@loopyc/cli

loopc — the Monkey D Loopy command-line factory for runnable, crash-resumable agent loops.

npm i -g @loopyc/cli

loopc quickstart                     # honest 100: prove + run + observe + inspect + vendor
loopc blueprints                    # starting points, one per loop pattern
loopc recipes                       # verified product workflows
loopc new repo-check --recipe repo-health-doctor
loopc new my-watch --blueprint poll-until
loopc validate my-watch.loop.yaml   # refuses unbounded / unreachable loops
loopc verify   my-watch.loop.yaml   # dry-run proof: bounded · deterministic · resume-stable
loopc score    my-watch.loop.yaml   # 0-100 scorecard (termination grounding included)
loopc compile  my-watch.loop.yaml --target all --out ./out
loopc run      my-watch.loop.yaml --inputs ./inputs.json --out ./run

When the verified path depends on effect output, pass deterministic data-only mocks with loopc verify spec.yaml --fixtures fixtures.json and loopc score spec.yaml --fixtures fixtures.json. No shell, HTTP, or model effect executes during verification.

No-install first run: npx --yes @loopyc/cli@latest quickstart.

The load-bearing rule: the compiler will not emit an unbounded loop. Termination is required, caps are mandatory, and the dry-run proves both before anything real executes. Compile targets: standalone (add --vendor for a zero-install bundle), babysitter, claude-code, claude-native, and n8n.

For Claude Code-native loops, loopc compile --target claude-native emits .claude/skills/<loop>/SKILL.md plus the embedded LoopSpec reference. Copy the generated .claude/ directory into a project and invoke the loop as /<loop> run; compile with --target all when you want the skill to delegate to a sibling standalone artifact for hard runtime guarantees.

Full command reference: docs/cli.md.

Jev authoring and iterative refinement (0.9.0)

These capabilities are available starting with 0.9.0. See availability and source setup and confirm that the installed CLI/MCP tool list includes the feature before using it.

| CLI | MCP | Purpose | |---|---|---| | recommend | recommend_workflow | Rank eligible catalog structures for a brief | | design | design_workflow | Create a validated, mock-verified scaffold from an explicit selection | | refine | refine_workflow | Compare current YAML and authored revisions with feedback/evidence; preserve history |

Jev requires TYPESAFE_API_KEY in the process environment and explicit provider selection (allowExternal: true in MCP). Env files are not loaded automatically. Refinement sends supplied eligible YAML, feedback and evidence to TypeSafe; keep secrets out. Offline performs local checks and retains current, without claiming semantic improvement. These authoring calls never activate a workflow or execute its real effects.

Read the workflow designer, agent guide, and complete agent context for exact contracts, setup, repeated rounds and limitations.