The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data such as logs, database exports, and transcripts, which represents an attack surface for instructions embedded within that data.
Ingestion points: Files are read using read_file as specified in the quality bar section of SKILL.md.
Boundary markers: The instructions mandate a 'not_observed' status until a scan is completed and recommend using line limits for initial inspection.
Capability inventory: The skill utilizes read_file, search_files, and execute_code capabilities to process data and run scans.
Sanitization: The skill requires text normalization (NFKC) and explicitly instructs that detector scripts must print only hit counts, never raw values, to prevent sensitive data from entering the session logs.
[DYNAMIC_EXECUTION]: The skill generates and executes local scripts to perform residual scans for PII shapes using regular expressions.
Evidence: Instructions in references/data-share-method.md describe running a 'detector script' through execute_code to count regex matches for identifiers like SSNs and mobile numbers.
[COMMAND_EXECUTION]: The skill references specific CLI tools for operational logging and scanning.
Evidence: SKILL.md contains examples of omh runtime record for status tracking and references using rg -P (ripgrep) for PCRE-compatible pattern matching.

omh-data-share-review