Phottly CLI exposes 7 tools via STDIO with minimal schema documentation. Tool names lack action verbs (login, logout, credits, avatars are nouns or generic). Descriptions are present but brief (10-50 chars), below the 50-200 char LLM-optimized baseline. Only 2 of 7 tools (generate, status, save) have visible input schemas; the others (login, logout, credits, avatars) show no schema in source. Schemas that exist are incomplete: generate lacks enum constraints for 'mode' and 'resolution' despite documented valid values in code comments. No output schemas documented. Error handling is minimal, most tools exit(1) with generic messages rather than guiding recovery. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Composition is reasonable (each tool does one thing), but parameter naming is inconsistent (avatarAssetId vs avatar, generationMode vs mode).

• Rename tools with action verbs: login→authenticate_oauth, logout→clear_session, credits→get_credits, avatars→list_avatars, generate→create_generation, status→get_generation_status, save→save_generation.
• Expand descriptions to 50-200 chars. E.g. 'Authenticate via OAuth 2.1 + PKCE. Opens browser for consent. Saves credentials to ~/.phottly/credentials.json. Required before any other command.'
• Add input schemas for login, logout, credits, avatars (even if empty object {}). Declare enums for generate.mode (posts|selfies|ultrareal_selfies|ultrareal_fits|ultrareal_new|custom) and generate.resolution (1K|2K|4K).
• Document output schemas. E.g. generate returns {generationId: string, creditsUsed: number}; status returns {generationId: string, state: string, frames: array}; avatars returns {assets: [{id, name, createdAt}]}.
• Add error guidance. E.g. 'Not logged in - run: phottly login' should be returned as structured error {error: 'UNAUTHENTICATED', message: '...', recovery: 'Call authenticate_oauth first'} so LLMs can self-correct.