chore: remove agents experiment flag and mark feature as beta#24432
Conversation
deansheather
force-pushed
the
remove-agents-experiment-flag
branch
2 times, most recently
from
0f034ed to
4a61a18
Compare
There was a problem hiding this comment.
Pull request overview
This PR graduates the Agents feature from an experiment flag to a default (beta) feature across backend routing, CLI surface area, frontend navigation/UX, and documentation, removing all ExperimentAgents gating while keeping the /api/experimental/* API paths.
Changes:
- Removed
ExperimentAgentsgating across backend routes, SDK experiment definitions, and frontend feature checks; made Agents visibility/roles unconditional. - Promoted the Agents CLI TUI from
coder exp agentstocoder agentsand regenerated CLI/docs artifacts accordingly. - Updated docs and UI to reflect “beta” status (including a beta badge in the Agents sidebar/header) and removed “enable experiment flag” instructions.
Reviewed changes
Copilot reviewed 36 out of 43 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| site/src/pages/WorkspacesPage/WorkspacesPage.tsx | Removes agents experiment gating for the chats-by-workspace query on Workspaces page. |
| site/src/pages/AgentsPage/components/Sidebar/AgentsSidebar.tsx | Adds beta stage badge next to the logo in the Agents sidebar. |
| site/src/pages/AgentsPage/components/Sidebar/AgentsSidebar.test.tsx | Wraps sidebar tests with TooltipProvider to support new tooltip-using badge. |
| site/src/pages/AgentsPage/components/AgentPageHeader.tsx | Adds beta stage badge next to the logo in the mobile Agents header. |
| site/src/modules/dashboard/Navbar/NavbarView.tsx | Removes experiment check from AgentsNavItem; renders based on canCreateChat only. |
| site/src/modules/dashboard/Navbar/NavbarView.stories.tsx | Removes storybook experiment parameter now that Agents is not gated. |
| site/src/api/typesGenerated.ts | Regenerates types to remove "agents" from Experiment union/list. |
| site/site.go | Sets AgentsTabVisible HTML state to true unconditionally. |
| plans/CODAGT-156-remove-experimental-flags.md | Adds implementation plan/verification checklist for removing Agents experiment flag. |
| enterprise/coderd/roles_test.go | Removes Agents experiment setup from enterprise roles test. |
| enterprise/coderd/exp_chats_test.go | Removes Agents experiment setup from enterprise chats tests. |
| docs/reference/cli/index.md | Adds agents command to generated CLI index. |
| docs/reference/cli/agents.md | Adds generated reference doc for coder agents. |
| docs/reference/api/schemas.md | Updates generated API schema docs to remove agents from experiment enum values. |
| docs/manifest.json | Removes “experimental” phrasing from Chats API description in docs manifest. |
| docs/ai-coder/agents/getting-started.md | Removes experiment enablement step; renumbers setup steps. |
| docs/ai-coder/agents/early-access.md | Removes --experiments=agents instructions; notes Agents available by default. |
| docs/ai-coder/agents/chats-api.md | Updates Chats API note from experimental-gated to beta. |
| codersdk/deployment.go | Removes ExperimentAgents constant and related display/known-experiments entries. |
| coderd/x/chatd/integration_test.go | Removes Agents experiment setup from chatd integration tests. |
| coderd/x/chatd/chatd_test.go | Removes Agents experiment setup from chatd tests. |
| coderd/roles.go | Always includes AgentsAccessRole in assignable site roles. |
| coderd/mcp_test.go | Removes Agents experiment setup and updates comments in MCP tests. |
| coderd/exp_chats_test.go | Removes Agents experiment setup from chats tests. |
| coderd/coderd.go | Removes experiment middleware from chat/MCP routes; makes blob: img-src CSP addition unconditional. |
| coderd/apidoc/swagger.json | Regenerates Swagger JSON to remove agents experiment enum/docs. |
| coderd/apidoc/docs.go | Regenerates embedded Swagger template to remove agents experiment enum/docs. |
| cli/testdata/coder_agents_--help.golden | Adds golden help output for coder agents. |
| cli/testdata/coder_--help.golden | Updates root help golden to include agents command. |
| cli/root.go | Moves agents command into core subcommands (out of experimental). |
| cli/agents_test.go | Renames tests and updates model constructor naming to non-exp. |
