[deanq]

Description

Fixes the queue-based (/runsync) request examples across the READMEs so they work when copy-pasted into flash dev. Two independent bugs made them fail:

1. Missing input envelope (422). QB routes require {"input": {"<function-param-name>": ...}}. The docs showed the raw body. The inner key is the worker function's parameter name (input_data, payload, or prompt), so it varies per example.
2. Wrong route for multi-function modules (404). When a module defines more than one @Endpoint function, flash dev mounts /<module>/<function>/runsync; /<module>/runsync only exists for single-function modules. Affected: 03_mixed_workers (cpu_worker), 01_text_to_speech, 01_autoscaling (README curls, load-test examples, and load_test.py's DEFAULT_ENDPOINT).

Also wraps the placeholder payloads in CLI-REFERENCE.md, CONTRIBUTING.md, and docs/cli/commands.md, and load_test.py's DEFAULT_PAYLOAD (the load test previously 422'd on every request).

Load-balanced routes (@endpoint.post/get, e.g. /pipeline/classify) take the body directly and are unchanged. Single-function modules (01_hello_world, 02_cpu_worker, 01_network_volumes, 05_volume_warm_cache, mixed_workers gpu_worker) keep /<module>/runsync.

Linear: SLS-346

Type of Change

• Bug fix
• Documentation update

Testing

Run with flash dev. Runpod Flash is 1.17.0 unless noted.

Route	Old payload / path	New payload / path
02_cpu_worker /cpu_worker/runsync	422	200 COMPLETED
autoscaling /cpu_worker/cpu_scale_to_zero/runsync	422	200 COMPLETED
autoscaling old path /cpu_worker/runsync (1.17.0 and 1.20.0)	404	n/a

• Re-checked on runpod-flash 1.20.0 (latest): every /runsync route's OpenAPI request schema requires input, the unwrapped payload still 422s, and multi-function routes are still /<module>/<fn>/runsync.
• Checked every edited path against the route table flash dev prints at startup. All exist.
• GPU and text-to-speech routes were only checked against the route table. I didn't call them, to avoid provisioning GPUs.
• make quality-check passes.
• Endpoints created during testing were torn down, and the account API confirms they are deleted.

cd 04_scaling_performance/01_autoscaling
uv run flash dev
curl -X POST http://localhost:8888/cpu_worker/cpu_scale_to_zero/runsync \
  -H "Content-Type: application/json" \
  -d '{"input": {"payload": {"text": "Hello autoscaling"}}}'

🤖 Generated with Claude Code