Skip to content
Daran Docs

Incident Handoffs and Autonomous PRs

An Incident Handoff is the main artifact Daran produces. It is designed for the moment when a responder or downstream AI coding agent needs enough context to continue an investigation without redoing the first pass manually. In Phase 1, a Handoff can also point to a Git branch and GitHub PR when Daran can prepare an application-side fix inside a scoped repository.

What A Handoff Answers

Section titled “What A Handoff Answers”

A useful Handoff should answer:

  • What is happening right now?
  • Which services, environments, or users appear affected?
  • Which alerts, logs, metrics, traces, deploys, pull requests, or configuration changes are relevant?
  • What are the leading hypotheses?
  • Which evidence supports each hypothesis?
  • Which evidence weakens a hypothesis or remains uncertain?
  • What should be checked next?
  • Are there action candidates worth considering, such as rollback, configuration change, or a code fix?
  • If Daran created a PR, which branch, files, validation results, and rollback or revert steps should be reviewed?

How Daran Builds One

Section titled “How Daran Builds One”

The first release follows a narrow investigation path:

Signal ingestion
-> Operational Context assembly
-> Evidence selection
-> Hypothesis generation
-> Handoff generation
-> Action Planning
-> Application-side Agent Worktree
-> Git branch
-> GitHub PR
-> Dashboard, CLI, or workflow output

Daran keeps deterministic correlation separate from AI-generated reasoning. Evidence is collected with source links and raw references first; hypotheses and Handoff text are generated from that inspectable context.

Release Boundaries

Section titled “Release Boundaries”

Phase 1 is focused on investigation, review, and evidence-backed application PRs. It includes signal ingestion, context assembly, evidence selection, hypothesis generation, Handoff generation, Action Planning, scoped branch / PR creation, feedback, and output surfaces such as dashboard, CLI, GitHub, Slack, and issue-tracker integrations.

Phase 1 does not include on-call scheduling, status pages, incident room management, postmortem management, or unapproved production, cloud, database, Kubernetes, or Terraform operations.