The `useEffect` that syncs `chatRecord.status` from React Query
unconditionally overwrites the store's `chatStatus`. The `chat(chatId)`
query has no `staleTime` (defaults to 0), so it refetches on window
focus, remount, etc. If the REST response catches a transient
`"pending"` status (e.g. between multi-step tool-call cycles), it
regresses `chatStatus` from `"running"` to `"pending"`.
Since `shouldApplyMessagePart()` drops ALL parts when status is
`"pending"` or `"waiting"`, every incoming `message_part` event is
silently discarded — not even buffered. Parts are visible on the
WebSocket but nothing renders, and the UI shows "Response is taking
longer than expected". A page reload fixes it because a fresh REST fetch
returns the current status.
**Fix:** Add `wsStatusReceivedRef` — once the WebSocket delivers a
status event, it becomes the authoritative source and REST refetches can
no longer overwrite it. This mirrors the existing
`wsQueueUpdateReceivedRef` pattern already used for queued messages. The
ref resets on chat change.
> Generated with [Coder Agents](https://coder.com/agents)
Aligns the copy/edit action bar so both user and assistant messages use
the same hover-to-reveal pattern.
## Changes
- Replace bifurcated copy UX (inline `afterResponseSlot` for assistant,
floating toolbar for user) with a single unified action bar using
`CopyButton` + optional edit `Button`
- Remove `BlockList` `afterResponseSlot` prop and related machinery
- Remove per-message `copyHovered`/`useClipboard` state and left-border
highlight effect
- Remove `lastAssistantPerTurnIds`/`isTurnActive` computation — all
messages with content get actions on hover
- Hide actions on mid-chain assistant messages (only last in consecutive
chain shows buttons)
- Reduce inter-message gap from `gap-3` to `gap-2`
- Shrink action buttons to `size-6` for tighter vertical spacing
- Add 8px sticky top offset for user messages
> 🤖 Generated by Coder Agents
## Problem
MCP servers configured in `.mcp.json` with stdio transport are
discovered successfully (tools appear) but die immediately after
connection, making all tool calls fail.
## Root Cause
In `connectServer`, the subprocess is spawned with `connectCtx` — a
30-second timeout context whose `cancel()` is deferred:
```go
connectCtx, cancel := context.WithTimeout(ctx, connectTimeout)
defer cancel()
if err := c.Start(connectCtx); err != nil { ... }
```
The mcp-go stdio transport calls `exec.CommandContext(connectCtx, ...)`.
When `connectServer` returns, `cancel()` fires, and
`exec.CommandContext` sends SIGKILL to the subprocess. The process
immediately becomes a zombie.
Confirmed by checking `/proc/<pid>/status` after context cancellation:
```
State: Z (zombie)
```
## Fix
Pass the parent `ctx` (which is `a.gracefulCtx` — the agent's long-lived
context) to `c.Start()`. `connectCtx` continues to bound only the
`Initialize()` handshake. The subprocess is cleaned up when the Manager
is closed or the parent context is canceled.
## Regression Test
Added `TestConnectServer_StdioProcessSurvivesConnect` which:
- Spawns a real subprocess (re-execs the test binary as a fake MCP
server)
- Calls `connectServer` and lets it return (internal `connectCtx` gets
canceled)
- Verifies the subprocess is still alive by calling `ListTools`
The test **fails** on the old code with `transport error: context
deadline exceeded` and **passes** with the fix.
> Generated with [Coder Agents](https://coder.com/agents)
## Summary
Move `ConvertMessagesWithFiles` into the `g2` errgroup so prompt
conversion runs concurrently with instruction persistence, user prompt
resolution, MCP server connections, and workspace MCP tool discovery.
## Problem
In `runChat`, the setup before the first LLM `Stream()` call is
sequential across two errgroups:
```
g.Wait() // model + messages + MCP configs
ConvertMessagesWithFiles() // sequential — blocked on g2 starting
g2.Wait() // instructions + user prompt + MCP connect + workspace MCP
```
`ConvertMessagesWithFiles` can take non-trivial time on conversations
with file attachments (batch DB resolution), and it was blocking g2 from
starting.
