The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source text and locale files, which is an inherent surface for indirect prompt injection in translation tasks. 1. Ingestion points: Source text, existing locale files, and glossary files (SKILL.md, "Expected inputs"). 2. Boundary markers: The skill explicitly instructs the agent to "Preserve exact tokens first" (SKILL.md, "Structural Flow") and includes verification steps for protected syntax. 3. Capability inventory: The skill uses READ and WRITE primitives for local file access and codebase navigation (SKILL.md, "Logical Operations"). 4. Sanitization: Verification steps are required to ensure placeholders, code spans, and URLs remain unchanged.
[NO_CODE]: The skill consists entirely of instructional markdown and configuration templates; no executable scripts or binary files are included.
[SAFE]: The instructions are designed to maintain high fidelity and structural integrity, with no evidence of malicious intent, data exfiltration, or obfuscation.

oma-translation