The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/apply-optimizations.sh performs automated patching of JavaScript files within the user's extension directory (~/.antigravity/extensions).
Evidence: The script uses sed -i '' to modify code within index.js, scripts.js, and wait-state.js to change MAX_TIMEOUT and other constants.
Context: While intended to optimize performance for the vendor's sidecar extension, modifying executable code at runtime is a risky pattern that can lead to application instability or unintended behavior.
[PROMPT_INJECTION]: The analyze_code function in ag_client.py ingests untrusted data from a user-provided project directory and includes it in prompts sent to the AI, creating a surface for indirect prompt injection.
Ingestion points: ag_client.py (via the project_path argument in the analyze_code function).
Boundary markers: Absent; the project path and the resulting analysis request are interpolated via direct string concatenation without delimiters or instructions to ignore embedded commands.
Capability inventory: Subprocess execution of local binaries (ag_client.py, scripts/antigravity_cli.py), and file system write/copy operations (ag_client.py, scripts/antigravity_mcp_client.py).
Sanitization: Absent; the content of the analyzed files is not validated or sanitized before being sent to the AI service.
[COMMAND_EXECUTION]: Multiple components execute a local binary via the command line.
Evidence: Scripts such as ag_client.py, scripts/antigravity_cli.py, and scripts/generate_image.py use asyncio.create_subprocess_exec to run the binary located at ~/.config/antigravity-mcp/bin/antigravity-mcp-server.
Context: This is the primary mechanism for the skill's functionality but relies on the existence and integrity of a local binary on the user's system.

antigravity-mcp