The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides endpoints to harvest local authentication credentials and export sensitive tokens. Evidence includes POST /api/providers/agy-auth/apply-local (described as 'Auto-detect and import the local Antigravity CLI (agy) login from disk') and POST /api/providers/{id}/claude-auth/apply-local which target local configuration files for credential extraction. Export endpoints such as /api/providers/{id}/claude-auth/export allow retrieving plaintext authentication tokens into the agent's context.
[COMMAND_EXECUTION]: The skill triggers the execution of local system binaries via its endpoints. The description for POST /api/providers/validate states it 'Spawns a fixed ACP summarizer binary to validate credentials', a behavior also seen in the bulk and import endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill has significant attack surfaces for data poisoning through untrusted file ingestion. 1. Ingestion points: ZIP archives (/api/providers/agy-auth/zip-extract) and external token files (/api/providers/agy-auth/import). 2. Boundary markers: No explicit instruction-ignoring delimiters are defined. 3. Capability inventory: The skill possesses high-privilege capabilities including reading local secrets (e.g., agy-auth), performing network requests (e.g., connection tests), and spawning binaries (e.g., validate endpoint). 4. Sanitization: There is no indication of content validation or sanitization for the data extracted from ZIP files or imported credential files.

omni-providers