This package helps ChatGPT, Claude, Claude Code, Codex, Cursor, and similar clients understand how to use oper8r tools safely and consistently. It does not contain credentials, does not connect to oper8r by itself, and does not grant tool access. Users still need an oper8r MCP/API connection and org-granted tools in oper8r.

Included Skills

| Skill | Purpose | | --- | --- | | oper8r-workflows-search-explorer | Guides LLM clients when using oper8r workflow, search, graph exploration, and integration-status tools for grounded account work. | | oper8r-employee-graph-curator | Guides LLM clients when curating the shared organization-internal employee graph: entities, facts, external IDs, temporals, invalidations, and reconciliation. |

Install

Install everything for filesystem-based clients:

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

List available skills:

npx @oper8r-io/agent-skills list

Install one skill:

npx @oper8r-io/agent-skills install oper8r-workflows-search-explorer --client codex --scope user

Client-Specific Setup

Codex

Installs skills into ~/.codex/skills:

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

Restart or reload Codex after installing.

Cursor

Installs Cursor project rules into .cursor/rules from your project root:

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

Cursor rules are project-scoped by design. Commit them only if you want everyone working in that repo to receive the same oper8r guidance.

Claude Desktop / Filesystem Skills

Installs skills into ~/.claude/skills:

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

For a project-local install:

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

Claude Code Plugin

The oper8r app exposes the skills marketplace endpoint; you can add it in Claude Code:

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

ChatGPT And Claude Web

ChatGPT and Claude web-style skill installs typically use uploaded ZIP artifacts rather than this npm CLI. Use the oper8r MCP integration panel to download each skill ZIP, then upload it in the client skill settings.

The hosted artifacts are also available from the oper8r app:

• https://oper8r.io/api/mcp/skills
• Zip files
  • https://oper8r.io/api/mcp/skills/oper8r-workflows-search-explorer/zip
  • https://oper8r.io/api/mcp/skills/oper8r-employee-graph-curator/zip
• Raw files
  • https://oper8r.io/api/mcp/skills/oper8r-workflows-search-explorer/raw
  • https://oper8r.io/api/mcp/skills/oper8r-employee-graph-curator/raw
• Cursor rules
  • https://oper8r.io/api/mcp/skills/oper8r-workflows-search-explorer/cursor-rule
  • https://oper8r.io/api/mcp/skills/oper8r-employee-graph-curator/cursor-rule

Scope Rules

The graph curator skill is intentionally strict.

Use it for:

• Source-backed, durable, organization-internal account and employee knowledge.
• Shared entities such as accounts, companies, contacts, opportunities, products, teams, employees, territories, vendors, and related records.
• External IDs from trusted systems such as Salesforce, HubSpot, Slack, Google Drive, Gmail, Linear, GitHub, or other approved integrations.
• Temporal corrections when newer evidence updates a mutable fact.
• Reconciliation before overwriting or invalidating conflicting facts.

Do not use it for:

• User-private graphs or individual personal memory.
• Customer-facing or public graphs.
• Passwords, API keys, auth tokens, secrets, or raw credential metadata.
• Personal preferences, private notes, transient chat instructions, guesses, unsupported summaries, or speculative claims.
• Customer profile or account facts that belong in a separate customer/public graph rather than the organization-internal employee graph.

Every graph-curation tool response should include a scope contract equivalent to:

{
  "graph": "employee",
  "accessScope": "employee",
  "privateGraphAllowed": false,
  "customerGraphAllowed": false
}

External ID Examples

When creating or updating entities, preserve external IDs as stable provider identifiers. Prefer structured IDs over plain text references.

{
  "name": "Acme Corp",
  "type": "account",
  "externalIds": [
    {
      "provider": "salesforce",
      "id": "0018c00002abcDEF",
      "type": "Account"
    },
    {
      "provider": "hubspot",
      "id": "982734",
      "type": "company"
    }
  ]
}

For people:

{
  "name": "Jane Smith",
  "type": "person",
  "externalIds": [
    {
      "provider": "slack",
      "id": "U06ABC123",
      "type": "user"
    },
    {
      "provider": "salesforce",
      "id": "0038c00001xyzABC",
      "type": "Contact"
    }
  ]
}

Do not store OAuth tokens, refresh tokens, API keys, private email contents without business relevance, or raw provider payloads as external IDs.

CLI Reference

oper8r-agent-skills list
oper8r-agent-skills install <skill|all> [--client claude|codex|cursor|all] [--scope user|project] [--cwd path]

Defaults:

• --client all
• --scope user
• --cwd is the current working directory

Install locations:

| Client | User Scope | Project Scope | | --- | --- | --- | | Claude | ~/.claude/skills/<skill> | .claude/skills/<skill> | | Codex | ~/.codex/skills/<skill> | .codex/skills/<skill> | | Cursor | .cursor/rules/<skill>.mdc | .cursor/rules/<skill>.mdc |