🤖 feat: interactive harness init approval flow #1807
Conversation
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
446b377 to
43951c5
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 446b377437
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
|
Follow-ups pushed:
Note: Chromatic "UI Review" / "UI Tests" are still pending (require baseline acceptance). |
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
7867a5e to
922f9e5
Compare
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
922f9e5 to
497091a
Compare
|
@codex review Changes:
Validation:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 497091ad40
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
497091a to
a7bb8f0
Compare
|
@codex review Addressed “Preserve harness state when stat fails”:
Validation:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 98297a50f0
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
|
@codex review Addressed “Guard harness load errors to avoid stuck loop state”:
Validation:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dfbe15ec01
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
dfbe15e to
82d1f5d
Compare
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
82d1f5d to
d67aca6
Compare
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d459251934
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
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".
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1a11933565
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
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".
| const prefix = workspaceName.trim().length > 0 ? workspaceName.trim() : "workspace"; | ||
|
|
||
| return { | ||
| configPath: joinForRuntime(runtimeConfig, workspacePath, HARNESS_DIR, `${prefix}.jsonc`), | ||
| progressPath: joinForRuntime( |
There was a problem hiding this comment.
For in-place workspaces metadata.name can be an absolute path (see comment in getRuntimeAndWorkspacePath). Here prefix is taken directly from workspaceName, then passed to path.join via joinForRuntime. If workspaceName starts with /, path.join treats it as absolute and discards workspacePath/.mux/harness, so the harness config/progress files resolve to something like /Users/me/project.jsonc instead of .mux/harness/.... This breaks harness detection and can write files outside the repo. Consider normalizing workspaceName to a safe relative basename (or stripping leading slashes) before joining.
Useful? React with 👍 / 👎.
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
1a11933 to
010bbdf
Compare
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
010bbdf to
6f8a76e
Compare
Adds workspace-local harness config (checklist + gates) and an opt-in Ralph loop runner. - Backend services: WorkspaceHarnessService, GateRunnerService, GitCheckpointService, LoopRunnerService - ORPC: workspace.harness + workspace.loop endpoints - UI: RightSidebar Harness tab + command palette actions for gates/checkpoint/loop Signed-off-by: Thomas Kosiewski <tk@coder.com> --- _Generated with • Model: openai:gpt-5.2 • Thinking: high • Cost: 0.17_ Change-Id: I99428a620b0bd65e9b9a2bb9023b9dd9e0843bc1
Change-Id: I15d81ab1136b5437df531ba6cb3e23cf84c321a0 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: Ide9e2ac1fa93252310350441843ae4d7eaa0ad25 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I0f684cca69decbe2756577ec54c321ea0e13b182 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: Iebbcc21aaa8a919be5e1217c0d44b6cee070d782 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Include the workspace plan file path in harness reset/loop bearings summaries. Signed-off-by: Thomas Kosiewski <tk@coder.com> --- _Generated with `mux` • Model: `openai:gpt-5.2` • Thinking: `xhigh` • Cost: $47.81_ Change-Id: I89cf61ac2e147042882b58297d0bf9dde49835fd
Change-Id: Icf5963d92a65300117de0c264272f8ca3952c4e0 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: Ie569d9a08cf122c8d7dce626003d1620a6e37bf9 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I88bf5879b908141790c6119d99f93983071a6b5e Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: Ic9c7e77915dcf5662b2cf767f93202c928d13c91 Signed-off-by: Thomas Kosiewski <tk@coder.com>
- Add lightweight workspace.harness.exists endpoint - Remove legacy harness filename support - Conditionally show/remove Harness tab in right sidebar and persisted layout Signed-off-by: Thomas Kosiewski <tk@coder.com> --- _Generated with `mux` • Model: `openai:gpt-5.2` • Thinking: `xhigh` • Cost: $74.55_ Change-Id: Icfd5f621eefc533c855a202e8f65739b3194791a
Change-Id: I19674059830f6d2a447a96dfef6cebb64c65143e Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I5d1a99f30bf1c8997053eab48894cac50f4d7317 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I44047b8a001042fec6b21d46baf7502a231540ce Signed-off-by: Thomas Kosiewski <tk@coder.com>
Replace planFileOnly with mode/agentId/allowedEditPaths and update tests + docs. Change-Id: If00ef62e3877084ef650822af645e254371c9d3b Signed-off-by: Thomas Kosiewski <tk@coder.com>
Ensure harness parent dirs exist before writing config/journal and update journal hint + tests. Change-Id: Id05475a1414d0e5b2a26efa3b63c49f619390b62 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Include nested harness glob in harness-init edit allowlist and update docs/tests. Change-Id: I5385802f1711dec7af09cc4526d8abc38c25df09 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I1f40d2ed2df5614ef7513e8a2efc61ac13fc7e73 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Change-Id: I8c417c433ed340807524e11609d2c4e39cdf07bb Signed-off-by: Thomas Kosiewski <tk@coder.com>
- Require Harness Init to delegate exploration to explore subagents\n- Inject <harness_output_path> derived from workspace name\n- Remove harness-from-plan agent + startFromPlan RPC Change-Id: Ie2b72ffc3bd7705c08d1ce861e73093518a14078 Signed-off-by: Thomas Kosiewski <tk@coder.com>
ThomasK33
force-pushed
the
mux-agents-gmwq
branch
from
6f8a76e to
9f4d0a2
Compare
1 participant
Adds an interactive Harness-from-Plan workflow so Ralph loop runs with an explicit, user-reviewed harness.
