The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by parsing external files (PNG, WebP, JSON) and extracting fields such as system_prompt and personality to define agent behavior.
Ingestion points: Processes user-provided files like character.png, character.json, and world-info.json from the workspace.
Boundary markers: The skill lacks boundary markers or sanitization, explicitly instructing the agent to 'Use character's system_prompt as behavior rules'.
Capability inventory: The agent can execute shell commands, perform file I/O, and install external dependencies.
Sanitization: No content validation or sanitization is performed on the extracted strings before they are injected into the agent's instruction context.
[COMMAND_EXECUTION]: The skill utilizes several shell commands to locate, copy, and execute its parsing logic.
Evidence: Instructions include using cp, find, and node to manage and run the parse-character-card.js script. The skill also provides 'Fallback Code' for manual script creation.
[EXTERNAL_DOWNLOADS]: The skill requires downloading external dependencies at runtime via the Node Package Manager.
Evidence: Instructs the agent to run npm install to acquire the png-chunks-extract and png-chunk-text packages.

story-roleplay