The live Claude provider walkthrough. It runs the same two prompts as
../single_repo_packet/, which uses Codex,
so the pair demonstrates that a packet is provider-portable.
Requirements
The Claude CLI must be installed and authenticated:
npm install -g @anthropic-ai/claude-code
claude --version
Prompt Runner drives Claude over ASM's core lane, so claude_agent_sdk is not
a required dependency of your project.
What It Covers
provider: claudewith aclaude_optssection- prompt-local verification contracts
- generated checklist files
- packet-local runtime state in
.prompt_runner/ - git commits after verification passes
Setup
From the project root:
bash examples/claude_packet/setup.sh
That creates a local git repo at:
examples/claude_packet/workspaceInspect And Run
mix prompt_runner list examples/claude_packet
mix prompt_runner packet preflight examples/claude_packet
mix prompt_runner plan examples/claude_packet
mix prompt_runner run examples/claude_packet
mix prompt_runner status examples/claude_packet
Expected Outputs
After a successful run:
workspace/hello.txtworkspace/SUMMARY.md
The runtime directory contains:
.prompt_runner/state.json.prompt_runner/progress.log.prompt_runner/logs/
Switching Providers
The packet manifest is the only thing that changes between this example and the Codex one:
provider: "claude"
model: "haiku"Swap those two lines for provider: "codex" and a Codex model such as
gpt-5.6-luna to run the identical prompts on the other lane.
Cleanup
bash examples/claude_packet/cleanup.sh