Claude
Reason through the product shape first.
npm create clawletter-connect@latest -- --agent claude
One-click package
Install the bridge, copy the prompt, launch the agent.
Setup creates the files. Run starts the selected agent.
setup
npm create clawletter-connect@latest -- --agent codex
run
npm create clawletter-connect@latest -- --agent codex --run
local repo
npm run connect:codex
ClawLetter CLI
ClawLetter CLI connects every coding agent.
Choose Claude, Codex, OpenClaw, or a private runner. Get the same handoff every time.
Claude
Codex
OpenClaw
Custom stdio
intent
agent
engine
app
app-conversion.session
live handoff
$ claw convert --agent codex --from repo --to app
intent "turn onboarding ops into a field console"
grep routes, auth, billing, inbox schema
read product notes + connected source graph
patch /apps/field-console + preview script
return app URL, changed files, reviewer notes
Input
repo + prompt
Runner
Codex CLI
Output
working app
Direct connectors
Four agent paths, one conversion contract.
Codex
Patch the repo and return the review trail.
npm create clawletter-connect@latest -- --agent codex
OpenClaw
Bind conversion requests to a durable relay.
npm create clawletter-connect@latest -- --agent openclaw
Custom CLI
Wrap any internal runner over stdio.
CLAWLETTER_AGENT_RUNNER=./agent-runner npm create clawletter-connect@latest -- --agent custom
Agent modes
The bridge changes shape around the runner.
Spec intelligence
Ask Claude for the product cut, then pass the exact handoff to the engine.
claw convert --agent claude --handoff product-spec
Useful when the raw ask needs product judgment before a repo patch exists.
Copy prompts
Prompts that make the bridge behave like a skill.
App conversion
Convert this repository into a focused app experience. Inspect the repo first, preserve existing patterns, build the smallest working surface, verify it, and return changed files plus the next runnable command.
Design pass
Improve this app as a real product surface. Prioritize information hierarchy, responsive spacing, useful controls, no overlapping text, and a visual system that makes the workflow feel connected.
Review trail
Return a conversion packet: preview URL or run command, changed files, tests or checks run, risks, and one follow-up prompt that another coding agent can continue from.
Conversion protocol
One command becomes a reproducible app handoff.
The CLI is intentionally small: it captures intent, selects a runner, packages context, and receives the app result without forcing every agent into the same runtime.
Capture the ask
Repo, docs, source graph, screenshots, or free-form instructions become a compact intent packet.
claw intent write --from repo --from notes
Bind the runner
Claude, Codex, OpenClaw, or a private executable gets a scoped token and a stable invocation.
claw agents add codex --command "codex"
Run conversion
The engine asks for a plan, patch, prototype, runbook, or deployable app depending on the target.
claw convert --agent codex --target app
Return proof
Every session returns changed files, commands run, test notes, and a human-readable review trail.
claw sessions inspect latest --format review
What comes back
Not a chat transcript. A working conversion packet.
PreviewLocal or hosted URL with the converted app experience.
PatchScoped file changes or a generated project the user can review.
RunbookInstall, test, deploy, and handoff commands for the next agent.
TraceInputs, runner, commands, and notes kept outside fragile chat history.
OpenClaw optional, CLI universal