Sanitization: No explicit sanitization or escaping mechanisms are defined for the content interpolated into the contract fields.
[COMMAND_EXECUTION]: The skill's verification protocol suggests the execution of common development CLI tools.
Evidence: Mentions the use of npx tsc --noEmit, npx eslint src/, and npx vitest run within the Verification Criteria section to ensure code quality and correctness.

agent-delegation-contract