Agent skills

Choose a workflow

Use creator research when you already have a shortlist of public handles to compare. Use API integration when you are connecting your application or agent to SignalShelf. Both are original first-party instructions, downloadable without signing in.

Install in your agent

Download the chosen SKILL.md and place it in a folder named exactly like its frontmatter name, using your agent's documented skill location. For example, the creator skill folder is signalshelf-creator-research and contains SKILL.md. Supported agent locations and reload behavior vary, so follow your client's instructions.

If your agent accepts project instructions rather than skill folders, you can provide the Markdown as project guidance. This is a manual instruction workflow, not a promise of a native integration for every client.

Connect your instance

Set SIGNALSHELF_BASE_URL to your SignalShelf deployment and SIGNALSHELF_API_KEY using your agent's environment or secret manager. The production base URL is https://signalshelf.dev.

Create a key in the SignalShelf workspace. Never paste a real key into a skill, public repository, report, frontend bundle or shared prompt.

SIGNALSHELF_BASE_URL=https://signalshelf.dev
SIGNALSHELF_API_KEY=YOUR_SERVER_SIDE_KEY

Try a bounded task

Ask: Compare these three public handles against my campaign brief. Accept results up to 24 hours old. Show public source URLs, available metrics, fetch times and credits charged. Do not send outreach.

A fresh successful lookup costs one credit. Accepted cached results, samples and failures are free. A sample result is illustrative and must never enter a real creator ranking. Discovery, transcripts and recent-video performance analysis require their own supported endpoints.

Connect tools with MCP

If your client supports Streamable HTTP MCP, connect the authenticated /mcp endpoint. The skills can then guide how the agent uses the exposed tools. A downloadable skill alone does not create that connection.