This dashboard is your agent's work and control view. Give assignments in your AI workspace (for example Claude Cowork or Hermes); the agent uses SharksAPI A2A/MCP tools and updates the cards.
1 · View the right period
Date range filters every data-backed card, including statistics and posts.
Compare uses the previous period of equal length. Tables show that period in a separate section underneath.
Strategy and task cards are not date statistics, so they remain visible.
2 · Give the agent a concrete brief
Review my Copywriter dashboard. Use the last 30 days of GSC and WordPress data, identify three content opportunities, avoid duplicating existing articles, and create the strongest article as a WordPress draft. Do not publish before my approval.
Include the goal, period, audience, constraints, and expected deliverable. Publishing, ad spend, and other external actions should go through approval.
3 · Install and use Skills
Ask the agent to read /llms.txt and fetch its recommended role skills from GET /api/v1/a2a/skills.
The agent saves each skill in its skill directory. For Claude: .claude/skills/<name>/SKILL.md. In other agents, use their skills/skill.md mechanism.
Before doing related work, the agent loads and follows the matching playbook — for example blog-writer, content-strategy, seo-audit, or command-center-agent.
Run check_skill_updates regularly and replace stale skills with current catalog versions.
Read https://sharksapi.ai/llms.txt. Fetch the SharksAPI skill catalog, install the skills recommended for my role, update stale skills, and load the appropriate skill before each task.
4 · Connections, memory, and approvals
If a required tool is unavailable, create the connection link and ask me to approve OAuth or access.
Store durable brand, budget, and workflow rules in project shared memory; never store passwords or tokens there.
Send publishable work through the draft/approval workflow and continue after my decision.
