The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/generate_image.py contains a hardcoded OAuth client secret.
Evidence: ANTIGRAVITY_CLIENT_SECRET = "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" in scripts/generate_image.py.
[DATA_EXFILTRATION]: The skill scans the local filesystem for sensitive authentication files belonging to the 'OpenClaw' and 'OpenCode' applications to harvest OAuth refresh tokens.
Evidence: find_all_antigravity_credentials function in scripts/generate_image.py attempts to read from:
~/.openclaw/agents/main/agent/auth-profiles.json
~/.openclaw/auth-profiles.json
~/.openclaw/credentials/google-antigravity.json
~/.config/opencode/antigravity-accounts.json
[EXTERNAL_DOWNLOADS]: The skill communicates with official Google OAuth and CloudCode API endpoints to refresh tokens and generate image content.
Evidence: Requests are made to oauth2.googleapis.com and cloudcode-pa.googleapis.com.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it interpolates user-provided prompts directly into requests sent to a remote image generation model without sanitization.
Ingestion points: User-provided --prompt argument in scripts/generate_image.py (line 332).
Boundary markers: None present in the API payload construction.
Capability inventory: Network access via requests and file system writes via pathlib (line 392).
Sanitization: No input validation or escaping is performed on the prompt string.

nano-banana-antigravity