## Fix
`ConvertMessagesWithFiles` only reads the `messages` slice (available
after `g.Wait()`) and resolves file references via the database. No g2
task reads or writes the `prompt` variable. This makes it safe to
overlap with g2:
```
g.Wait()
g2.Wait() // now includes ConvertMessagesWithFiles in parallel
```
The `InsertSystem` call for parent chats and the `promptErr` check are
deferred to after `g2.Wait()`, preserving correctness.
<details><summary>Decision log</summary>
- `ConvertMessagesWithFiles` is read-only on `messages` — no mutation,
safe for concurrent access
- `prompt` and `promptErr` are written only by the conversion goroutine,
read only after `g2.Wait()` — no data race
- Error from prompt conversion is checked immediately after `g2.Wait()`,
before any code that uses `prompt`
- `chatloop.Run` now uses `:=` instead of `=` since the prior `err`
declaration from `prompt, err :=` was removed
</details>
> Generated by Coder Agents
Piggybacks on #23878. Moves instruction file reading and skill discovery
from `chatd` (server-side, via multiple `LS`/`ReadFile` round-trips
through the agent connection) to the agent itself (local filesystem
access).
This intentionally drops backward compatibility with older agents that
don't support the context-config endpoint. Agents and server are
deployed together; there is no rolling-update contract to maintain here.
## What changed
The agent's `GET /api/v0/context-config` response now returns
`[]ChatMessagePart` directly — the same types chatd persists. This
eliminates intermediate type conversions and makes the protocol
extensible.
| Field | Type | Description |
|---|---|---|
| `parts` | `[]ChatMessagePart` | Context-file and skill parts, ready to
persist |
| `working_dir` | `string` | Agent's resolved working directory |
Removed from the response: `instructions_dirs`, `instructions_file`,
`skills_dirs`, `skill_meta_file`, `mcp_config_files` — the agent reads
files locally and returns their content as parts.
Removed from chatd: all legacy `LS`/`ReadFile` fallback code
(`readHomeInstructionFile`, `readInstructionDirFile`, `DiscoverSkills`
via LS, etc).
## Why
The previous architecture had the agent resolve paths, serve them over
HTTP, then `chatd` make N+1 round-trips back through the agent
connection to read files. The agent has direct filesystem access and
should just read the files.
## Key design decisions
- **Agent returns `ChatMessagePart` directly** — same types chatd
persists. No intermediate `InstructionFileEntry`/`SkillEntry` types
needed.
- **`SkillMeta.MetaFile`** — persisted via `ContextFileSkillMetaFile` on
the skill part, so custom meta file names
(`CODER_AGENT_EXP_SKILL_META_FILE`) survive across chat turns.
- **No pre-read body** — `read_skill` always dials the workspace to
fetch the skill body on demand. Simpler than caching the body in the
response.
- **MCP config paths kept agent-internal** — `MCPConfigFiles()` getter,
not sent over the wire.
- **No backward compat fallback** — old agents that don't support
context-config get no instruction files. This is acceptable since agent
and server deploy together.
Following on from #23989#24018
- We also no longer want to collect `IsBusiness` demographic data
- Newsletter fields no longer allow `nil` as a value, instead default to
false
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
New `IndustryType` and `OrgSize` enums were added in #23989, but they
are no longer desired in the onboarding/marketing telemetry data. This
removes them.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Two new columns added to aibridge_token_usages:
- cache_read_input_tokens (BIGINT, default 0)
- cache_write_input_tokens (BIGINT, default 0)
Migration backfills existing rows by extracting values from the metadata
JSONB column (cache_read_input, input_cached, prompt_cached for reads
(max value selected since only 1 should be set), cache_creation_input
for writes).
All references to data from metadata were updated to reference new
columns. No other changes then changing where data is extracted from.
Requires aibridge library version bump to include:
https://github.com/coder/aibridge/pull/229
Fixes: https://github.com/coder/aibridge/issues/150
Add optional demographic and newsletter preference fields to the setup
page: business use (yes/no), industry type, organization size, and two
newsletter toggles (marketing, release/security updates).
