Skip to content
Workflows

Miner workflow

A deterministic four-step loop. Each step is pure metadata; each output is structured JSON your agent can consume.

The mirrored loop

Each step on the left is what the contributor runs; the matching step on the right is what the maintainer sees in the repo at the same point.

Miner
What the contributor does
Maintainer
What the repo owner sees
  1. Plan
    Pull a decision pack — lane context, repo targets to pursue or avoid, freshness, and ranked next actions.
    gittensory-mcp agent plan --login your-login --json
    How scoreability works
    Plan
    Nothing visible in the repo. The plan step is private MCP context for the contributor; no public comments or labels are emitted.Privacy boundary
  2. Analyze
    Metadata-only branch analysis on the current branch — refs, changed-file metadata, labels, linked issues, commit messages, validation summaries.
    gittensory-mcp analyze-branch --login your-login --json
    Branch analysis reference
    Analyze
    Still silent in the repo. Branch analysis runs locally against the API; no source is uploaded and no check runs are created.What we don't upload
  3. Preflight
    Combine branch analysis with account/queue context to surface branch blockers, account blockers, and maintainer-fit notes.
    gittensory-mcp preflight --login your-login --json
    Common preflight blockers
    Preflight
    On confirmed-miner PRs you can later request the same view with @gittensory preflight — the response is sanitized for the PR thread.All @gittensory commands
  4. Packet
    Produce a public-safe PR packet — a description that reads cleanly to maintainers, with no private scoring or risk language leaking out.
    gittensory-mcp agent packet --json
    Set up your MCP client
    Packet
    At most one sticky sanitized comment and one configured label per confirmed-miner PR. Private scoring, reward, and risk language never appear in the thread.Install the GitHub App
Cleanup first. When the preflight reports queue pressure or unsquashed commits, prefer cleaning open work over opening more — risk-adjusted priority is part of the score model.