The Agent Skills Directory

[SAFE]: The skill consists of instructional markdown and configuration metadata. It contains no executable code, shell scripts, or external dependencies that would pose a security risk. All referenced scripts and domains belong to the skill author (itsoltech) and are used for documentation and context purposes.
[SAFE]: Guidance within the reference files explicitly promotes security best practices, such as verifying model artifact checksums, pinning dependencies, and warning against the execution of untrusted model formats like pickles.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and review external data, such as ML pull requests, manifests, and registry records. This represents a potential indirect injection surface if an attacker provides malicious instructions within those artifacts.
Ingestion points: Analyzes model contracts, schemas, and PR content as described in SKILL.md and references/01-serving-contracts-parity.md.
Boundary markers: The instructions in references/01-serving-contracts-parity.md recommend explicit validation at the serving boundary and handling of malformed inputs.
Capability inventory: The skill uses tools like itsol-current-tech-context to fetch documentation but does not include instructions for writing to the filesystem or performing unauthorized network operations.
Sanitization: references/04-security-privacy-fairness.md provides explicit instructions for the agent to validate tool-call arguments, escape model output, and avoid executing generated code without a sandbox.

ml-serving-mlops-review