The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled binary named ocean-watch-launcher (or ocean-watch_windows_amd64.exe on Windows) to perform its core functions. This binary handles all interactions with the Ocean Engine API, including project creation, account synchronization, and reporting. The launcher path is resolved dynamically based on the skill's installation directory.
[EXTERNAL_DOWNLOADS]: The skill specifies a dependency on the third-party Python package f2 (version 0.0.1.7) for specific metadata resolution tasks related to Douyin. The instructions include environment checks via a doctor command to verify the presence of this dependency rather than performing automated, unverified installations at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the Ocean Engine platform, such as video titles, creator metadata, and performance reports, to inform campaign strategies.
Ingestion points: External data enters through MCP tools such as search_marketing_creator_materials and report_marketing_plans (documented in workflow-reference.md).
Boundary markers: The skill explicitly mandates that all plan mutations default to dry-run/preview mode and requires explicit user confirmation before any online write operations are submitted.
Capability inventory: The skill possesses capabilities for creating and updating marketing projects, promotion units, and budget settings via the binary launcher.
Sanitization: Instructions require the agent to compare external candidate titles against product context and flags ambiguous matches for manual user confirmation.
[SAFE]: The skill communicates exclusively with well-known official Ocean Engine API domains (api.oceanengine.com and ad.oceanengine.com). It demonstrates high security awareness by explicitly forbidding the AI from printing or requesting OAuth secrets, access tokens, or cookies in the chat interface, instead utilizing secure OS-level credential storage (Keychain/DPAPI).

ads-plan-monitor