The new data flows through telemetry via a FirstUserOnboarding struct in
the snapshot payload, sent once when the first user is created. The
telemetry-server and BigQuery schema changes are required separately to
persist this data.
---------
Co-authored-by: default <davidiii@fraley.us>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Improves the copy button on the last assistant message in agent chat
conversations.
**Changes:**
- Copy icon aligned to the left edge of the content area
- No margin-top gap between message content and the copy button
- Hovering the copy button reveals a 2px vertical indicator line
spanning the copyable content
- Tooltip repositioned to bottom-left with compact sizing
- Extra padding below the copy button to visually separate from the next
user prompt
- Hover state triggers only on the button itself, not the entire row
> This PR was authored by Mux on behalf of @ibetitsmike.
Chromatic tests for `ChatModelAdminPanel` were failing because
synchronous
`getByRole`/`getByLabelText` calls were used after `userEvent.click()`
actions
that trigger navigation to a conditionally-rendered detail view. The
switches
and inputs in `ProviderForm` aren't in the DOM immediately after click,
so the
synchronous queries fail with "Unable to find an accessible element."
Changed all post-click `getByRole`/`getByLabelText` calls to their async
`findByRole`/`findByLabelText` equivalents across three stories:
`ProviderWithUserKeysEnabled`, `EnvPresetProviders`, and
`CreateAndUpdateProvider`.
## Problem
When a prebuilt workspace is claimed, the agent reinitializes via a
single fire-and-forget pubsub event over SSE. If the agent's SSE
connection is interrupted at claim time, the event is permanently lost —
the workspace is stuck with no self-healing path.
Additionally, regular (non-prebuild) workspaces had no way to opt out of
the `/reinit` polling loop — agents would reconnect indefinitely to an
endpoint that would never send them anything useful.
## Root Cause
`workspaceAgentReinit` fetches the workspace (with its current
`owner_id`) via `GetWorkspaceByAgentID`, but never checked whether a
claim already happened. It only subscribed to pubsub for future events.
The database already has durable claim state (`owner_id` changes from
`PrebuildsSystemUserID` to the real user), but no layer ever consulted
it on reconnection.
## Solution
### Server-side durable check with first-build-initiator gating
**TOCTOU-safe ordering**: Subscribe to pubsub claim events *before* any
durable checks, so a claim that fires during the check is buffered in
the channel rather than lost.
**First-build-initiator gating**: When `!workspace.IsPrebuild()` (owner
is no longer the system user), look up the first build's `InitiatorID`.
The prebuild reconciler always uses `PrebuildsSystemUserID` as the
initiator. This distinguishes claimed prebuilds from regular workspaces
without any SQL schema changes.
- **Regular workspace** (first build initiator ≠ system user) → **409
Conflict**, agent stops reconnecting
- **Claimed prebuild, build completed** → pre-seed channel with reinit
event and close it, transmitter delivers one-shot then exits
- **Claimed prebuild, build in-progress** → fall through to pubsub
subscription, agent waits for completion event
- **Unclaimed prebuild** → pubsub subscription (existing happy path)
### Declarative reinit events (defense-in-depth)
- Added `UserID` field to `ReinitializationEvent` with JSON tags
- Switched pubsub serialization from raw string to JSON (with
backward-compat fallback for rolling upgrades)
- Populated `UserID` at both the publish site and the durable check
### Agent SDK: 409 handling
`WaitForReinitLoop` detects 409 Conflict from the server and closes the
`reinitEvents` channel, cleanly exiting the retry goroutine.
### Agent CLI: fixed two bugs + added reinitCtx
- **Closed channel (`!ok`)**: now blocks on `<-ctx.Done()` instead of
`continue`, keeping the current agent running. Previously this would
leak agents by skipping `agnt.Close()` and re-entering the loop.
- **Duplicate owner reinit**: cancels `reinitCtx` (stops the reinit
goroutine), then blocks on `<-ctx.Done()`. Previously `continue` would
skip cleanup and create a new agent on the next loop iteration.
- **`reinitCtx`**: a cancellable child of `ctx` passed to
`WaitForReinitLoop`, allowing the agent to stop the reinit HTTP polling
after reinit completes.
