junior briefs

Use junior briefs to inspect and tune Brief generation without changing production data. A Brief records the intent, outcome, decisions, open decisions, durable facts, and evidence links for one Conversation.

Pull snapshots

Create a personal API token in the Junior dashboard. Then download one or more Conversations:

pnpm exec junior briefs pull <conversation-id> \
  --base-url https://junior.example.com \
  --token "$JUNIOR_API_TOKEN" \
  --out ./brief-snapshots

You can omit --token when JUNIOR_API_TOKEN is set. The command downloads the detail report and follows all older event pages. The saved JSON does not contain the API token.

You can pull any public Conversation. You can pull a private Conversation only when your token can view its content. Expired and redacted transcripts cannot generate a Brief.

Generate Briefs

Run the default prompt and the app’s configured default model against a snapshot:

pnpm exec junior briefs run ./brief-snapshots/<conversation-id>.snapshot.json \
  --out ./brief-output

The command writes <conversation-id>.brief.json and <conversation-id>.brief.md. The Markdown file ends with an evidence check. It shows link counts, dropped model URLs, and model cost.

Without --model, the command resolves the model for the current default profile when the run starts. Use --model <id> or --prompt <file> to compare generation settings. Add --turn-by-turn to generate one version for each completed turn in the snapshot.

Brief generation needs Vercel OIDC or AI_GATEWAY_API_KEY. Snapshot download needs only the dashboard token.