
agent-host-chat-contributions
Build and review cross-cutting agent-host chat behavior through lifecycle contributions. Use when adding turn lifecycle side effects, prompt or context injection, restored-history transformation, protocol-action observation, or when reviewing changes that add code to AgentSideEffects or AgentService.
- Installs
- 41
- Last 24h
- —
- Trending rank
- —
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/microsoft/vscode --skill agent-host-chat-contributions- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads agent-host-chat-contributions automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
Agent Host Chat Contributions
Use the contribution model for a self-contained behavior that crosses an agent-host chat lifecycle boundary. Read these files before changing the model:
src/vs/platform/agentHost/common/agentHostChatContributionsService.tssrc/vs/platform/agentHost/node/agentHostChatContributionsService.tssrc/vs/platform/agentHost/node/chatContributions/TODO.mdsrc/vs/platform/agentHost/node/chatContributions/builtInChatContributions.ts
The rule that prevents the old failure mode
New cross-cutting features go in contributions, not in
AgentSideEffectsorAgentService.
AgentSideEffects has shed well over 700 lines by extracting seventeen behaviors. Do not put another lifecycle side effect, outgoing-prompt addition, restored-turn enrichment, admission check, or action observer back into either large orchestration class. AgentSideEffects dispatches the lifecycle and retains routing and send mechanics; AgentService owns session lifecycle and invokes hydration and memento eviction. See agentSideEffects.ts and agentService.ts.
This is also a review rule: code added to either file needs a specific reason that it is routing, a correctness invariant, provider-shaped behavior, or a dispatcher call site. "It was convenient to add here" is not a reason.
Decide before you code
Write a contribution when the behavior answers yes to this test:
- Does it react to a lifecycle moment the hooks already name — a turn request asking to proceed, a completed turn, a dispatched action, a turn about to be sent, a restored turn list, or a chat being restored?
- Can an explicitly registered, dependency-injected unit own it without changing protocol routing or provider mapping?
- Does it compose with other behavior in a defined order?
Examples are admission gating, local-command interception, checkpoint and changeset work, queue draining, GitHub-reference attachment, title handling, persisted usage, session input-needed aggregation, session flag persistence, chat drafts, a chat's own config, worktree announcements, unread state, rich-link guidance, and chat-surface guidance. Their directories under node/chatContributions/ are the reference implementations.
Do not use a contribution for the following:
- Subagent signal routing and buffering in
node/agentSideEffects.tsare routing fabric and correctness invariants. - Turn-id remapping in
node/agentSideEffects.tsis also routing fabric. - The three SDK event mappers and three attachment serializers are provider-shaped.
- Provider-shaped overlays applied during restore, such as seeding a draft's model from
IAgent-supplied session metadata in_doRestoreSession. Moving one would push provider metadata intoIHydrationContextfor one consumer.
If the answer is "this must route every signal correctly before state changes" or "this differs by SDK," it is not a contribution. Open work and standing caveats are tracked in node/chatContributions/TODO.md.
Keep the hook surface small
IAgentHostChatContribution has seven hooks in common/agentHostChatContributionsService.ts. Only add another method when no existing payload can express the behavior.
The set was once reduced from six to four:
onTurnConsumablebecameonTurnEndwithTurnEndReason.kind === 'localCommand'.onUserMessagebecameonOutgoingTurn.
Prefer adding a discriminant or payload to an existing type, such as a new TurnEndReason variant, over adding a hook. A proposed single-purpose hook with one caller and one implementer is the smell that produced both removals. Do not create an abstraction merely to move one direct call.
The three hooks added since clear a specific bar: no existing payload could express them because they fire at a different moment or from a different source, not merely with different data.
onHydrateChatrestores host-owned chat state before a chat enters the catalog.onHydrateTurnsruns lazily, after a provider returns turns, so reusing it would leave a restored peer chat untitled until it was opened.onIncomingRequestgates a turn before it reaches a provider, and can reject it. No observer hook can refuse.onDidDispatchActionobserves server-dispatched and rejected actions, whichonDidApplyClientActionstructurally cannot see.
Adding a { kind: ... } discriminant to an existing hook was considered and rejected for onHydrateChat: the discriminant guidance covers payload variants within one moment, the way TurnEndReason discriminates outcomes of a single turn ending.
