Aspire Orchestration

Aspire Orchestration

MANDATORY COMPLIANCE — This skill prevents agent self-harm in Aspire projects. Violating these rules causes file locks, orphaned processes, and user frustration (#15801).

Prerequisites

Use the installation method owned by the user's environment. npm, Nix, Homebrew, WinGet, mise, and the install scripts are supported alongside the NativeAOT .NET global tool.

Detection

Activate when ANY signal is present:

See detection.md for detailed fingerprinting.

VS Code AppHost Lifecycle

When the agent host exposes aspire_apphost_start or aspire_apphost_stop, use the matching tool before running aspire start or aspire stop in a terminal, except when starting from a git worktree requires --isolated and the tool cannot request it. If the tool is listed as deferred, load its contract with the host's tool-discovery mechanism first; do not treat an unloaded deferred tool as unavailable.

Pass the exact selected appHostPath discovered by Aspire to editor lifecycle tools. In a multi-root workspace, that tool contract may require a selector such as repo-a~1/MyApp.AppHost/MyApp.AppHost.csproj.

The CLI --apphost flag does not understand that selector namespace. Before any CLI start/stop fallback, resolve the selected AppHost to its actual filesystem project path and pass that path to --apphost; never copy a multi-root selector verbatim. That CLI project path may be workspace-relative (MyApp.AppHost/MyApp.AppHost.csproj) or absolute (/workspaces/repo-a/MyApp.AppHost/MyApp.AppHost.csproj or C:\workspaces\repo-a\MyApp.AppHost\MyApp.AppHost.csproj); use the current platform's native path syntax. If several AppHosts are discovered and the user's target is unclear, ask which one to use instead of guessing, invoking the tool for every AppHost, or issuing an unscoped CLI command.

If the selected appHostPath is already a normal workspace-relative project path such as MyApp.AppHost/MyApp.AppHost.csproj, reuse it unchanged for CLI fallbacks; do not convert it to an absolute path just for the CLI.

An unclear target is a hard stop. Ask one clarifying question and wait for the user to name an AppHost. Do not call a lifecycle tool or terminal command until the target is resolved, and do not offer commands that bypass this gate. Stop multiple AppHosts only when the user explicitly requests all of them.

For starts, call aspire_apphost_start with mode run and the exact selected appHostPath unless the user explicitly asks to attach a debugger. In a git worktree, the editor tool does not currently request Aspire's isolated state. Resolve the selected AppHost to its filesystem path and use aspire start --non-interactive --isolated --apphost <filesystem-path> instead.

run mode has no debugger attached, but it still has an editor-owned Aspire session that aspire_apphost_stop can stop. After a stop call, follow this result matrix exactly:

The rows are mutually exclusive. Act only on the current result; do not offer a command from another row as a speculative future workaround. Re-evaluate only after a new tool result is returned.

ambiguousSession is a terminal safety refusal. Do not run or offer a CLI fallback, and do not ask whether the user wants one. User confirmation cannot make an ambiguous editor session safe to terminate from the CLI.

Use direct Aspire CLI lifecycle commands only when the matching editor tool is unavailable, for isolated worktree starts, or for a stop result explicitly marked as allowed above. When a CLI fallback is allowed, keep the target exact by resolving the selected AppHost to its filesystem path first.

Safety Guardrails

See safety-guardrails.md for detailed rules and recovery patterns.

Default Workflow

1. Confirm workspace is Aspire — identify the AppHost
2. If the workspace is a git worktree, resolve the selected AppHost to a filesystem project path and use aspire start --non-interactive --isolated --apphost <filesystem-path> even when aspire_apphost_start is available, because the editor tool cannot request isolation. Otherwise, start with aspire_apphost_start in mode run using the exact selected appHostPath when available; if that tool is unavailable, resolve the selected AppHost to a filesystem project path and use aspire start --non-interactive --apphost <filesystem-path>
3. aspire wait <resource> before interacting with any resource
4. aspire describe to inspect state, then work
5. If AppHost code changed, restart through the same lifecycle routing; if only one resource changed, prefer the resource's commands/watch/HMR/debug workflow
6. Stop with aspire_apphost_stop using the exact selected appHostPath when available; use aspire stop --non-interactive --apphost <filesystem-path> only for the documented fallback outcomes

Quick Reference

Error Handling

🔒 File-Lock Recovery (MSB3491 / CS2012) — Always Stop the AppHost First

When a build fails with error MSB3491: Could not write to output file ... or error CS2012: Cannot open ... for writing, the project itself is healthy — Aspire is running and holding file locks on the resource's output assemblies. The recovery is always the same:

Resolve the exact AppHost and call aspire_apphost_stop first when available. If a CLI fallback is permitted, resolve the selected AppHost to its filesystem path before using the commands below; ambiguousSession and all other no-fallback results end the recovery attempt.

# ✅ CLI stop only after the result matrix permits fallback
aspire stop --non-interactive --apphost <filesystem-path>  # release the locks
# ... then either rebuild / restart one resource if the resource exposes commands ...
aspire resource <name> rebuild   # example: C# project resource with rebuild command
# ... or restart the whole AppHost through aspire_apphost_start. If unavailable:
aspire start --non-interactive --apphost <filesystem-path>
# In a git worktree:
aspire start --non-interactive --isolated --apphost <filesystem-path>

The same rule applies to any "file in use", "cannot access the file", or "another process is using" error during a build of an Aspire-managed resource.

Handoff Rules

Runtime Settings And Environment

TypeScript AppHost Note

Detection covers current apphost.mts and legacy apphost.ts, but all TS AppHost authoring is delegated to aspireify. The CLI-driven legacy migration is project maintenance owned by this orchestration skill, not authoring. Explain that it also updates Aspire packages, config, tsconfig, and imports; offer aspire update --migrate --yes --non-interactive only after approval for the full change, then hand back to aspireify only if source authoring remains. Current rules to apply when handing off:

Skill Routing — In-Plugin Sibling Skills

After aspire init drops a skeleton AppHost + aspire.config.json, route AppHost wiring (scan repo → propose resource graph → edit AppHost → wire Aspire.ServiceDefaults / OTel → validate via aspire start) to the in-plugin aspireify skill: aspireify/SKILL.md. For first-run flows that only need the skeleton drop, see the in-plugin aspire-init skill: aspire-init/SKILL.md. This orchestration skill stays focused on lifecycle (start/stop/wait/restart) and never edits AppHost code itself.

Project-Local Skill Precedence

If .agents/skills/aspire/SKILL.md exists (from aspire agent init), defer to it for: C# AppHost editing, TS AppHost editing, Playwright handoff, investigation workflows. Safety guardrails from this plugin ALWAYS apply.

If .agents/skills/aspireify/SKILL.md exists project-locally (installed by aspire init in current Aspire), warn the user that a project-local aspireify skill is present and defer to it for AppHost wiring instead of the in-plugin sibling. Same precedence rule as the project-local aspire skill above: project-local wins, plugin guardrails still apply.

References

• safety-guardrails.md — Detailed rules and recovery patterns
• detection.md — Project fingerprinting
• app-commands.md — App lifecycle and bootstrap commands
• resource-management.md — Resource wait, restart, and operations
• agent-workflows.md — Common agent investigation, integration, TypeScript, and handoff workflows