| cli/agents_styles.go | Adds shared TUI styles for the promoted agents command. |
| cli/agents_stream_test.go | Adds tests for streaming output consumption helper. |
| cli/agents_render_test.go | Renames render test to non-exp naming. |
| cli/agents_render.go | Adds transcript/tool rendering logic for agents TUI. |
| cli/agents_model.go | Renames the main TUI model from exp-prefixed to stable naming. |
| cli/agents_list.go | Adds chat list UI model (search, expand/collapse, navigation). |
| cli/agents_helpers.go | Adds terminal text sanitization helper. |
| cli/agents_e2e_test.go | Renames E2E tests and updates invocation to coder agents. |
| cli/agents_e2e_helpers_test.go | Updates backend/session helpers for new coder agents command path. |
| cli/agents_cmds.go | Adds tea messages/commands for loading chats, history, tool results, and streaming. |
| cli/agents_chat.go | Adds chat view model (streaming, rendering, tool-call overlays like ask_user_question). |
| cli/agents.go | Promotes command definition to coder agents and removes the agent alias. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| const chatsByWorkspaceQuery = useQuery({ | ||
| ...chatsByWorkspace(workspaceIds), | ||
| enabled: agentsEnabled && workspaceIds.length > 0, | ||
| enabled: workspaceIds.length > 0, | ||
| }); |
There was a problem hiding this comment.
Done in 04fb69d — gated the query on permissions.createChat, matching the existing canCreateChat pattern used elsewhere in the codebase.
Address Copilot review feedback on #24432: after dropping the experiment flag, the Workspaces page was fetching chatsByWorkspace for every user with workspaces, even users who can't create chats. The endpoint still runs a DB query + RBAC post-filter and the downstream UI (AgentsNavItem, chat link icon in the workspaces table) is already hidden for users without createChat, so the fetch is pure waste for them. Gate the useQuery on permissions.createChat, matching the existing canCreateChat pattern used in Navbar.tsx and AgentCreatePage.tsx.
deansheather
force-pushed
the
remove-agents-experiment-flag
branch
from
4a61a18 to
d37c1ee
Compare
|
/coder-agents-review |
![coder-agents-review[bot]](https://avatars.githubusercontent.com/in/3454270?s=60&v=4){kind=small}
There was a problem hiding this comment.
Clean experiment removal across 46 files. The ExperimentAgents flag is fully excised from Go, TypeScript, docs, and tests, with no orphaned references. The architectural decision to replace the two-layer gate (experiment flag + RBAC) with RBAC-only is sound: the AgentsAccessRole remains the access control mechanism, and the permissions.createChat frontend guard is tighter than the old experiment flag (skips useless API calls for users without the role). The emdash exclusion for golden files is a proactive fix for a pre-existing gap. Commit messages are well-structured.
Severity count: 1 P2, 3 P3, 1 P4, 2 Nit.
The P2 is a docs inconsistency where getting-started.md still says "experimental" while chats-api.md was updated to "beta." The P3s are a stale PR description that contradicts the diff, a stale doc comment on mcpDeploymentValues, and a misleading doc heading.
[DEREM-2] The PR description states: "the implementation files are still named cli/exp_agents_*.go and internal identifiers still use exp prefixes." The diff renames every file and every identifier. Update the description to reflect what was delivered.
"This is the gap between stated intent and actual action." (Mafu-san)
PS. 11 of 17 reviewers independently flagged the AgentsTabVisible metadata pipeline as dead code. The backend goroutine, HTML meta tag, metadata registration, and test mock are all orphaned. The mock still says false while production returns true. This needs a human decision: file a cleanup ticket or accept the dead plumbing.
docs/ai-coder/agents/getting-started.md:247
P2 [DEREM-1] chats-api.md line 4 was updated to say "The Chats API is in beta." This line still reads "The Chats API is experimental and may change without notice." Same API, two maturity labels in the same doc set, authored in the same PR. An operator reading the getting-started guide sees "experimental," clicks through to the Chats API reference, and sees "beta."
