The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source text and user voice samples for translation. It includes a robust 'Verification Gate' (Stage 4) and a 'Critical Review' stage (Stage 5) to ensure integrity and naturalness, serving as a mitigation against simple adversarial content in the input text.
Ingestion points: SKILL.md (Expected inputs: source text, voice samples).
Boundary markers: SKILL.md (Stage 4: Verification Gate
Placeholder and structure parity checks).
Capability inventory: SKILL.md (Actions table: READ, WRITE, VALIDATE tools; file editing mentioned for translation application).
Sanitization: SKILL.md (Mechanical checks, translation rubric, and anti-AI pattern taxonomy).
[COMMAND_EXECUTION]: The skill instructs the agent to use rg (ripgrep) to search for context in sibling locale files. This is a standard development operation used appropriately within the codebase scope.
[DATA_EXPOSURE]: The skill accesses project-specific configuration files (.agents/oma-config.yaml) and locale resources (resources/lang/*.md). These are internal application resources necessary for the skill's primary function and do not involve sensitive system credentials.

oma-translation