模組 1 · Symphony Baseline and the Comparison Frame
第 1 節,共 5 節 · 4 分鐘 · 英語 · 唯讀
Symphony is an open-sourced agent orchestrator: it turns a project-management board into a control plane for coding agents, so teams manage work rather than supervise individual agent sessions. This lesson names the capability categories that define that baseline and separates orchestration-level concerns from the model-level work the agents themselves perform.
Symphony is an agent orchestrator that turns a project-management board like Linear into a control plane for coding agents. In the published specification, it is a long-running automation service that continuously reads work from a configured issue tracker, creates an isolated workspace for each issue, and runs a coding agent session for that issue inside the workspace. The repository describes the same idea from the team's side: Symphony turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising coding agents.
That framing already tells you where the boundary sits. Symphony does not itself answer prompts or generate code; it decides which work is eligible, prepares a place for an agent to run, launches that agent, and tracks what happens. The coding agent is the component that does the inference and edits. Symphony's feature set is therefore about control flow and coordination, not about model quality.
來源支援
The specification lists four operational problems the service solves, and each one maps onto a capability category rather than onto a model feature. It turns issue execution into a repeatable daemon workflow instead of manual scripts, which is the orchestration category. It isolates agent execution in per-issue workspaces so agent commands run only inside per-issue workspace directories, which is the isolation category. It keeps the workflow policy in-repo in a `WORKFLOW.md` file so teams version the agent prompt and runtime settings with their code, which is the configuration category. And it provides enough observability to operate and debug multiple concurrent agent runs, which is the operations category.
The specification also states the boundary explicitly: Symphony is a scheduler, runner, and tracker reader. Ticket writes such as state transitions, comments, and PR links are typically performed by the coding agent through provider-native tools executed by Symphony with the configured tracker credential. That split is the clearest evidence for the orchestration-versus-model distinction: the orchestrator decides and dispatches, while the agent acts.
來源支援
Given the feature description "routes a user request to a specialized agent and returns the result," classify it as orchestration-level or model-level, and justify the classification using the Symphony baseline categories. Assume the description is a hypothetical feature statement, not a quoted Symphony capability.
Start by asking which component performs the work in the description. Routing a request to a specialized agent is a dispatch decision: something must decide which agent should handle the request and then hand the work over. In the Symphony baseline, dispatch decisions belong to the orchestration category, which the specification describes as owning the poll tick, the in-memory runtime state, and the decision about which issues to dispatch, retry, stop, or release.
Next, check whether the description requires any inference. Returning the result is a handoff of output, not a generation step; nothing in the description says a model produced the answer. Compare that with the coding agent, which the specification places in the execution layer and which is the component that actually runs inside the workspace.
Because the description is entirely about selecting and handing off work, and contains no inference step, it is orchestration-level. The same test applies in reverse: a description such as "generates a patch for the failing test" names an inference and editing action, so it is model-level or agent-level work that Symphony coordinates rather than performs.
來源支援
來源支援
想一想
Pick one capability from the baseline list and describe a feature that would look similar but actually belong to the coding agent rather than to Symphony. What in the feature description tells you which side of the boundary it falls on?
關係摘要
These edges describe composition and contrast, not execution order. The orchestration category is what Symphony itself provides; the coding agent is the separate component whose inference and edits Symphony coordinates.
Symphony baseline
The capability categories that define the offering
Agent orchestration
Coordinating agents, workspaces, and runs
是其中一部分 Symphony baseline
Tracker intake
Continuously reads work from a configured issue tracker
是其中一部分 Agent orchestration
Workspace isolation
Creates an isolated workspace per issue
是其中一部分 Agent orchestration
Agent session runner
Runs a coding agent session inside that workspace
是其中一部分 Agent orchestration
Workflow policy
Repository-owned WORKFLOW.md contract
是其中一部分 Symphony baseline
Observability
Structured logs and operator-visible status
是其中一部分 Symphony baseline
Coding agent
The component that does the inference and edits
來源支援