Skip to content

Latest commit

 

History

History
79 lines (58 loc) · 3.4 KB

File metadata and controls

79 lines (58 loc) · 3.4 KB

Getting started

From nothing to a closed loop on your own repository. Every step below is one command, each command tells you what it decided, and nothing here asks you to assemble anything — a step that leaves you holding instructions is a defect in the step, not homework.

1. Bring Sync up

git clone https://github.com/stroland02/Sync
cd Sync
npm start

npm start is the whole install. It checks for updates and fast-forwards a clean checkout, picks the route that works on your machine — the container where Docker answers, a user-space install where it cannot — sets up Python, Postgres, the schema, and the console's dependencies, then serves the console. Rerun the same command after a reboot; it restarts what stopped and touches nothing else. (npx @stroland02/sync-up delivers the same launcher from the registry and hands you the clone command above.)

When it finishes, the console is at the address it prints, and the Setup screen shows each prerequisite as measured — not assumed.

2. Put your own code on the screen

uv run sync index --repo <your-repo-remote>

INDEX reads your repository's call sites into the API Dependency Graph, offline, and the console shows your vendors and your call sites. An empty result is reported as exactly that — the console never fills the gap with somebody else's data.

3. Pick what to watch

The integration catalog lists every vendor Sync serves today and every one it recognizes. A supported vendor is watched by a registered adapter; a recognized one is named honestly as unwatched, with its page saying what adding it takes — often one line in generated-vendors.yaml.

uv run sync run --vendor stripe \
    --from-version <pinned> --to-version <target> --repo <your-repo-remote>

This joins the vendor's changes against your call sites. Findings land in the console, each carrying the provenance rung it arrived at — static, resolved, or observed — because a finding that cannot be attributed cannot be fixed.

4. Rehearse the loop before you spend anything

uv run sync rehearse

Rehearsal executes the full pipeline — locate, patch, verify — against a local, zero-remote fixture repository. No network, no vendor account, no pull request. It is how you watch the machinery work before pointing it at anything real.

5. Close the loop for real

The remediation loop needs two credentials, and it needs them from you because Sync holds no secrets of its own:

  • An authenticated gh CLI — pull requests are opened as you, onto a branch you configured.
  • An Anthropic API credential — the patch agent that writes fixes runs on it. The agent is contained: it works inside a clone, its tools are gated, and nothing reaches a pull request without passing tsc and then your repository's own CI.

Then a run that produces a mechanically-safe finding carries it through patch and verification to a pull request in your repository, and the console shows every step of the reasoning — including the steps that gave up, because abandoned runs are data.

What Sync will tell you it cannot tell you

The readiness meter on the console reports axes it has no evidence for as CANNOT TELL, not as green. That is deliberate and permanent: an unmeasured axis is absence, not a passing grade, and a tool that paints absence green is the tool this one exists to replace.