fix(opencode): stabilize Agent Builder /agents flow - #314
Conversation
|
The following comment was made by an LLM, it may be inaccurate: |
|
@codex review |
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
- Hosted PR #314 run 28459035332 job 84341579388 failed because the Run Agent Swarm e2e tests step timed out after 10 minutes. - The uploaded JUnit report for that run shows 66 tests, 65 passed, 1 skipped, 0 failed, and 612.534s elapsed, so the failure is CI budget rather than a test assertion failure. - git blame traces the timeout to fork-owned commit 5cc1a69 (ci: run Agent Swarm TUI e2e tests); upstream/dev has no Agent Swarm e2e job. - PR #314 adds Agent Builder mode and Plan handoff coverage, so raise only the Agent Swarm e2e step timeout to 20 minutes and keep the test command unchanged.
Refs #314. Wait for the recovered Build follow-up request after the startup Plan approval point while still asserting that the recovered prompt text reaches the native request. This keeps the e2e proof stable without weakening the Plan-to-Build routing assertion.
|
@codex review |
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
- Hosted PR #314 run 28459035332 job 84341579388 failed because the Run Agent Swarm e2e tests step timed out after 10 minutes. - The uploaded JUnit report for that run shows 66 tests, 65 passed, 1 skipped, 0 failed, and 612.534s elapsed, so the failure is CI budget rather than a test assertion failure. - git blame traces the timeout to fork-owned commit 5cc1a69 (ci: run Agent Swarm TUI e2e tests); upstream/dev has no Agent Swarm e2e job. - PR #314 adds Agent Builder mode and Plan handoff coverage, so raise only the Agent Swarm e2e step timeout to 20 minutes and keep the test command unchanged.
1e45bb1 to
63678b1
Compare
Refs #314. Wait for the recovered Build follow-up request after the startup Plan approval point while still asserting that the recovered prompt text reaches the native request. This keeps the e2e proof stable without weakening the Plan-to-Build routing assertion.
707d7f7 to
5d334d3
Compare
- clean stale mode command assertions and docs wording - add focused TUI E2E for server-free Build/Plan, compact, reopen, and dead Run fallback - keep Tab scoped to agent switching inside the current mode
- hydrate pending TUI questions from question.list during sync - cover startup Plan handoff when the live question event is missed
- Recover missed pending Plan approval questions on active session reopen - Keep question recovery best-effort and workspace-routed - Cover repeated-arrow approval, declined approval, stale question sync, and missed-event recovery
- Hosted PR #314 run 28459035332 job 84341579388 failed because the Run Agent Swarm e2e tests step timed out after 10 minutes. - The uploaded JUnit report for that run shows 66 tests, 65 passed, 1 skipped, 0 failed, and 612.534s elapsed, so the failure is CI budget rather than a test assertion failure. - git blame traces the timeout to fork-owned commit 5cc1a69 (ci: run Agent Swarm TUI e2e tests); upstream/dev has no Agent Swarm e2e job. - PR #314 adds Agent Builder mode and Plan handoff coverage, so raise only the Agent Swarm e2e step timeout to 20 minutes and keep the test command unchanged.
Refs #314. Wait for the recovered Build follow-up request after the startup Plan approval point while still asserting that the recovered prompt text reaches the native request. This keeps the e2e proof stable without weakening the Plan-to-Build routing assertion.
- Move Plan, Build, and Run selection into /agents and remove /modes - Keep Tab on Build/Plan outside Run and target switching inside Run - Remove production-only question test hook and preserve pending Plan approval recovery - Update user flows, QA coverage, README, and fork changelog
- open Build when a detected local Agent Swarm project has an entry import/startup failure after the environment is ready - keep the same project available for Run after the user fixes it in Build - cover the fallback with source and binary terminal E2E plus launcher tests
aaf0264 to
8547ed6
Compare
- strip stale launch-time Agency Swarm routing after external /connect from Build fallback - keep prepared local Run config authoritative when returning to Run from Build - extend focused terminal E2E assertions for stale local-agency config
8547ed6 to
bb4ab88
Compare
- keep startup Build slash commands usable with prefilled repair text - keep TUI input ownership when switching back to local Run after dependency refresh - document and cover the real terminal recovery cases
e8b23d5 to
53ee119
Compare
- keep startup fallback prompt generic for any project startup error - shorten the prefilled Build prompt while preserving the real error - update focused assertions and flow docs
|
@codex review |
|
Codex Review: Didn't find any major issues. More of your lovely PRs please. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
- show a visible starting state while /agents Run prepares the local swarm - cover the slow Build fallback to Run path in focused TUI E2E - document the user-visible wait state in USER_FLOWS and FORK_CHANGELOG
Issue for this PR
Closes #313
Type of change
What does this PR do?
This PR makes
/agentsthe Agent Builder control point for Plan, Build, Run, and Run-mode swarm targets./agentsand removes the separate/modesdialog.Starting...state while/agentsstarts or reconnects local Run, so the TUI does not look frozen.OPENCODE_CONFIG_CONTENTandenabled_providers/disabled_providersstate from overriding repaired local Run config, while preserving explicit/connectchoices from the Build fallback.USER_FLOWS.md,FORK_CHANGELOG.md, QA coverage notes, terminal E2E coverage, sync/npx tests, and the Agent Swarm E2E CI timeout for the larger suite.How did you verify your code works?
The diff adds or updates coverage for
/agentsmode rows, native Plan/Build routing, Run server-backed routing, visible Run startup feedback, Tab behavior, Plan handoff approval/decline/recovery, startup/import failure fallback into Build, same-project Run after Build repair, stale env config and provider filter handling, dead Run server repair, and pending-question workspace routing.Latest local proof on head
00e474707:bun test --timeout 240000 --max-concurrency=1 ../../e2e/agent-swarm-tui/terminal-tui.test.ts -t "launcher opens Build after startup failure and restarts repaired local Run from stale env config"passed.AGENTSWARM_TUI_E2E_BINARY_PATH=/Users/nick2/.codex/worktrees/pr314-agent-builder-final/agentswarm-cli/packages/opencode/dist/agentswarm-cli-darwin-arm64/bin/agentswarm.bun test test/cli/tui/dialog-agent.test.tsx --timeout 30000passed.bun typecheckfrompackages/opencodepassed.bun x prettier --check FORK_CHANGELOG.md USER_FLOWS.md e2e/agent-swarm-tui/terminal-tui.test.ts packages/opencode/src/cli/cmd/tui/component/dialog-agent.tsxpassed.git diff --checkpassed.bun run buildrebuilt the Darwin arm64 binary;agentswarm --versionreturned1.4.26.bun turbo typecheckpassed.origin/devreported no actionable correctness issues.Hosted checks remain the live source of truth for CI status on each PR head.
Screenshots / recordings
N/A. This is terminal TUI behavior covered by terminal E2E tests.
Checklist