First use: from a demo to your own configuration

中文 · Documentation · Current release

1. Run the no-key example. Follow the customer-form project. Reproduce input loss, apply the supplied patch and verify: two original failures, two fixed passes, three regression passes, one source-verified experience. This needs no DSH or model call and is not an automatic-learning or benefit demonstration.
2. Choose a compatible artifact. Check release status and the exact matrix. This alpha.2 release supports exactly DSH 0.2.0-rc.2 and 0.2.1-rc.5; historical alpha.1 supports only the former. While public publication is unverified, begin with the standalone example; do not substitute alpha.1 or downgrade business data.
3. Bind read-only access first. Follow installation sections 1–2. Keep learning, execution and sharing disabled. Use the same profile for install and launch, then open the complete newly printed authentication URL. The plugin should show the configured repository and read-only mode; an empty experience list is expected. Keys belong in DSH, not plugin JSON.
4. Authorize learning separately. For a trusted Python repository, provide independent acceptance tests outside the repository, the exact Python executable and in-project regressions. Read the authorization guide. Only the owner can explicitly enable execution, background verification, sharing, automatic learning and a bounded extraction grant for an actually registered model. Do not copy a grant ID or reset usage by renaming it. Extraction call limits are not a currency budget; ordinary DSH dialogue can also incur charges. DSH can converse normally with plugin learning disabled.
5. Verify before drawing conclusions. Inspect capture, original/fixed/regression checks, extraction and independent review in learning progress. A failure, skip or unknown receipt retains its reason. A later reference record proves a version was offered, not adoption or task improvement. Those require separate acceptance and controlled evaluation.

Start with a small non-private repair. Before upgrading, read data operations; use the FAQ when blocked. Report only redacted versions, failed steps and expected versus actual behavior.

Alpha.2 provides setup and doctor; the following entry points work from the built source checkout used for the example (historical alpha.1 lacks these commands). No globally installed CLI is assumed:

node dist/src/cli.js setup --repo /absolute/project --output /private/config/reforge.json --python /absolute/python
node dist/src/cli.js doctor --config /private/config/reforge.json

setup refuses existing files and enables no permission or extraction grant; optional --data-dir selects data storage. doctor reports bounded categories/counts without private paths or config values, model calls, migrations or new databases. Host identity and model connection are explicitly untested warnings, not successful connection checks. Fix reported errors without deleting data. Configure credentials locally in DSH.