The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute local scripts and binaries, such as tools/package.mjs for archive assembly and cargo run for character pack validation. These tools are intended to run within the user's workspace or the skill's bundle.
[DYNAMIC_EXECUTION]: The workflow involves the execution of local JavaScript (.mjs), TypeScript (.ts), and Rust source code during the conversion and packaging steps. The instructions explicitly state that the packaging tool should reject unsafe paths and not download remote media.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external manifest.json files and legacy character pack data. This creates an attack surface for indirect prompt injection if malicious instructions are embedded in the character metadata, although the skill includes specific instructions to validate these files against a schema and native validator.
Ingestion points: Processes manifest.json, legacy 2.0 character manifests, and WebP artwork files from the local filesystem.
Boundary markers: The skill relies on external schemas and native validation logic to enforce data integrity.
Capability inventory: File system read access (manifests/art), file system write access (ZIP packaging), and local command execution (Node.js/Cargo).
Sanitization: The packaging script is described as having logic to reject unsafe paths and duplicate keys.

dramagic-character-release