The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.
TypeScript 3d-modelingagent-frameworkagent-harnessai-agentscadclaude-codecodexcoding-agentsdeveloper-toolsesp32game-developmentmcp-servermulti-agentopen-hardwareopencodeopenharnesspcbroboticsterminaltmux
/** The portable harness ABI is files + a shell, not an engine's plugin format.
* Each adapter names the project instruction files its engine actually reads, in precedence order.
* Skills and executable tools are exposed through the session's context, so native skill/MCP
* support is an optional optimization rather than a condition for running a harness.
* Contracts and primary sources: store/spec/portability.md.
*/
import { harnessAdapters } from '../engines/launches.js'
import type { AgentEngine, ProcessEngine } from '../engines/types.js'
export interface HarnessAdapter {
instructionFiles: readonly string[]
/** A session-local instruction flag, when the engine provides one. */
contextArgs?: (contextFile: string) => string[]
/** The session's variables as flags, for an engine whose commands do not inherit its own environment. */
envArgs?: (env: Record<string, string>) => string[]
}
export const HARNESS_ADAPTERS = {
...harnessAdapters,
cursor: { instructionFiles: ['AGENTS.md'] },
opencode: { instructionFiles: ['AGENTS.md', 'CLAUDE.md'] },
pi: { instructionFiles: ['AGENTS.md', 'CLAUDE.md'], contextArgs: (file: string) => ['--append-system-prompt', file] },
omp: { instructionFiles: ['AGENTS.md', 'CLAUDE.md'], contextArgs: (file: string) => ['--append-system-prompt', file] },
hermes: { instructionFiles: ['.hermes.md', 'AGENTS.md', 'CLAUDE.md', '.cursorrules'] },
commandcode: { instructionFiles: ['AGENTS.md'] },
devin: { instructionFiles: ['AGENTS.md'] },
muse: { instructionFiles: ['AGENTS.md', 'CLAUDE.md'] },
amp: { instructionFiles: ['AGENTS.md'] },
kilo: { instructionFiles: ['AGENTS.md'] },
grok: { instructionFiles: ['AGENTS.md'] },
agy: { instructionFiles: ['AGENTS.md', 'GEMINI.md'] },
copilot: { instructionFiles: ['AGENTS.md'] },
} satisfies Record<ProcessEngine, HarnessAdapter>
/**
* Every project file Harness itself may write into: any adapter's instruction files (the session
* bootstrap, `runtime.ts`) and the files the saved-API notes go in (`lib/apiInstructions.ts`). Asked of
* the workspace's SCM before those writes (`prepareScmWrite`), for an SCM that holds tracked files
* read-only. Not every name exists in a given workspace.
*/
export const PROJECT_INSTRUCTION_FILES: readonly string[] = [...new Set([
...Object.values(HARNESS_ADAPTERS).flatMap(adapter => adapter.instructionFiles), 'AGENTS.md', 'CLAUDE.md', 'GEMINI.md',
])]
/** A terminal has no instruction loader or agent tools. It remains a Coding-only choice. */
export function harnessAdapter(engine: AgentEngine): HarnessAdapter {
if (engine === 'terminal') throw new Error('A harness needs an agent engine; Terminal cannot run harness instructions.')
return HARNESS_ADAPTERS[engine]
}
export const HARNESS_BOOTSTRAP = `<!-- harness:runtime v1 -->
## Harness session context
If HARNESS_CONTEXT_FILE is set in your environment, read that file before working:
\`\`\`sh
if [ -n "$HARNESS_CONTEXT_FILE" ]; then cat "$HARNESS_CONTEXT_FILE"; fi
\`\`\`
It contains the instructions and skill index for this session's selected harness.
If the variable is absent, this is an ordinary coding session with no selected harness.
<!-- /harness:runtime -->
`