Fix: change to match chats-api.md:
> The Chats API is in beta and may change without notice.
(Leorio P2, Mafuuu P3)
🤖
coderd/mcp_test.go:21
P3 [DEREM-3] The comment reads "mcpDeploymentValues returns deployment values with the agents experiment enabled, which is required by the MCP server config endpoints." The function body no longer sets any experiment, and the MCP routes no longer require it. The comment actively contradicts the code.
Fix: rewrite to "mcpDeploymentValues returns deployment values for MCP tests." or inline coderdtest.DeploymentValues(t) at each call site.
(Knov P3, Bisky Nit, Kite Nit)
🤖
coderd/rbac/roles.go:897
Nit [DEREM-7] The comment reads "e.g. when the agents experiment is enabled" but the experiment no longer exists. The caller (coderd/roles.go:47) now includes this role unconditionally.
(Kurapika)
🤖
🤖 This review was automatically generated with Coder Agents.
| @@ -42,25 +42,9 @@ redesign. | |||
|
|
|||
| ## Enable Coder Agents | |||
There was a problem hiding this comment.
P3 [DEREM-4] The heading says "Enable Coder Agents" but the body says "Coder Agents is available by default. No experiment flags are required." The heading implies there is an enablement procedure; the body says there isn't one. The heading is a vestigial label from when the section had real enablement instructions.
Consider renaming to "Get started with Coder Agents" or "Set up Coder Agents" to match the body.
(Razor P3, Leorio Note)
🤖
There was a problem hiding this comment.
DEREM-4: Done in eeb8201 — heading renamed to 'Set up Coder Agents'.
| agentsTabVisible = experiments.Enabled(codersdk.ExperimentAgents) | ||
| } | ||
| data, err := json.Marshal(agentsTabVisible) | ||
| data, err := json.Marshal(true) |
There was a problem hiding this comment.
P4 [DEREM-5] The goroutine marshals a constant true into state.AgentsTabVisible. The HTML template emits it as a meta tag. useEmbeddedMetadata.ts registers it. No production component reads the value. The AgentsNavItem uses permissions.createChat, not this metadata. Compare with tasks-tab-visible, which IS consumed by TasksNavItem.
The mock MockAgentsTabVisible in entities.ts:565 is still false while production is true. The entire pipeline (backend goroutine, HTML template field, metadata registration, stale mock) is dead weight.
11 of 17 reviewers flagged this independently. "Delete AgentsTabVisible from htmlState, remove the goroutine in populateHTMLState, remove the agents-tab-visible registration from useEmbeddedMetadata.ts, and remove the mock. This eliminates the class of bug where a stale or mismatched embedded boolean silently controls UI visibility without any consumer." (Meruem)
🤖
There was a problem hiding this comment.
DEREM-5: Done in 74140b6 — removed the entire pipeline (htmlState field, populateHTMLState goroutine, HTML meta tag, useEmbeddedMetadata registration, MockAgentsTabVisible).
|
|
||
| if _, ok := runModel.(expChatsTUIModel); !ok { | ||
| if _, ok := runModel.(chatsTUIModel); !ok { | ||
| return xerrors.New(fmt.Sprintf("unknown model found %T (%+v)", runModel, runModel)) |
There was a problem hiding this comment.
Nit [DEREM-6] xerrors.New(fmt.Sprintf(...)) should be xerrors.Errorf(...), which is the idiomatic pattern and used everywhere else in this file.
return xerrors.Errorf("unknown model found %T (%+v)", runModel, runModel)(Leorio)
🤖
There was a problem hiding this comment.
DEREM-6: Done in eeb8201 — switched to xerrors.Errorf and dropped the now-unused fmt import.
Address Copilot review feedback on #24432: after dropping the experiment flag, the Workspaces page was fetching chatsByWorkspace for every user with workspaces, even users who can't create chats. The endpoint still runs a DB query + RBAC post-filter and the downstream UI (AgentsNavItem, chat link icon in the workspaces table) is already hidden for users without createChat, so the fetch is pure waste for them. Gate the useQuery on permissions.createChat, matching the existing canCreateChat pattern used in Navbar.tsx and AgentCreatePage.tsx.
