The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong behavioral directives intended to override the agent's default conversational style (e.g., "Drop high-token conversational behavior", "Acknowledge and Act", "Never use generic terminal commands"). These instructions reconfigure the agent's operational logic based on specific triggers like "planning mode" or "fast path". While functional for coding tasks, they represent a significant override of standard interactive behavior.
[COMMAND_EXECUTION]: The skill permits the use of terminal commands for specific lifecycle tasks such as running builds (npm run build), executing tests (flutter test), and installing packages. It explicitly restricts the use of these tools for file modification or reading, which is a defensive measure to ensure more controlled file interactions via built-in tools.
[DATA_EXPOSURE]: The skill focuses on local codebase exploration and modification. It instructs the agent to retrieve file contexts and create project artifacts like implementation_plan.md or system_architecture.md. No evidence of targeting sensitive system paths (~/.ssh, ~/.aws) or credentials was found.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface.
Ingestion points: The agent is instructed to read and search codebase files to identify line numbers and understand dependencies (SKILL.md sections 1.1, 2.A, 2.B.1, 2.C.1).
Boundary markers: No specific boundary markers or instructions to ignore embedded commands within the processed files are present.
Capability inventory: The agent has the capability to modify files (edit/replace tools), create new files (artifacts/save states), and execute shell commands (for builds and tests).
Sanitization: The skill does not mention validation or sanitization of data retrieved from the codebase before processing or using it in tool calls.

antigravity-protocol