Anatomy and registration
Give every contribution its own directory under node/chatContributions/<feature>/. It owns its implementation, helpers, and tests as it grows. A contribution must have:
static readonly id, unique across registrations.- An explicit
order, even when it currently appears independent. - A constructor whose first parameter is
IAgentHostChatContributionContext, followed by injected services. - A single registration in
node/chatContributions/builtInChatContributions.ts.
This is the complete shape of the existing Markdown plan rich-link contribution, adapted from markdownPlanRichLinks/markdownPlanRichLinksContribution.ts:
export class MarkdownPlanRichLinksContribution extends Disposable implements IAgentHostChatContribution {
static readonly id = 'markdownPlanRichLinks';
readonly order = 100;
constructor(
protected readonly _context: IAgentHostChatContributionContext,
@IAgentConfigurationService private readonly _agentConfigService: IAgentConfigurationService,
) {
super();
}
onOutgoingTurn(turn: IOutgoingTurn): ISendContribution | undefined {
return this._agentConfigService.getRootValue(platformRootSchema, AgentHostMarkdownPlanRichLinksEnabledConfigKey)
? { instructions: [createMarkdownPlanRichLinksInstruction(turn.chat)] }
: undefined;
}
}
Add its constructor to the one built-in list with contributions.registerContribution(MarkdownPlanRichLinksContribution). That list is activated by activateAgentHostContributions in node/agentHostContributions.ts; do not add a competing registration site. The dispatcher constructs each contribution through IInstantiationService, rejects duplicate ids, and disposes it on unregistration.
The seven hooks
| Hook | Fires when | Must not assume | Existing example |
|---|---|---|---|
onIncomingRequest | A turn request asks to proceed to a provider, before any side effect. Synchronous, and fails closed. | It decides admission, so a throw rejects rather than being skipped. Return accept, handled, or reject; the first non-accept wins and later contributions are not consulted. | localCommand 50, turnAdmission 100 |
onTurnEnd | Any terminal outcome a started turn can reach, plus a request refused before its turn started. | Discriminate rejected from error: the former has no started turn to finalize and no checkpoint to capture. Do not throw. It does not fire for an agent-emitted terminal action arriving with no active turn, because the reducer no-ops for those. | checkpointAndChangeset, queueDrain, githubReferences, sessionTitle, markUnread |
onDidApplyClientAction | A client-dispatched action was applied to host state. | It never sees server-dispatched or rejected actions. Anything it sees definitely reduced. | queueDrain tracks queued senders; sessionTitle persists user title changes. |
onDidDispatchAction | Any action was dispatched and its outcome is known, from any origin. | It also delivers rejected actions that never reduced — check rejectionReason. Implementing this and onDidApplyClientAction together sees every client action twice. | sessionInputNeeded, persistedTurnUsage, chatConfig, sessionFlags |
onOutgoingTurn | Awaited after admission and provider lookup, before the turn is sent. | Rejected turns and noAgent failures never reach it. IOutgoingTurn carries the full Message; contributions can replace its text in order and add instructions, but cannot replace its already-committed attachments, model, agent, origin, or metadata. Prefer the message-text channel for context injection: changing instructions invalidates the provider prompt cache and increases user cost. Do not bypass the send path. | markdownPlanRichLinks, chatSurface, sessionTitle, sideChat |
onHydrateTurns | A provider has returned the complete restored turn list for a chat. | It is not limited to default chats and must return a list for the next stage. Do not assume its input is pristine provider output because earlier contributions may have transformed it. | turnDelegation, persistedTurnUsage, worktreeAnnouncement, sideChat |
onHydrateChat | A chat is being restored, eagerly, before it enters the session catalog. | It runs with no provider — it is a metadata-only read — and before the chat exists, so a chat memento taken here would outlive a failed registration. | sessionTitle, chatDraft, chatConfig |
The call sites are in node/agentSideEffects.ts, node/agentHostTurnStarter.ts, and node/agentService.ts.
Ordering is behavior
order is load-bearing. The dispatcher sorts lower values first; registration order breaks ties only. Ordering is per hook, so the 100-series can be reused for hydration independently of outgoing turns. See node/agentHostChatContributionsService.ts.
