The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external game assets which serve as ingestion points for untrusted data that could potentially influence agent behavior.
Ingestion points: Game dialogue, scripts, and resource metadata files processed during translation as described in references/engine-assets.md and references/image-text.md.
Boundary markers: The instructions mandate strict folder separation between original assets and translation work in references/project-structure.md, and emphasize immutable source identity.
Capability inventory: The skill uses scripts/inventory_assets.py for file discovery and scripts/payload_manifest.py for integrity checks, granting it read/write access to the project workspace.
Sanitization: While the scripts use strict hash-based validation, there is no explicit sanitization of natural language content extracted from game files before it is processed by the agent.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts for asset management and manifest creation.
Evidence: SKILL.md instructs the agent to run python3 scripts/inventory_assets.py to index game files and python3 scripts/payload_manifest.py to generate and verify file-integrity manifests.
Safety Features: The scripts/payload_manifest.py utility includes a safe_path function designed to prevent directory traversal by validating that paths do not leave the intended payload directory.
[EXTERNAL_DOWNLOADS]: The documentation references established third-party tools for game dumping and decryption.
Evidence: references/private-inputs.md provides links to the GitHub repositories for nxdumptool (DarkMatterCore) and hactool (SciresM), which are standard tools within the game modification community.

switch-localize