Per coder-agents-review[bot] feedback: - DEREM-1 (P2): docs/ai-coder/agents/getting-started.md said the Chats API was 'experimental'; chats-api.md was already updated to 'beta' in the rebase. Sync getting-started.md so both docs agree. - DEREM-3 (P3): coderd/mcp_test.go's mcpDeploymentValues docstring still claimed it returned values 'with the agents experiment enabled'. Drop the stale claim and simplify the now-trivial function body. - DEREM-4 (P3): docs/ai-coder/agents/early-access.md heading 'Enable Coder Agents' implied an enablement procedure, but the body says 'No experiment flags are required.' Rename to 'Set up Coder Agents'. - DEREM-6 (Nit): cli/agents.go used xerrors.New(fmt.Sprintf(...)); switch to xerrors.Errorf(...) to match the rest of the file. Drop the now-unused fmt import.
Per PR #24432 review feedback (DEREM-5): the AgentsTabVisible metadata pipeline (Go htmlState field, populateHTMLState goroutine, HTML meta tag, useEmbeddedMetadata registration, MockAgentsTabVisible) has zero production consumers. Compare with tasks-tab-visible, which NavbarView.tsx:220 actually reads. AgentsNavItem already gates on permissions.createChat. The whole pipeline is now dead weight, with a stale mock value (false) that disagrees with the unconditional production value (true). Removing the entire chain rather than just toggling the mock eliminates the class of bug where a stale embedded boolean silently controls UI visibility without any consumer. - site/site.go: drop the AgentsTabVisible field and its goroutine. - site/index.html: drop the agents-tab-visible meta tag. - site/src/hooks/useEmbeddedMetadata.ts: drop the registration. - site/src/hooks/useEmbeddedMetadata.test.ts: drop the test cases. - site/src/testHelpers/entities.ts: drop MockAgentsTabVisible.
deansheather
force-pushed
the
remove-agents-experiment-flag
branch
from
d37c1ee to
74140b6
Compare
|
Thanks for the review! Addressed in the latest force-push:
Also rebased onto latest
|
|
/coder-agents-review |
There was a problem hiding this comment.
Round 2. All 7 R1 findings (DEREM-1 through DEREM-7) confirmed fixed. The AgentsTabVisible dead code pipeline (DEREM-5) was cleaned thoroughly: zero references remain across Go, TypeScript, HTML, and test mocks. The docs fixes (DEREM-1, DEREM-4), stale comment updates (DEREM-3), PR description correction (DEREM-2), and xerrors.Errorf fix (DEREM-6) are all verified. DEREM-7 was resolved by the rebase absorbing PR #24438.
Severity count (new this round): 1 P3, 1 Nit.
The P3 is three stale Experimental(agents) comments in coderd/coderd.go that survived the cleanup. The PR already removed one identical pattern at line 1156 in R1; three siblings at lines 771, 2155, and 2157 were missed. Line 771 also has a latent emdash (U+2014) that will break make lint/emdash the moment anyone edits that line.
"The line 771 comment also contains an emdash, which violates the project's make lint/emdash rule. It is not caught today because the emdash check runs in changed mode and this line was not in the diff, but editing the line for any reason will trigger the lint failure." (Pariston)
coderd/coderd.go:771
P3 [DEREM-8] Three stale Experimental(agents) comments survive the cleanup:
- Line 771:
{ // Experimental: agents — chat daemon and git sync worker initialization. - Line 2155:
// Experimental(agents): chatDaemon handles background processing... - Line 2157:
// Experimental(agents): gitSyncWorker refreshes stale chat diff statuses...
This PR already removed the identical pattern at line 1156 (// Experimental(agents): chat API routes gated by ExperimentAgents.). These three siblings were missed.
Line 771 also contains an emdash (U+2014), which is a latent make lint/emdash failure. It won't fire today because the emdash check runs in changed mode, but editing this line for any reason will trigger it.
Fix: reword all three to drop the Experimental prefix and the emdash. For example: { // Chat daemon and git sync worker initialization.
(Pariston)
🤖
docs/ai-coder/agents/getting-started.md:3
Nit [DEREM-9] "enabling Coder Agents" is stale since the enablement step was removed. The remaining steps are configuring a provider, granting a role, and starting an agent.
