Skip to main content
Specific is designed to be operated by coding agents. Your agent writes the code and defines the infrastructure to run it, with the same specific.hcl and the same CLI. This page covers how to connect an agent and how agents consume Specific’s documentation.

​
Connect your agent

Inside your project, run: 
specific init
This writes Specific usage instructions into the right place for your agent, auto-detected from your project (or pass --agent explicitly):
AgentWhat’s written
Claude Code (claude)Instructions appended to CLAUDE.md, plus permissions for specific docs and specific check in .claude/settings.local.json.
Codex (codex)Instructions appended to AGENTS.md.
Cursor (cursor)A rule file at .cursor/rules/specific.mdc.
Other (other)Instructions appended to AGENTS.md.
specific init --agent claude cursor
It also adds .specific and specific.local to .gitignore. From then on, the agent knows to define infrastructure in specific.hcl, validate with specific check, run everything with specific dev, and consult specific docs when it needs details.

​
Docs built into the CLI

The CLI ships its own LLM-optimized documentation, so agents never depend on stale training data or web access: 
specific docs                      # the documentation index
specific docs services             # a specific topic
specific docs integrations/nextjs  # framework-specific guidance
specific docs migrations/supabase  # migrating from other platforms
The docs are nested and cross-linked; agents navigate them by path. They ship with the CLI binary, so they always match the installed version.

​
Agent skill

For agents that support skills, a Specific skill is available at docs.specific.dev/skill.md. It contains the product summary, install command, and essential commands: enough for an agent to bootstrap itself, install the CLI, and pull deeper documentation via specific docs.

​
This documentation site

Every page on docs.specific.dev is also available to agents and LLM tools:
  • /llms.txt - an index of all pages for LLM consumption, and /llms-full.txt for the full content.
  • Append .md to any page URL for clean markdown.
  • The contextual menu on every page offers copy-as-markdown and open-in-ChatGPT/Claude shortcuts.

​
Agents operating remotely

Specific runs fully headless, so agents in remote environments like OpenClaw or Hermes can install the CLI, build, and deploy without a human at the machine. See the OpenClaw setup instructions for a ready-to-use prompt that configures such an environment. Two pieces make this work:
  • Tunnels: an agent on a machine you can’t browse from can still show you the running app. specific dev --tunnel exposes public services at https://<subdomain>.tunnel.spcf.app.
  • Device-code login: logging in without a browser works via the device-code flow. specific deploy and specific login print an authorization link and code, and specific login --device-code <code> completes it.

​
What stays the human’s job

  • Confirming the CLI installation.
  • DNS changes for custom domains - the CLI prints the records, but you configure them at your registrar.
  • Production management in the dashboard: scaling, secrets rotation, domain removal, billing.