Skip to content

AI Discovery

Agents and generators should start from runtime metadata instead of scraping source files or examples:

timeout 120s nix run . -- metadata cl-weave/test --reporter json --output cl-weave-metadata.json

The metadata payload advertises CLI commands, typed options, finite choices, command-specific choices, environment variables, CI quality gates, public package exports, matchers, mutation operators, mop-architecture-assertions, capabilityMatrix, artifactSchemas, and distributionChannels. For runtime self-diagnostics, doctor --reporter json emits a structured doctor-report artifact without requiring an ASDF system argument. The report separates bundled cl-weave visibility, optional requested-system resolution, workspace .asd discovery, and output-target configuration so CI and agents can distinguish environment drift from an actually missing target system.

artifactSchemas is the contract for structured artifacts such as JSON run results, JSONL run events, JSON test plans, JSONL plan entries, doctor reports, and mutation reports. Each entry declares the artifact kind, producing commands, supported reporters, artifact-local schemaVersion, streaming mode, and field map, so agents can plan parsers and CI integrations without hard-coding reporter internals. Result artifacts intentionally advertise both run and watch, because watch --once emits the same machine-readable shape as a normal run. qualityGates exposes validation commands as argv vectors with explicit timeouts and expected artifacts, so agents can reproduce CI without scraping prose.

distributionChannels is the canonical install and run table for source checkout execution, local Nix packaging, and remote Nix packaging. Agents should prefer its installCommand and runCommand vectors over inferring entrypoints from surrounding prose examples. The maintainer-facing verification and scope boundary for those channels lives in Distribution Policy.

capabilityMatrix is the readiness table: each entry links a high-level feature to implemented status, representative public APIs, validation gates, and canonical documentation. The complete artifact and capability lists are intentionally discovered from the command output; documentation examples are illustrative.

See AI Contract for the full machine-readable normalization contract, and Doctor Report for the self-diagnostic artifact shape.