The Agent Skills Directory

[SAFE]: A comprehensive analysis of the skill instructions and configuration revealed no evidence of prompt injection, data exfiltration, obfuscation, or unauthorized command execution. The skill operates within a controlled environment using local file paths and internal routing protocols. The use of internal vendor prefixes (afa-*) is consistent throughout the document.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface as it ingests raw user requests and writes data to internal memory files. This surface is considered safe as it adheres to the project's structured output and interaction protocols.
Ingestion points: The user_request field passed from the Hub component in the context contract.
Boundary markers: Output is constrained by the _system/output-format.md specification and structured YAML completion blocks.
Capability inventory: The skill performs routing to specialized worker skills and writes state to ./brand-brain/{file}.md.
Sanitization: Not explicitly implemented within the skill logic, relying instead on system-level protocols and the referenced preamble for initialization and rule enforcement.

afa-organic