### Agent-side idempotency
Tracks `lastOwnerID` in the agent reinit loop — duplicate events for the
same owner are skipped.
## Testing
- **"unclaimed prebuild receives reinit via pubsub"**: prebuild owned by
system user, pubsub event triggers reinit
- **"claimed prebuild receives one-shot reinit on reconnect"**: first
build by system user, owner changed, build completed → immediate reinit
(no pubsub needed)
- **"claimed prebuild waits during in-progress claim build"**: claimed
but build still running → no reinit until build completes
- **"regular workspace gets 409"**: first build by real user → 409
Conflict, agent stops polling
- Updated claim publisher/listener tests: verify `UserID` survives JSON
round-trip + backward compat with raw string payloads
- Updated SSE round-trip test: verify `UserID` survives transmit →
receive cycle
Fixes#22359
## Rolling upgrade note
During a rolling deploy where old coderd instances coexist with new
ones, the pubsub `ReinitializationEvent` has a new `workspace_id` field
(JSON key `workspace_id`). Old publishers send a raw reason string
instead of JSON; the new listener gracefully falls back by treating the
entire payload as the reason and filling in `WorkspaceID` from context.
The only visible effect during the upgrade window is that `WorkspaceID`
may be the zero UUID in agent-side logs — this is cosmetic and resolves
once all instances are updated.
Add a nullable `value_key_id` column to the `user_secrets` table with a
foreign key to `dbcrypt_keys`. This is the column dbcrypt uses to track
which encryption key encrypted a given secret's value. This is required
for encryption of user secret values.
The column was missing from the original migration (000357).
Frontend for provider key policies (backend in #23751).
## Changes
**Admin provider form**: Three policy toggles (central API key, user API
keys, central fallback) with cross-field validation and conditional
visibility. Form resets properly after save.
**User settings page**: New `/settings/providers` route for personal API
key management. Conditional sidebar item (visible only when providers
allow user keys). Status badges, masked key input, save/remove actions
with confirmation. Read-only model list per provider. Gated behind
`agents` experiment flag.
**Model selector**: Distinguishes user-fixable (`user_api_key_required`)
from admin-fixable (`missing_api_key`) empty states. Links to
`/settings/providers` when user action is needed. Applied to both chat
detail and agent create flows.
**API client**: Query/mutation hooks for user provider configs. Cache
invalidation across provider configs and model catalog.
When a user tries to archive-and-delete a chat from /agents but the
workspace is already gone, the UI showed a "Failed to look up workspace
for deletion" toast and blocked the archive. This change detects the
workspace-gone response and archives the chat without attempting
deletion.
## Changes
The backend returns 410 Gone for soft-deleted workspaces and 404 for
workspaces that do not exist or the user cannot access.
`isWorkspaceNotFound()` detects both status codes.
`resolveArchiveAndDeleteAction()` now returns `"archive"` when the
workspace preflight fetch gets a 404 or 410, and the page branches on
that action to call the existing archive mutation directly. The
`archiveAndDeleteMutation` also tolerates these status codes from
`deleteWorkspace()` to handle the race where the workspace disappears
between the preflight lookup and the actual delete call.
The mutation body was extracted into a testable
`archiveAndDeleteWorkspace()` utility so the tolerance logic has direct
test coverage. A `navigateAfterArchive()` helper consolidates the
post-archive redirect logic that was previously duplicated across the
proceed, confirm, and archive paths.
## Pre-existing patterns preserved
- The `"proceed"` and `"confirm"` archive-and-delete paths use
`onSettled` for navigation, matching the existing behavior before this
change. Only the new `"archive"` path uses `onSuccess` since it has no
workspace deletion step that should still navigate on partial failure.
- `isWorkspaceNotFound()` uses the same `isAxiosError(error) &&
error.response?.status` pattern already used in several places in
`site/src/api/api.ts`. The backend 404 ambiguity (deleted vs
unauthorized) is documented in the JSDoc.
- The pre-existing double-submit race during the async preflight window
is unchanged.
This PR modifies the `wait_agent` tool call card to display screen
recordings of computer use subagents. The backend logic was added in
https://github.com/coder/coder/pull/23894.