Consider changing to "setting up" or "configuring" to match the actual flow. The early-access.md heading was already updated from "Enable" to "Set up" for the same reason (DEREM-4).
(Leorio)
🤖
🤖 This review was automatically generated with Coder Agents.
Per coder-agents-review[bot] round 2 feedback:
- DEREM-8 (P3): Drop the 'Experimental(agents)' / 'Experimental:
agents' prefix from three stale comments in coderd/coderd.go that
the round-1 cleanup missed:
- line 771: '{ // Experimental: agents [emdash] chat daemon and...'
- line 2155: '// Experimental(agents): chatDaemon...'
- line 2157: '// Experimental(agents): gitSyncWorker...'
Line 771 also contained a literal emdash that would have tripped
make lint/emdash the next time anyone edited the line; the new
wording uses an ASCII period instead.
- DEREM-9 (Nit): Reword 'enabling Coder Agents' in
docs/ai-coder/agents/getting-started.md to 'setting up Coder
Agents'. The enablement step was removed in the round-1 cleanup,
matching the same change made to early-access.md's heading
(DEREM-4).
|
Round 2 addressed in 63a27a1:
|
|
/coder-agents-review |
There was a problem hiding this comment.
Round 3. All 9 findings (DEREM-1 through DEREM-9) confirmed fixed across three rounds. No new findings from Netero or the panel (7 reviewers).
Clean experiment removal. The ExperimentAgents flag is fully excised, the AgentsTabVisible dead code pipeline is gone, docs are consistent ("beta" everywhere), stale comments are updated, the CLI graduated from coder exp agents to coder agents, and the emdash lint exclusion for golden files is correctly scoped.
"The problem is correctly understood. The solution is proportional. The fix is at the right level." (Pariston)
🤖 This review was automatically generated with Coder Agents.
Remove the ExperimentAgents feature flag so the Agents feature is always available without requiring --experiments=agents. The feature is now in beta. Backend: - Remove RequireExperimentWithDevBypass middleware from chat and MCP server routes in coderd/coderd.go - Always include AgentsAccessRole in assignable site roles - Always set AgentsTabVisible to true in site/site.go - Make blob: CSP img-src addition unconditional - Remove ExperimentAgents constant, DisplayName case, and ExperimentsKnown entry from codersdk/deployment.go Frontend: - Remove experiment check from AgentsNavItem (show when user has canCreateChat permission) - Remove agentsEnabled experiment check from WorkspacesPage - Add FeatureStageBadge with contentType=beta next to the Coder logo in the Agents sidebar (desktop and mobile) Docs: - Remove experiment flag setup instructions from early-access.md and getting-started.md - Update chats-api.md to say beta instead of experimental Tests: - Remove ExperimentAgents setup from all test files Existing deployments that still pass --experiments=agents will get a harmless 'ignoring unknown experiment' warning on startup.
Move the agents TUI command from the experimental command tree to the core subcommands, so it is available as "coder agents" instead of "coder exp agents". - Move r.agentsCommand() from AGPLExperimental() to CoreSubcommands() in cli/root.go (alphabetically sorted at the top). - Drop the "agent" alias to avoid colliding with the hidden "coder agent" command used by the workspace agent binary. - Remove leftover ExperimentAgents references introduced during rebase on main in cli/exp_agents_e2e_helpers_test.go and coderd/x/chatd/chatd_test.go. - make gen regenerated docs/reference/cli/agents.md, the cli index, and the --help golden file. The files are still named exp_agents_*.go for now; renaming them and internal identifiers (expChatsTUIModel, etc.) is a mechanical cleanup that can follow in a separate PR.
The agents command graduated from the experimental command tree in the previous commit; update startExpAgentsSession so it invokes the new top-level command path.
Follow-up to "remove agents experiment flag and mark feature as beta": - coderd/x/chatd/integration_test.go, coderd/mcp_test.go: drop stale 'with the agents experiment' wording from comments that no longer configure the experiment. - docs/manifest.json: remove 'experimental' from the Chats API sidebar description now that docs/ai-coder/agents/chats-api.md calls the API 'beta'.
