The Agent Skills Directory

Indirect Prompt Injection Surface: The skill defines structured, machine-readable diagnostic output (JSON) explicitly intended for AI agents to branch on. This creates a multi-step pipeline where the tool's output influences downstream agent behavior. If the input triggering a diagnostic (such as a developer-authored template or CLI input) contains instructions disguised as data, they could potentially influence the agent parsing the resulting error report. (1) Ingestion points: CLI output in JSON format as described in SKILL.md Section 6. (2) Boundary markers: The skill advocates for stable, machine-readable identifiers (codes) and structured JSON objects to separate machine logic from human-readable descriptions. (3) Capability inventory: The documentation itself has no execution logic, but it provides guidance for a framework operating in FFI, WASM, and CLI contexts. (4) Sanitization: The conventions mandate plain-text output free of ANSI escape codes to prevent breaking machine parsers.

diagnostics