There's one big inefficiency in the current implementation: to display
video thumbnails, the frontend downloads the entire video files from the
backend. Our backend does not support HTTP range requests to only fetch
the first frame. I'll be fixing that in a later PR.
https://github.com/user-attachments/assets/684cea8b-66a9-45f8-96b2-57433da41c1c
This PR introduces screen recording of the computer use agent using the
virtual desktop.
- Screen recording is triggered by a `wait_agent` tool call. Recording
is stopped by a successful `wait_agent` tool call or when there hasn't
been any desktop activity for 10 minutes.
- Recordings are handled by the `portabledesktop` cli via the `record`
command. The videos are sped up in periods of inactivity.
- Recordings are saved to the database to the `chat_files` table.
There's a hard limit of 100MB per recording. Larger recordings are
dropped.
- A successful `wait_agent` on a computer use subagent tool call returns
a `recording_file_id`, later allowing the frontend to display the
corresponding video.
`isContextLimitKey` had a fallback heuristic that matched any key starting with `"max"` containing `"context"`, causing false positives on keys like `"max_context_version"`. A provider returning such metadata would have the value parsed as a context limit.
Replace substring matching on the separator-stripped key with word-level matching. A new `metadataKeyWords` function tokenizes keys by splitting on separators and camelCase boundaries, then the fallback requires
`"context"` paired with a limit-related word (`"limit"`, `"window"` + qualifier, `"length"` + qualifier, or `"tokens"` + qualifier). Known exact forms like `"context_window"` remain in the fast-path switch.
Closes https://github.com/coder/coder/issues/23332
Add `-x` to backport script `git cherry-pick` command to include a
commit message reference to the original commit. This makes it easier to
trace where a cherry picked commit actually came from.
- Extend `TestChatTemplateAllowlistEnforcement` to also exercise
`read_template` and `create_workspace` through the allowlist
- Mock LLM now chains 4 tool calls: list_templates, read_template
(blocked), read_template (allowed), create_workspace (blocked)
- Wire dummy `CreateWorkspace` config into test server so the tool
reaches the allowlist check
- Generalize tool result collection to support multiple calls per tool
name
> 🤖 Created by Coder Agents and reviewed by Kyle the human.
When the `agents` experiment is enabled, new users are automatically
granted the `agents-access` role at creation time so they can use Coder
Agents without manual admin intervention.
- Auto-assigns in `CreateUser()` — covers admin API, OAuth, and OIDC
creation paths
- Skips auto-assign for OIDC users when enterprise site role sync is
enabled (sync overwrites roles on every login; those admins should use
`--oidc-user-role-default` instead)
- CLI `create-admin-user` bypasses `CreateUser()` but creates `owner`
users who already have all permissions
> 🤖 Written by a Coder Agent. Will be reviewed by a human.
_Disclaimer: created using Claude Opus 4.6._
```
# Examples:
# ./scripts/backport-pr.sh 2.30 23969
# ./scripts/backport-pr.sh --dry-run 2.30 23969
```
Here's one I created: https://github.com/coder/coder/pull/23972
Signed-off-by: Danny Kopping <danny@coder.com>
- Change `errChatHasNoWorkspaceAgent` message from cryptic `"chat has no
workspace agent"` to actionable `"workspace has no running agent: the
workspace may be stopped. Use the start_workspace tool to start it, or
create_workspace to create a new one"`
- Update test assertions to match the new message substring
> 🤖 Written by a Coder Agent. Reviewed by a human.
Add language reference docs that the Modernization Reviewer reads
before reviewing TS/React code, matching the existing Go reference
(.claude/docs/GO.md).
- references/typescript.md: Modern TypeScript 5.0-6.0 RC patterns,
replacements, and new capabilities
- references/react.md: Modern React 18-19.2 + Compiler 1.0 patterns,
replacements, and new capabilities
SKILL.md updated to reference these docs in the Tier 2 file filters
and spawn prompt instructions.
