The Agent Skills Directory

[NO_CODE]: The skill is composed entirely of markdown instructions and YAML frontmatter. It does not include any scripts, executable binaries, or references to external packages.
[SAFE]: No patterns associated with prompt injection, data exfiltration, obfuscation, or persistence were found. The skill defines a logical procedure for manual or agentic analysis of benchmark validity.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data such as benchmark specifications and evaluation records.
Ingestion points: Processed via the evaluation_records and benchmark_specification inputs defined in SKILL.md.
Boundary markers: None present; the instructions do not specify delimiters for the input data.
Capability inventory: None. The skill does not authorize or instruct the use of any tools, shell commands, or network access.
Sanitization: None.
Risk Assessment: Although the skill processes untrusted external data, the complete absence of executable capabilities or tool access within the skill instructions eliminates the risk of privilege escalation or malicious command execution.

assess-construct-validity