Intermesh CLI

inter is a local CLI for two shipped workflows:

• Generate an installable CLI or local stdio MCP package from an OpenAPI specification and reviewed product tasks.
• Evaluate a CLI or MCP interface from one authored eval.json and retain its evidence locally.

Install

npm install --global @malmhq/intermesh-cli
inter --help

Requires Node.js 22.14 or newer and npm 11 or newer. Intermesh does not require an account or hosted service.

Generate

Create a named generation with one OpenAPI file, then review its product tasks.

inter generate setup --name my-api --openapi ./openapi.json --target cli
# Review .intermesh/generations/my-api/generation.json
inter generate run --name my-api

Configuration flags like --model, --credential-env, and --target belong to generate setup, not generate run — generate run reads them from the saved generation.json and reports unknown options if you pass them again.

Generation captures the reviewed file and OpenAPI, validates every requested task, then builds and verifies the package in the same invocation. Unsupported tasks stop before building and identify what must be changed.

Generation requires an explicit provider/model selection and its named credential environment variable. OpenAPI and task inputs are sent to that selected provider for analysis and generation. Each workflow publishes its extracted package and installable tarball under .intermesh/generations/<name>/outputs/<run-id>/. Run summaries and diagnostics remain in private machine-local Intermesh storage.

Evaluate

An evaluation lives in the workspace:

.intermesh/evaluations/my-evaluation/
└── eval.json

Run inter eval setup my-evaluation once to install the project Skills if needed; if eval.json is missing, the command reports its expected location. Use the inter-eval-setup Skill to create or repair eval.json. It defines the candidate, interface, scenarios, clients, environment-variable names and run settings. It never contains secret values.

inter eval setup my-evaluation
inter eval run my-evaluation
inter eval results my-evaluation

Every finalized run writes a self-contained report.html, versioned results.json and redacted public traces under its private ~/.intermesh/ run directory. To export the same public artifacts immediately after a completed run:

inter eval run my-evaluation --yes --export ./runs

To export a saved run later, select its exact ID:

inter eval results my-evaluation --run <run-id> --export ./runs

The export creates ./runs/<run-id>/ and refuses to overwrite an existing run folder. Review the export before publishing; reports can contain task prompts, findings, agent-reported feedback and redacted interaction traces.

inter eval run asks for confirmation. After explicit authorization, use --yes for automation. The workflow requires Docker, its Harbor/Python runtime, the selected model credential variable, and the dedicated target credential declared by the plan. Never put credential values in flags, plans, or reports.

Today, setup validates configuration, while candidate preparation, execution and result reading are being connected to the eval module. The commands explain that boundary clearly and do not fall back to the retired plan-based evaluator.

Automation

Use --json for scripts and coding agents. JSON mode never prompts: provide required choices and use --yes only after the exact plan and inputs have been reviewed.

Use --dry-run --json before setup or generation mutations to preview their local effects. For eval, use inter --dry-run eval setup <name> to validate the configuration or inter --dry-run eval run <name> to preview run scope.

Safety

Intermesh reads .env only from the selected workflow folder; exported environment values take precedence. Keep .env out of Git and limit its local permissions. Generation runs with the operating-system permissions of the user who starts it, so use only trusted inputs. Evaluate with scoped target credentials.

Command reference

inter generate setup --help
inter eval setup --help
inter status --help
inter generate --help
inter eval --help