Agent Skills

Agent skills are installable instructions that help agent clients use oper8r tools correctly. They are separate from MCP tool registration. MCP exposes the tools; skills teach the client when to use them, what scope applies, and what not to store.

Included Skills

npm Installer

For filesystem-based clients:

npx @oper8r-io/agent-skills install all --client all

Install for Codex:

npx @oper8r-io/agent-skills install all --client codex --scope user

Install Cursor project rules from your project root:

npx @oper8r-io/agent-skills install all --client cursor --scope project

List available skills:

npx @oper8r-io/agent-skills list

Hosted Artifacts

The oper8r app exposes a skill manifest and downloadable artifacts:

/api/mcp/skills
/api/mcp/skills/:slug/zip
/api/mcp/skills/:slug/raw
/api/mcp/skills/:slug/cursor-rule
/api/mcp/skills/claude-marketplace

ChatGPT and Claude web-style installs usually use uploaded ZIP artifacts. Cursor can use generated .mdc rules. Claude Code can install from the oper8r marketplace endpoint.

Claude Code

/plugin marketplace add https://YOUR_OPER8R_HOST/api/mcp/skills/claude-marketplace
/plugin install oper8r-agent-skills@oper8r-io

Skill Scope

The public oper8r skill is intentionally narrow. It should guide the client to:

• Search the oper8r knowledge base before answering.
• Cite source material when generating answers.
• Use oper8r_answer for cited answers, including RFP, security, compliance, and questionnaire responses.
• Use { "items": [...] } for every tool request and read { "results": [...] } from every response.
• Avoid storing private user memory, credentials, unsupported summaries, guesses, or unrelated personal data.

Recommended Rollout

1. Enable the relevant oper8r MCP tools for the organization.
2. Install the matching skill into the client.
3. Test search first.
4. Test a cited answer with a human reviewer.