Coding Harness Integrations
These guides cover technical and operational integration between Rapid OS v3 and popular AI-assisted coding environments and agents.
1. Core principle: declared capabilities versus actual capabilities
In Rapid OS, a harness profile (HarnessProfile) models the standard capabilities associated with a tool in a declarative and conservative way.
Product truth rule: A built-in profile (
builtin:cursor,builtin:codex, and others) does not grant a model magical permissions or guarantee that the local environment can perform an operation. A coding harness can only perform actions the developer has allowed in the terminal, sandbox, or local configuration.
2. Canonical capability matrix
Rapid OS evaluates exactly 11 canonical capability identifiers:
| Canonical ID | Capability | Typical built-in status |
|---|---|---|
context.consume | Harness can read and consume the Context Bundle | supported |
repository.read | Read repository files and directories | supported |
workspace.current | Operate in the active workspace | supported |
repository.write | Modify files on disk | unknown (requires local permissions) |
shell.execute | Run commands in the terminal or shell | unknown (subject to sandbox and approval) |
tests.execute | Run automated test suites | unknown (depends on the local test runner) |
git.inspect | Inspect branches, diffs, and Git status | unknown |
git.modify | Create commits or branches, or modify Git state | unknown |
workspace.isolated | Work in isolated branches or Git worktrees | unknown |
mcp.invoke | Invoke tools through the MCP protocol | unknown |
subagents.delegate | Delegate work to subagents | unknown |
3. Main workflow: shortcuts
The main workflow for any of these harnesses uses /rapid-<mode> shortcuts ($rapid-<mode> in Codex). Run rapid init --shortcuts or rapid shortcuts install --harness <id> to generate them for your tool. The agent guides the run (rapid next → rapid start → rapid check → rapid finish) and stops when a person needs to act. The v3 commands (rapid spec, rapid run, rapid evidence, rapid eval, and others) remain available in Advanced mode for CI, debugging, and auditing. See the shortcuts guide for harness locations and invocation details.
4. Harness guides
Each guide starts with the main shortcut workflow and retains the advanced mode. Choose your tool:
- OpenAI Codex / CLI: Headless execution workflows and governed scripts.
- Anthropic Claude Code: Interactive CLI integration and terminal permissions.
- Cursor IDE: Composer integration,
.cursorrulesproject rules, and context injection. - Visual Studio Code: GitHub Copilot Chat, extensions, and integrated terminals.
- Google Antigravity: Multi-agent orchestration, autonomous subagents, and advanced sandboxing.