What changed
harness-initagent (repo-aware, interactive harness authoring) with UI styling.harness-initedits to.mux/harness/*.jsoncvia tool-layerallowedEditPathsenforcement.harness-initsub-agent spawning to read-onlyexploretasks.harness-initand request a harness proposal.propose_harnesstool + UI card with Approve & Start to start the loop in Exec mode.Validation
make static-check📋 Implementation Plan
Interactive Harness-from-Plan (repo-aware + chat-first approval)
Goals
rg, inspect CI config) before proposing checklist + gates.exploresub-agents to answer: “what parts of the repo are affected?” and “what gates/commands exist here?”Recommended approach — True inline “Harness Mode” (hidden harness-init agent)
Net LoC (product code): ~400–650
1) Add a hidden harness-init agent (interactive, repo-aware)
harness-init) that is:ui.hidden: true(not selectable in the agent picker / command palette)subagent.runnable: true; add a denylist check in thetasktool for defense-in-depth)--color-harness-init-mode(and optional hover/alpha variants) insrc/browser/styles/globals.css.src/browser/components/AgentModePicker.tsxto show Harness Init in that color.src/browser/components/ChatInput/index.tsxso the Send button usesbg-harness-init-modewhen the active agent isharness-init.src/node/builtinAgents/harness-from-plan.mdand the newharness-initprompt to share the same guidance:Makefile,justfile,package.jsonscripts,.github/workflows/*)2) Constrain edits to harness files (but allow in-place diffs)
file_edit_*tools forharness-init, but enforce a tool-layer allowlist so it can only edit.mux/harness/*.jsonc.ToolConfigurationwithallowedEditPathsand enforce it insrc/node/services/tools/fileCommon.ts.allowedEditPathsinsrc/node/services/aiService.tswhen the active agent isharness-init..mux/harness/<workspace>.jsoncwhen harness-init starts so the agent can make small diffs without re-outputting the whole file.propose_harness, assert the working tree has no changes outside.mux/harness/*.jsoncto catch accidental edits (including viabash).3) Enable explore subagents safely
task+task_awaitforharness-init.src/node/services/tools/task.tssoharness-initcan only spawnagentId: "explore".4) Wire “Start Ralph Loop” to switch agent + send an initiating message
src/browser/components/tools/ProposePlanToolCall.tsx:api.workspace.loop.startFromPlan()call with a Plan-Mode-like transition:updatePersistedState(getAgentIdKey(workspaceId), "harness-init")api.workspace.sendMessage({ message: "Generate a Ralph harness from the current plan and propose it" })5) Add
propose_harnesstool + approval UI (mirrorspropose_plan)propose_harness:recordFileStateso the UI can detect out-of-band editsProposeHarnessToolCall.tsx):api.workspace.harness.getif it already returns enough; otherwise add a dedicatedgetHarnessContentendpoint)execworkspace.loop.startendpoint if one doesn’t already exist)6) Tests
allowedEditPathsenforcement (can edit harness; cannot edit other files)propose_harnessvalidation.Alternative (less refactor)
Option — Dedicated “Harness Review” child workspace
Net LoC: ~300–500
Generated with
mux• Model:openai:gpt-5.2• Thinking:high• Cost: $60.54