Grep platform skills

Teach your coding agent to use Grep: discover existing agents, turn repetitive work into reusable agents, run them on new inputs, and improve repeated execution with workflows and code. Research, documents, apps, slides, and spreadsheets remain part of the library.

Install the published release

npx grep-research-skills@latest

grep-research-skills is the npm package this repository publishes; its version is the version in package.json once a release ships. Check what is published with npm view grep-research-skills version. A second npm package, grepai-skills, was published from an earlier revision of this repository; it is not the package package.json names, so do not expect releases there. Neither 0.2.0 build contains the platform workflows (grep-platform, grep-agentify, grep-agents, grep-optimize); they are in the 0.3.0 source. Node 18+ is required.

Install this revision from source

git clone https://github.com/Parcha-ai/grepai-skills.git ~/.grep-research-skills
cd ~/.grep-research-skills
./setup

This installs the checked-out source, including unreleased changes. For an unreleased branch, check out that branch before setup. This revision adds a grep-skills executable alias in the same package.

The installer links skills for Claude Code and detects Codex, Cursor, and OpenClaw installations. It preserves real user-owned skill directories. Connect Grep through your client's MCP setup at https://api.grep.ai/api/v2/mcp, or set GREP_API_KEY securely for REST. Do not commit API keys in project configuration.

Start with /grep-platform, or ask: “Agentify the company research procedure we have been repeating.”

Main workflows

Repetition detection uses context available to the coding agent. It is not an always-on watcher. Cross-session reuse comes from a project .grep/agents.json reference, not hidden transcript access. A suggestion does not authorize a paid batch, schedule, purchase, or public sharing. Existing user authorization remains valid.

Current platform contract

Platform reference records the verified contract and auth differences. Fetch live capabilities with:

node scripts/grep-api.js capabilities
node scripts/grep-api.js agents
node scripts/grep-api.js agent-build --file=build.json
node scripts/grep-api.js agent-build-status BUILD_ID
node scripts/grep-api.js research "Research Acme" --expert-id=AGENT_ID --idempotency-key=UNIQUE_RUN_KEY
node scripts/grep-api.js result RUN_ID --no-wait

build.json contains domain, optional context (up to 5,000 characters), and depth (standard or deep). Build success may register a draft: inspect its lifecycle and activate it before testing. See skills/grep-agents/SKILL.md.

Production verified on 2026-09-28 exposes /api/v2/run; creation/configuration still uses /api/v2/experts/*. MCP retains research_* / expert_* names. A public optimize endpoint is not currently advertised. Workflow generation produces a candidate; it is not proof of lower cost. The library checks deployment capabilities and never claims a proposed optimization API is live.

API keys or v2 resource-scoped OAuth tokens support the platform. GREP_ACCESS_TOKEN supplies such an OAuth token. Legacy email-OTP sessions retain v1 research compatibility but cannot masquerade as v2 OAuth credentials. GREP_API_BASE selects the API host; GREP_UI_BASE can select the matching UI host. Preserve IDs after timeouts and resume existing work.

Cowork / Claude.ai

npm ci
npm run build:cowork

Upload dist/grep-research-skills-v0.3.0.zip using the client's skill-upload interface. The archive retains the research entrypoint for compatibility but routes the whole platform. References are generated from skills/, and scripts/resources are bundled. It needs permitted network access to the configured Grep API and auth provider.

Development

npm test
npm run sync:cowork
npm run build:cowork

The Node test suite uses mocked HTTP and temporary installation roots; it creates no remote agents or paid runs. Source skills are canonical. Do not edit generated Cowork references directly. Version numbers in npm/plugin/discovery metadata move together; publishing remains a separate release step. scripts/update-check.js checks only by default; use --update to explicitly update from npm.

{
  "name": "grep-research-skills",
  "owner": {
    "name": "Parcha Labs",
    "url": "https://grep.ai"
  },
  "metadata": {
    "description": "Grep platform skills for reusable agents, workflows, and research."
  },
  "plugins": [
    {
      "name": "grep-research-skills",
      "source": "./",
      "version": "0.3.0",
      "description": "Grep platform skills for reusable agents, workflows, and research.",
      "author": {
        "name": "Parcha Labs",
        "url": "https://grep.ai"
      },
      "homepage": "https://grep.ai",
      "repository": "https://github.com/Parcha-ai/grepai-skills",
      "license": "MIT",
      "keywords": [
        "research",
        "deep-research",
        "grep",
        "ai-agent"
      ]
    }
  ]
}