Use explicit orders to preserve a required sequence:
LocalCommandContributionis 50 andTurnAdmissionContributionis 100 ononIncomingRequest. That order is load-bearing: local commands were always intercepted before the read-only guard, so/renamestill works in a read-only or archived chat.CheckpointAndChangesetContributionruns at 100 so it captures a checkpoint before later changeset-related work.QueueDrainContributionis 200 andSessionInputNeededContributionis 200.GitHubReferencesContributionandChatSurfaceContributionare 300 where applicable.SessionTitleContributionis 400,MarkUnreadContributionandSideChatContributionare 500,ChatDraftContributionis 600,ChatConfigContributionis 650, andSessionFlagsContributionis 700.
markUnread and sideChat both declare 500 and both implement onTurnEnd, so their relative sequence rests on registration order. Give a new contribution a distinct order rather than joining that tie.
A silent reordering regression already happened. The built-in-sequence regression tests in test/node/chatContributions.test.ts protect the original turn-end, outgoing-turn, and hydration sequences. Add or update those tests when ordering changes intentionally.
Depend on services, not a growing context
Never add callbacks or services to IAgentHostChatContributionContext. It is only contribution identity plus centrally evicted mementos. Coordinate through @Injectable services.
IAgentHostChatContributionHost is intentionally the narrow bridge to AgentSideEffects: it has only hostLaunchKind and sendTurnMessage. It grew to seven members before being cut back to two by promoting collaborators to real services:
IAgentHostSessionTitleControllerIAgentHostProviderServiceIAgentHostTelemetryReporterIAgentHostTurnTrackerIAgentHostToolCallTrackerIAgentHostLocalCommands
IAgentHostWorktreeIsolation is injectable too. "It isn't injectable" was a wrong conclusion caused by service-registration order and later disproved. Trace registration from node/agentHostServices.ts through node/agentHostBootstrap.ts and createAgentServiceComposition in node/agentServiceComposition.ts. Register a real service before the instantiation path constructs contributions; do not smuggle it through the context or host bridge.
Adding a service is not free: roughly six hand-built test ServiceCollections construct AgentSideEffects or call registerBuiltInChatContributions, and each needs the new registration or fails with UNKNOWN service <id>. Two related traps, both hit for real:
- Only wrap a test registration in
disposables.add(...)when the class extendsDisposable; wrapping a plain object throwsd.dispose is not a functionat teardown. - If the collaborator is stateless, prefer a plain function taking
ServicesAccessor, invoked throughIInstantiationService.invokeFunction, and skip DI entirely.node/agentHostTurnStarter.tsis the reference: making it a service bought nothing and cost a registration in every graph.
Mementos have a lifecycle
Create keys with createChatMementoKey or createSessionMementoKey in common/agentHostChatContributionsService.ts. Contexts are contribution-scoped, so identical debug names in two contributions do not collide.
- Chat mementos are evicted when their chat is disposed.
- Session mementos and all supplied chat mementos are evicted when their owning session is disposed.
- Extra key segments are part of the identity. Use them for bounded identities such as a queued message id.
- Call
deleteMementowhen an extra-segment value is no longer needed. Setting its observable toundefinedleaves its map entry alive until chat or session disposal, so a long-lived chat otherwise accumulates one entry per segment value. onHydrateChatruns before its chat is registered, so a chat memento taken there would outlive a registration that then failed. No current contribution does this, but do not be the first without handling it.
queueDrain/queueDrainContribution.ts demonstrates the required delete on queued-message removal and consumption. Central eviction lives in node/agentHostChatContributionsService.ts and is called by node/agentService.ts.
Failure isolation, and the one exception
The dispatcher logs and isolates failures independently for every observer hook:
- A throwing
onTurnEnd,onDidApplyClientAction, oronDidDispatchActioncontribution does not stop later contributions. - A failed
onOutgoingTurncontribution adds no instructions and does not block the send or later contributions. - A failed
onHydrateTurnsoronHydrateChatcontribution deliberately passes the previous value to the next stage. Losing chat history, a title, or a draft is worse than losing one enrichment.
onIncomingRequest inverts this and fails closed. A contribution that throws there rejects the request with internalError at stage validation. That gate is the enforcement behind the UI hiding the composer, so failing open would let a buggy or remote client run work in a session that may no longer have its isolated worktree on disk. Losing one enrichment is survivable; letting a request past a guard is not.
