define-object

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via magick and node using arguments derived from external documentation files.
  • Evidence: The 'Normalise notes' and 'Recompose assembly' sections provide command templates like node <fabrication-cycle>/scripts/normalise-runtime-asset.mjs --target-height <Inventory Appearance Size> and magick ... -splice <left padding>x0. These placeholders are sourced from docs/game/assets.md and docs/game/world.md without explicit sanitization, potentially allowing for command injection if the project files are malicious.
  • [INDIRECT_PROMPT_INJECTION]: The skill reads and processes multiple documentation files that influence its behavior and output, creating an ingestion point for untrusted instructions.
  • Ingestion points: The skill reads docs/game/world.md, docs/game/assets.md, docs/game/story.md, docs/game/puzzles.md, and docs/game/screenplay.md.
  • Boundary markers: Absent; there are no instructions to use delimiters or to ignore instructions embedded within these files.
  • Capability inventory: The agent has the ability to write new TypeScript/JavaScript files (ObjectDefinition), generate artwork, and execute shell commands.
  • Sanitization: Absent; the skill does not specify how to validate or escape the content read from these documents before using it in fabrication steps or code generation.
  • [DYNAMIC_EXECUTION]: The skill dynamically generates and updates ObjectDefinition files in the project's source directory based on external data.
  • Evidence: Step 4 ('Author the definition') directs the agent to 'Write or update the ObjectDefinition under src/objects//'. While part of the intended workflow, this involves creating executable code from data processed from potentially untrusted project files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 05:01 PM