The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute multiple bundled Python scripts (e.g., scripts/policy.py, scripts/route_capabilities.py, scripts/validate_review_artifact.py) to manage project configuration, perform routing logic, and validate artifacts. These scripts perform filesystem operations within the .translation/ directory to track approvals and project state.
[DYNAMIC_EXECUTION]: The skill implements a sophisticated extension loading system. As described in SKILL.md (Workflow step 12) and scripts/route_capabilities.py, the agent is instructed to base64-decode file content provided in a JSON routing payload, verify it against SHA-256 digests, and then load the content as a new skill artifact. This dynamic loading of instruction-bearing content into the agent context is an inherent execution risk, albeit mitigated by the described verification protocol.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and translate external content, which it explicitly identifies as untrusted data that may contain instruction-like text or directives.
Ingestion points: Translation source artifacts, metadata, and content from third-party specialists processed during the translate and refine phases.
Boundary markers: The skill includes a dedicated Source trust boundary section in SKILL.md that instructs the agent to treat embedded directives as data and maintain structural/linguistic contracts without following them.
Capability inventory: The skill can execute local scripts via subprocess, write files to the project directory, and dynamically load external skill modules.
Sanitization: The review workflow uses 'sanitized challenge input' and requires deterministic validation of review artifacts to ensure conceptual and structural consistency.

translating-products