Refs #23500
*Disclaimer: implemented by a Coder Agent using Claude Opus 4.6*
## Summary
Two changes on the AI Bridge sessions page
(`/aibridge/sessions/<session>`):
1. **Updated header subtitle and link** — replaced the generic
"Centralized auditing for LLM usage across your organization. More about
AI Governance" with auditing-specific copy and a link to the [AI Bridge
audit docs](https://coder.com/docs/ai-coder/ai-bridge/audit).
2. **Added prompt attribution tooltip** — each user prompt now shows an
info icon with a tooltip explaining that prompt origin cannot be
reliably determined (human vs. agent), linking to the [attribution
docs](https://coder.com/docs/ai-coder/ai-bridge/audit#human-vs-agent-attribution).
## Changes
| File | What changed |
|------|-------------|
| `AIBridgeSessionsLayout.tsx` | Updated subtitle text and link target |
| `SessionTimeline.tsx` | Added `InfoIcon` + `Tooltip` next to the
"Prompt" label in `ThreadItem` |
<img width="954" height="318" alt="image"
src="https://github.com/user-attachments/assets/db3ca443-cb0f-426a-8457-4625c82fd6ba"
/>
---------
Signed-off-by: Danny Kopping <danny@coder.com>
Reuse the CopyButton component to let users copy plan content from
the propose_plan tool output. Follows the same pattern used by the
assistant message copy button.
- stabilize `TestAwaitSubagentCompletion/CompletesViaPubsub` by waiting
for durable completion state before sending the synthetic pubsub wake
- add coverage for successful subagent completion with an empty report
> 🤖 Written by a Coder Agent. Reviewed by a human.
Some clients (e.g. Claude) send a HEAD request without credentials as a
connectivity check before making actual API calls. This was logging at
`Warn` level, creating noise. Downgrade to Info for unauthenticated HEAD
requests and add the HTTP method to the logger for better observability.
Related to internal slack thread:
https://codercom.slack.com/archives/C0AEHQGLW22/p1775045200997309
## Description
Adds `provider_name` to aibridge interceptions to store the provider
instance name alongside the provider type. This allows distinguishing
between multiple instances of the same provider type (e.g. `copilot` vs
`copilot-business`).
## Changes
* Add `provider_name` column to `aibridge_interceptions` table with
backfill from `provider`.
* Add `provider_name` field to the proto `RecordInterceptionRequest`
message.
* Add `ProviderName` to the `codersdk.AIBridgeInterception` API
response.
_Disclaimer: initially produced by Claude Opus 4.6, modified and
reviewed by @ssncferreira ._
Closes https://github.com/coder/internal/issues/1432
Closes https://github.com/coder/internal/issues/1399
The test setup in `createWorkspaceWithApps` opens a short-lived RPC
connection to fetch the agent manifest before starting the real agent.
This connection used `ConnectRPC()` which sends no `role` parameter, so
the server treated it as a real agent connection and enabled connection
monitoring. When the helper closed, its monitor asynchronously wrote
`disconnectedAt` to the DB — racing with the real agent's monitor and
transiently marking the agent as disconnected.
The fix uses `ConnectRPCWithRole(ctx, "apptest-manifest")` so the helper
doesn't trigger connection monitoring. The server already has this
role-based distinction for non-agent clients like
`coder-logstream-kube`; the test helper just wasn't using it.
Both issues share this codepath: `setupProxyTest` →
`createWorkspaceWithApps` → the `ConnectRPC` call at `setup.go:518`.
Both test configurations have a non-empty `PrimaryAppHost`, so both
enter the affected block.
This is not masking a product issue — the "disconnected" state was
caused by two competing monitors writing to the same agent DB row, a
scenario that only exists in this test setup. No assertions were
weakened; the proxy still checks real agent connectivity on every
request.
The copy button on the last assistant message was showing even while the
turn was still in progress (agent streaming or running tool calls). The
content is not final at that point, so the button should be suppressed
until the turn completes.
The `lastAssistantPerTurnIds` computation unconditionally included the
trailing assistant message. Now it checks a new `isTurnActive` prop
derived from `isActiveChatStatus(chatStatus) || hasStreamState` and
skips the trailing ID when the turn is active. Completed turns (those
followed by a user message) are unaffected.
Kyle Carberry