…entifiers
The 'coder agents' TUI graduated out of the experimental command tree,
but the implementation files (cli/exp_agents_*.go) and internal symbols
still carried 'exp' prefixes. This is the mechanical cleanup promised
in the preceding "graduate coder exp agents" commit.
- Rename cli/exp_agents*.go -> cli/agents*.go (15 files, including the
newly added cli/exp_agents_diff{,_test}.go from #24512).
- Rename internal identifiers:
expChatsTUIModel -> chatsTUIModel
newExpChatsTUIModel -> newChatsTUIModel
expAgentsPtr -> agentsPtr
expAgentsSession -> agentsSession
setupExpAgentsBackend -> setupAgentsBackend
startExpAgentsSession -> startAgentsSession
TestExpAgents* -> TestAgents*
Two CI-driven cleanups bundled here so the rename commit passes
pre-commit under the new repo-wide 'no emdash/endash' lint rule:
- plans/CODAGT-156-remove-experimental-flags.md: normalize em-dashes
to colons/hyphens. The plan doc is new-vs-main so its lines are
scanned by scripts/check_emdash.sh.
- scripts/check_emdash.sh: exclude cli/testdata/*.golden and
enterprise/cli/testdata/*.golden from the scan. serpent's CLI help
renderer emits emdash borders around the 'Run coder --help' footer,
so every generated help golden file contains them. The newly added
coder_agents_--help.golden tripped the check; the exclusion matches
what's already true for all existing goldens.
Left intentionally untouched:
- expClient (local variable of type *codersdk.ExperimentalClient; the
experimental API client is a separate concept from the graduated CLI).
- coderd/exp_chats*.go and enterprise/coderd/exp_chats_test.go (the
chat API routes still live under /api/experimental/chats).
- Other cli/exp_*.go files (still-experimental commands like chat,
mcp, rpty, prompt, sync).
Address Copilot review feedback on #24432: after dropping the experiment flag, the Workspaces page was fetching chatsByWorkspace for every user with workspaces, even users who can't create chats. The endpoint still runs a DB query + RBAC post-filter and the downstream UI (AgentsNavItem, chat link icon in the workspaces table) is already hidden for users without createChat, so the fetch is pure waste for them. Gate the useQuery on permissions.createChat, matching the existing canCreateChat pattern used in Navbar.tsx and AgentCreatePage.tsx.
Per coder-agents-review[bot] feedback: - DEREM-1 (P2): docs/ai-coder/agents/getting-started.md said the Chats API was 'experimental'; chats-api.md was already updated to 'beta' in the rebase. Sync getting-started.md so both docs agree. - DEREM-3 (P3): coderd/mcp_test.go's mcpDeploymentValues docstring still claimed it returned values 'with the agents experiment enabled'. Drop the stale claim and simplify the now-trivial function body. - DEREM-4 (P3): docs/ai-coder/agents/early-access.md heading 'Enable Coder Agents' implied an enablement procedure, but the body says 'No experiment flags are required.' Rename to 'Set up Coder Agents'. - DEREM-6 (Nit): cli/agents.go used xerrors.New(fmt.Sprintf(...)); switch to xerrors.Errorf(...) to match the rest of the file. Drop the now-unused fmt import.
Per PR #24432 review feedback (DEREM-5): the AgentsTabVisible metadata pipeline (Go htmlState field, populateHTMLState goroutine, HTML meta tag, useEmbeddedMetadata registration, MockAgentsTabVisible) has zero production consumers. Compare with tasks-tab-visible, which NavbarView.tsx:220 actually reads. AgentsNavItem already gates on permissions.createChat. The whole pipeline is now dead weight, with a stale mock value (false) that disagrees with the unconditional production value (true). Removing the entire chain rather than just toggling the mock eliminates the class of bug where a stale embedded boolean silently controls UI visibility without any consumer. - site/site.go: drop the AgentsTabVisible field and its goroutine. - site/index.html: drop the agents-tab-visible meta tag. - site/src/hooks/useEmbeddedMetadata.ts: drop the registration. - site/src/hooks/useEmbeddedMetadata.test.ts: drop the test cases. - site/src/testHelpers/entities.ts: drop MockAgentsTabVisible.
