The Agent Skills Directory

[NO_CODE]: The analyzed skill consists entirely of markdown documentation, schema definitions, and policy guidelines. There are no executable scripts, binaries, or active code components included in the repository.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for ingesting untrusted data and provides explicit defensive instructions to mitigate potential injection attacks.
Ingestion points: Data ingestion occurs through the processing of scraped page text, reviews, comments, and emails as defined in SKILL.md.
Boundary markers: The Unified Skill Contract requires evidence to be labeled by provenance (measured, user-provided, etc.) to distinguish between trusted and untrusted context.
Capability inventory: The skill does not contain instructions for command execution, file writing, or network exfiltration. Its primary capabilities are analytical and generative.
Sanitization: SKILL.md and references/skill-contract.md contain explicit instructions to treat scraped material as untrusted input and to ignore any embedded instructions found within that data.

message-system-architect