amplifier-shell
Skill5 starsChanged 5 days agoMITView on GitHub
Customize Amplifier Unified presentation or build shell extensions. USE WHEN changing layouts, themes, sidebar modules, or artifact viewers, interactive conversation surfaces, and validating hot-loaded components. DO NOT USE WHEN changing AmplifierSession tools, providers, or orchestration; use runtime bundle guidance.
In this repository4 other files it gives its agents.
- AGENTS.mdAGENTS.md
- appearance-authoringSkill
- coordinate-workSkill
- unified-appSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use this skill for the web shell.
- Very specific: line 14
Read app state using app_control(operation="get_state").
- Very specific: line 27
If working outside an app session, use the host's authenticated GET /api/actions and POST /api/actions interfaces through the user's …
- Distinctive: line 37
Change existing presentation: Read presentation.md. Reuse installed modules, layouts and themes before creating code.
- Distinctive: line 39
Create interactive UX for a conversation: Read conversation-surfaces.md.
- Distinctive: line 42
Build or replace a component: Read authoring.md, then the relevant contract in the shell guide.
- Distinctive: line 53
Use shared actions and the public SDK.
- Distinctive: line 59
Use current revisions and view generations from inspection.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Amplifier Unified shell
Use this skill for the web shell. A shell module changes presentation; an Amplifier behavior changes the agent's runtime configuration. This skill is delivered by a behavior so agents can learn the shell's separate extension API.
Discover the target
- Read app state using
app_control(operation="get_state"). Identify the user's intended browser client fromvisibleUI.clientId; never assume a global presentation or choose an arbitrary attached client. Follow$statePathreferences with bounded reads when the overview omits details. - Request current action schemas using
app_control(operation="list_actions", args={"prefix":"shell."})and, for artifact views,args={"prefix":"canvas.views."}. Readitems, followingnextOffsetwith the same prefix and revision. For an oversized schema, use itsschemaReadarguments and concatenate allschemaTextpages before parsing the action. Inspect the relevant shell or view before preparing a change. For shell components, use the returnedslots,registry,resolvedInstancesandcomponentCommands; only author against contracts that host exposes. Discover theme actions separately when a CSS skin is requested. - If working outside an app session, use the host's authenticated
GET /api/actionsandPOST /api/actionsinterfaces through the user's configured connection. If no connection exists, author locally and report activation as unverified. Never read credential files to invent a connection.
Success criteria: An explicit target client and the current supported actions are known, or work is clearly limited to an offline package.
Choose the workflow
- Change existing presentation: Read presentation.md. Reuse installed modules, layouts and themes before creating code.
- Create interactive UX for a conversation: Read conversation-surfaces.md. Keep one surface ID and refine its existing tab, preserving shared user/agent state.
- Build or replace a component: Read authoring.md, then the relevant contract in the shell guide. The public SDK is packages/shell-sdk.
These paths are relative to this skill's directory. The same resource layout
is included in installed app packages; a development checkout is not required
to read the guide, SDK, or examples. Locate companions relative to the path
returned by load_skill, not the conversation's workspace.
Preserve ownership and verify
Use shared actions and the public SDK. Canonical conversations, drafts,
sessions, resources and tool execution belong to the host. Keep passive
navigation passive. Read-only browsing must not select a chat or send context.
Declare unsaved component edits with host.setDirty; retain dirty components
until their edits are resolved. Never use recovery to silently discard edits.
Use current revisions and view generations from inspection. Retry uncertain mutations with the same request ID and identical arguments. Validation approval is distinct from activation, and activation is distinct from user acceptance. Finish with the changed client/module/view, browser evidence, preservation checks, and a usable revert path. State any missing live validation explicitly.
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
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.