It is also deliberately synchronous, for the same reason: a gate has to decide before the send path performs any await, so the state it reads cannot change between the decision and its effect. Every admission check the host makes today is already synchronous. An async version was tried first and broke three agentSideEffects tests by deferring the read-only guard one microtask past the assertions — which was the design saying the gate belongs before the first await.
Implement normal error handling where it improves the feature, but never depend on a thrown error to control lifecycle flow. The exact dispatch behavior is in node/agentHostChatContributionsService.ts.
Tests and extraction proof
Add focused dispatcher tests in test/node/chatContributions.test.ts for hook dispatch, ordering, memento eviction, and failure isolation. The same file owns the built-in-sequence regression tests.
When extracting existing behavior from AgentSideEffects, leave the existing test/node/agentSideEffects.test.ts suite unchanged. Its passing behavior is the proof that the extraction preserved semantics; test new dispatcher mechanics separately rather than rewriting the old assertions to fit the new structure.
One caveat that has bitten twice: some hand-built test graphs construct AgentSideEffects without calling registerBuiltInChatContributions, so they silently stop exercising behavior the moment it moves into a contribution. When an extraction breaks such a suite, add the missing registration so the graph mirrors production wiring — do not weaken the assertion. The failure means the test was verifying an incomplete graph, not that the extraction was wrong.
FAQ
What does the agent-host-chat-contributions skill do?
Build and review cross-cutting agent-host chat behavior through lifecycle contributions. Use when adding turn lifecycle side effects, prompt or context injection, restored-history transformation, protocol-action observation, or when reviewing changes that add code to AgentSideEffects or AgentService.
How do I install agent-host-chat-contributions?
Run npx skills add https://github.com/microsoft/vscode --skill agent-host-chat-contributions in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/agent-host-chat-contributions/ (personal) or .claude/skills/agent-host-chat-contributions/ (project).
Which AI agents can use agent-host-chat-contributions?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is agent-host-chat-contributions safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains agent-host-chat-contributions?
It is published by microsoft in microsoft/vscode. It has 41 recorded installs.
More from microsoft/vscode
memory-leak-audit
microsoft/vscode
Audit code for memory leaks and disposable issues. Use when reviewing event listeners, DOM handlers, lifecycle callbacks, or fixing leak reports. Covers addDisposableListener, Event.once, MutableDisposable, DisposableStore, and onWillDispose patterns.
azure-pipelines
microsoft/vscode
Use when validating Azure DevOps pipeline changes for the VS Code build. Covers queueing builds, checking build status, viewing logs, and iterating on pipeline YAML changes without waiting for full CI runs.
accessibility
microsoft/vscode
Primary accessibility skill for VS Code. REQUIRED for new feature and contribution work, and also applies to updates of existing UI. Covers accessibility help dialogs, accessible views, verbosity settings, signals, ARIA announcements, keyboard navigation, and ARIA labels/roles.
sessions
microsoft/vscode
Core principles and workflow router for changes to the Agents Window under src/vs/sessions.
author-contributions
microsoft/vscode
Identify all files a specific author contributed to on a branch vs its upstream, tracing code through renames. Use when asked who edited what, what code an author contributed, or to audit authorship before a merge. This skill should be run as a subagent — it performs many git operations and returns a concise table.
update-screenshots
microsoft/vscode
Update the committed blocks-ci screenshot hashes after the "Screenshots & Tests" check fails, or investigate a screenshot diff reported on a PR. Use when asked to update, accept, or refresh component screenshot baselines from CI. This skill should be run as a subagent.
Popular ai & agents skills
google-agents-cli-publish
google/agents-cli
This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).
google-agents-cli-eval
google/agents-cli
This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).
skill-creator
anthropics/skills
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
codex-pet
prime-skills/runcomfy-agent-skills
Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.
full-output-enforcement
leonxlnx/taste-skill
Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.
executing-plans
obra/superpowers
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available