Per coder-agents-review[bot] round 2 feedback:
- DEREM-8 (P3): Drop the 'Experimental(agents)' / 'Experimental:
agents' prefix from three stale comments in coderd/coderd.go that
the round-1 cleanup missed:
- line 771: '{ // Experimental: agents [emdash] chat daemon and...'
- line 2155: '// Experimental(agents): chatDaemon...'
- line 2157: '// Experimental(agents): gitSyncWorker...'
Line 771 also contained a literal emdash that would have tripped
make lint/emdash the next time anyone edited the line; the new
wording uses an ASCII period instead.
- DEREM-9 (Nit): Reword 'enabling Coder Agents' in
docs/ai-coder/agents/getting-started.md to 'setting up Coder
Agents'. The enablement step was removed in the round-1 cleanup,
matching the same change made to early-access.md's heading
(DEREM-4).
deansheather
force-pushed
the
remove-agents-experiment-flag
branch
from
63a27a1 to
9d48184
Compare
|
/coder-agents-review |
There was a problem hiding this comment.
Round 4. Rebase-only round (new base 07e86ae). No new findings from Netero or the panel (6 reviewers). All 9 prior findings remain fixed. The rebase introduced no regressions.
"Pulled every thread I could find across this diff. The removal is thorough and consistent." (Hisoka)
🤖 This review was automatically generated with Coder Agents.
The CODAGT-156 implementation plan was a working document that shouldn't have been committed to the repo. Implementation is complete; the plan no longer needs to live alongside the source.
Remove the
ExperimentAgentsfeature flag so the Agents feature isalways available without requiring
--experiments=agents. The featureis now in beta.
Existing deployments that still pass
--experiments=agentswill get aharmless "ignoring unknown experiment" warning on startup.
Changes
Backend:
RequireExperimentWithDevBypassmiddleware from chat and MCP server routesAgentsAccessRolein assignable site roles (later refactored to org-scoped on main; rebase keeps that)AgentsTabVisible = true, then drop the entire deadAgentsTabVisiblemetadata pipeline (Go htmlState field, populateHTMLState goroutine, HTML meta tag, useEmbeddedMetadata registration, mock); no production consumer reads it.AgentsNavItemalready gates onpermissions.createChat.blob:CSPimg-srcaddition unconditionalExperimentAgentsconstant,DisplayNamecase, andExperimentsKnownentryCLI:
coder exp agentstocoder agents(moved fromAGPLExperimental()toCoreSubcommands())agentalias so it does not collide with the hidden workspace-agent commandcli/exp_agents_*.go->cli/agents_*.goand internal identifiers (expChatsTUIModel->chatsTUIModel,newExpChatsTUIModel->newChatsTUIModel,setupExpAgentsBackend->setupAgentsBackend,startExpAgentsSession->startAgentsSession,expAgentsPtr->agentsPtr,expAgentsSession->agentsSession,TestExpAgents*->TestAgents*).expClient(the*codersdk.ExperimentalClientlocal) is kept;coderd/exp_chats*.goand other still-experimentalcli/exp_*.gocommands are intentionally untouched.Frontend:
AgentsNavItem- render whencanCreateChatis trueagentsEnabledexperiment check fromWorkspacesPage, then gatechatsByWorkspaceonpermissions.createChatso users without chat access don't trigger the per-page DB query (Copilot review feedback)FeatureStageBadge(beta) next to the Coder logo in the Agents sidebar (desktop + mobile)Docs:
early-access.mdandgetting-started.md(and renameearly-access.md's "Enable Coder Agents" heading to "Set up Coder Agents", since there is no enablement step left)chats-api.mdandgetting-started.md's Chats API note to say "beta" instead of "experimental"docs/manifest.json: drop "experimental" from the Chats API sidebar descriptionmake genregenerateddocs/reference/cli/agents.mdand the CLI indexscripts/check_emdash.sh: excludecli/testdata/*.goldenandenterprise/cli/testdata/*.goldenfrom the new repo-wide emdash lint, since serpent emits emdash borders in every generated--helpgolden fileTests:
ExperimentAgentssetup from all test files (14 occurrences across 7 files)coderd/x/chatd/integration_test.goandcoderd/mcp_test.go