Sortie

Sortie turns issue tracker tickets into autonomous coding agent sessions. Engineers manage work at the ticket level. Agents handle implementation. Single binary, zero dependencies, SQLite persistence.

Sortie assumes your coding agent already produces useful results when you run it manually. It handles scheduling, retry, isolation, and persistence around that agent — it does not improve the agent's output.

The Problem

Coding agents can handle routine engineering tasks — bug fixes, dependency updates, test coverage, feature work — when they have good system prompts, appropriate tool permissions, and have been tested on representative issues. But running validated agents at scale requires infrastructure that doesn't exist yet: isolated workspaces, retry logic, state reconciliation, tracker integration, cost tracking. Teams build this ad-hoc, poorly, and differently each time.

Sortie is that infrastructure.

How It Works

Define your WORKFLOW.md in a single file alongside the target repository:

---
tracker:
  kind: github
  api_key: $GITHUB_TOKEN
  project: acme/billing-api
  query_filter: "label:agent-ready"
  active_states: [todo, in-progress]
  in_progress_state: in-progress
  terminal_states: [done, wontfix]

agent:
  kind: claude-code
  max_turns: 10
  max_sessions: 3
  max_concurrent_agents: 4

workspace:
  root: ~/workspace/billing-api

hooks:
  after_create: |
    git clone --depth 1 [email protected]:acme/billing-api.git .
  before_run: |
    git fetch origin main
    git checkout -B "sortie/$SORTIE_ISSUE_IDENTIFIER" origin/main
  after_run: |
    git add -A && git diff --cached --quiet || \
      git commit -m "sortie($SORTIE_ISSUE_IDENTIFIER): automated changes"
    git push origin "sortie/$SORTIE_ISSUE_IDENTIFIER"
---

You are a senior Go engineer working on the billing-api service.

## {{ .issue.identifier }}: {{ .issue.title }}

{{ .issue.description }}

{{ if .run.is_continuation }}
Resuming work — review workspace state before continuing.
{{ end }}
{{ if .attempt }}
Retry attempt {{ .attempt }}. Check the workspace for partial work.
{{ end }}

Set GITHUB_TOKEN to a fine-grained PAT with Issues: Read and write permission scoped to the target repository. States are mapped to GitHub labels — create labels matching your active_states and terminal_states before starting Sortie. The query_filter scopes polling to issues with a specific label so Sortie only picks up work you explicitly mark as ready. See the GitHub adapter reference for full configuration details.

Sortie watches this file, polls for matching issues, creates an isolated workspace for each, and launches Claude Code with the rendered prompt. It handles the rest: stall detection, timeout enforcement, retries with backoff, state reconciliation with the tracker, and workspace cleanup when issues reach terminal states. Changes to the workflow are applied without restart.

See examples/WORKFLOW.md for a complete example with all hooks, continuation guidance, and blocker handling.

Copilot CLI

To use GitHub Copilot CLI instead of Claude Code, swap the agent block:

agent:
  kind: copilot-cli
  max_turns: 5
  max_concurrent_agents: 4

copilot-cli:
  model: gpt-5.3

The adapter requires a GitHub Copilot subscription and a valid GitHub token. See the Copilot CLI adapter reference for full configuration details and examples/WORKFLOW.copilot.md for a complete workflow.

Architecture

Sortie is a single Go binary. It uses SQLite for persistent state (retry queues, session metadata, run history) and communicates with coding agents over stdio. The orchestrator is the single authority for all scheduling decisions; there is no external job queue or distributed coordination. For full architectural details, see docs/architecture.md.

Issue trackers and coding agents are integrated through adapter interfaces. Adding support for a new tracker or agent is an additive change: implement the interface in a new package.

Supported trackers: GitHub Issues and Jira. Supported agents: Claude Code and Copilot CLI. See docs/decisions/ for detailed rationale on technology choices.

Documentation

Full configuration reference, CLI usage, and getting started guide: docs.sortie-ai.com

Prior Art

Sortie's architecture is informed by OpenAI Symphony, a spec-first orchestration framework with an Elixir reference implementation. Sortie diverges in language (Go for deployment simplicity), persistence (SQLite instead of in-memory state), extensibility (pluggable adapters for any tracker or agent, not hardcoded to Linear and Codex), and completion signaling (orchestrator-managed handoff transitions instead of relying solely on agent-initiated tracker writes).

Why "Sortie"

A sortie is a military and aviation term for a single mission executed autonomously. The metaphor is precise: the orchestrator dispatches agents on missions (issues), each with an isolated workspace, a defined objective, and an expected return. The name is short, two syllables, pronounceable across languages, and does not conflict with existing projects in this domain.

Roadmap

See our project board for current status and priorities.

License

Apache License 2.0