agentleFS
DROOdotFOODROOdotFOO/agent-skills/skills/raxol-symphony/SKILL.md

raxol-symphony

Skill1 starsChanged 2 months agoMITView on GitHub

Symphony: the Raxol tracker-driven coding-agent orchestrator (raxol_symphony, pre-alpha). Turns tracker issues into autonomous agent runs in isolated workspaces, surfaces evidence (CI/PR/asciinema), and supports human-in-the-loop paused runs across six surfaces. TRIGGER when: mix.exs lists :raxol_symphony; code imports Raxol.Symphony.*; there is a WORKFLOW.md driving an orchestrator; user asks about tracker-driven agent orchestration, Symphony runners (RaxolAgent/Codex), evidence capture, or paused-run resume with Raxol. DO NOT TRIGGER when: building a single Raxol agent or the core framework (use raxol skill); agent payments / ACP job sessions (use raxol-payments skill); generic multi-agent system design without Symphony (use agent-designer skill); the Claude Agent SDK (use claude-api).

  • Reads credentials
Files
SKILL.md

Raxol Symphony Skill

raxol_symphony (v0.1, pre-alpha) is an OTP port of OpenAI's Symphony: an orchestrator that turns tracker work into autonomous coding-agent runs. Each issue gets an isolated workspace, runs an agent until a workflow-defined handoff state, and surfaces evidence (CI, PR comments, asciinema walkthrough) so engineers manage outcomes, not prompts.

Config is a WORKFLOW.md (YAML front-matter + a Liquid prompt template), hot-reloaded on change. Pre-alpha: APIs may shift.

What You Get

  • Orchestrator loop: poll tracker -> isolate workspace -> run runner -> collect evidence
  • Runners: RaxolAgent (primary), Codex (JSON-RPC parity), AgentSession, Review, Noop
  • Trackers: Linear (GraphQL), GitHub (state labels), Memory (tests)
  • WORKFLOW.md hot-reload + a workflow graph adapter (saga-style)
  • Evidence framework: CI, PR comments, complexity (cloc), asciinema recordings
  • Six surfaces + a first-class paused-run substrate (pause/resume across surfaces)

Quickstart

{:ok, _pid} = Raxol.Symphony.start_link(workflow_path: "WORKFLOW.md")
<!-- WORKFLOW.md -->
---
tracker:
  kind: linear
  project_slug: demo
  api_key: $LINEAR_API_KEY
  active_states: ["Todo", "In Progress"]
  terminal_states: ["Done", "Cancelled"]
runner:
  kind: raxol_agent
  agent:
    backend: anthropic
    model: claude-sonnet-4-6
    max_turns: 20
---
You are working on issue {{ issue.identifier }}: {{ issue.title }}.

See also

  • raxol -- core agent/TUI framework Symphony runs agents on
  • raxol-payments -- ACP job sessions used by the paused-run resume flow
  • agent-designer -- general multi-agent orchestration patterns (non-Symphony)
  • prd-to-plan / qa -- feeding a tracker with well-formed issues

Reading Guide

TaskFile
Orchestrator, trackers, runners, configorchestrator.md
WORKFLOW.md, graph adapter, evidenceworkflows-evidence.md
Surfaces + pause/resume + sandboxessurfaces-pause.md

Key Conventions

  • One workspace per issue; PathSafety sanitizes identifiers and asserts paths stay inside the workspace root. Never run an agent outside its workspace.
  • A runner returns :ok (retry-continue), {:error, reason} (backoff retry), or {:pause, reason, token} (park in the orchestrator's paused map).
  • The Review runner's Contract deliberately carries no workspace reference -- reviewer isolation is an invariant.

Common Pitfalls

  1. Blocking on human input inline -- return {:pause, reason, token} and let a surface/Resumer resume; don't sleep the runner.
  2. Bad WORKFLOW.md -- WorkflowStore falls back to the last-known-good config and records last_error/1; check it after edits.
  3. Assuming production-ready -- v0.1 pre-alpha; pin versions and expect churn.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.