🎯 feat: Tolerant Workspace Edits with Host-Rendered Conflict Diagnostics by danny-avila · Pull Request #16520 · LibreChat-AI/LibreChat

danny-avila · 2026-09-29T18:39:11Z

Summary

edit_file in an attached (BYOM) workspace is the strictest of LibreChat's four edit_file variants, and the least informative when it fails. Over 10 days on one deployment, 125 of 2,935 calls (4.3%) failed with EDIT_CONFLICT, and 94 of those were multi-edit batches. Each failure reached the model the same way: the worker's JSON body with its newlines escaped, followed by a suffix saying "did not match exactly once… Re-read the file and retry". That suffix was appended to every 409, including a file that changed while the edit was in flight. The model could not tell which edit failed or why, so it retried blindly or re-read the whole file.

LibreChat-AI/code-interpreter#271 gives the worker per-edit diagnostics and two negotiated edit features: tolerant_match (whitespace-tolerant fallback) and replace_all. This PR is the LibreChat side:

Negotiation: it reads the worker's editFileFeatures from its status and carries them into the selected workspace.
Tolerant matching: it sends matching: 'tolerant' on edits and previews whenever the worker negotiated tolerant_match. So attached edits now tolerate trailing whitespace, CRLF and indentation shifts, as skill and sandbox edits already do. Operators can require exact matching with configSchema.edits.tolerantMatching: false. A match must still be unique unless replace_all is set.
replace_all: the edit_file schema gets replace_all, top level and per edit, for every variant.
- On attached workspaces it is sent only when the worker negotiated replace_all. Otherwise it is refused before dispatch with a clear message.
- Skill-file and sandbox edits support it in LibreChat's own matcher.
Conflict messages: a 409 never forwards worker text. parseEditConflict accepts only the worker's fixed diagnostic grammar and keeps bounded facts: which edits failed, not-found or ambiguous, counts, up to five line numbers, the strategy, and known hint kinds. Quoted file excerpts are dropped. formatEditConflict renders LibreChat's own sentences from those facts. A file that changed mid-edit gets its own message; anything outside the grammar gets generic retry guidance.
Bounded replace_all: exact matches are counted without retaining positions and rebuilt with split/join after a projected-size check. The whitespace-tolerant fallback stops collecting at 10,000 ranges.
Success text: it names any edit that matched by a non-exact strategy or replaced several locations. Edits that all matched exactly keep the current wording.
Tool descriptions: they describe replace_all, the whitespace-tolerant fallback on current workers, and the new failure reporting.

With an older worker or Code API, none of these fields are sent, and requests, results and messages stay exactly as they are today.

Depends on LibreChat-AI/code-interpreter#271 for the new worker behavior. Merging this first is safe: without the worker change, no features are advertised and nothing changes.

How it works

Code API worker status ── editFileFeatures ──> getCodeBridgeWorkerStatus (known names only)
  -> resolveCodeExecutionWorkspaceContext: codeWorkspace.editFileFeatures
    -> handleAttachedWorkspaceEditFileCall
         replace_all requested but not negotiated   -> refuse before dispatch
         tolerant_match negotiated (unless disabled)  -> matching: 'tolerant' (edit + protected preview)
         409 EDIT_CONFLICT                            -> parse facts -> host-rendered message / changed-file / generic
         result.matches                               -> per-edit strategy and occurrence notes

The transport (code/workspace.ts) accepts matching and per-edit replaceAll in requests.
Results must carry matches exactly when the request opted in: exact strategies only for non-tolerant requests, and more than one occurrence only for replaceAll edits. This mirrors the worker's own validation.
WorkspaceToolHttpError.editConflict exposes the worker's EDIT_CONFLICT text.
The feature and strategy names live in code/edits.ts, so bridge.ts and workspace.ts share them without an import cycle.

Type of change

Feature
Bug fix

Testing

Tested environments/configuration:

Node 24.16.0; unit and integration tests only. The live BYOM path needs code-interpreter#271 deployed to a worker.

Automated tests:

packages/api: npx jest src/agents/handlers.spec.ts src/code/workspace.spec.ts src/code/bridge.spec.ts src/code/capabilities.spec.ts src/agents/tools.spec.ts src/code/workspace.lanes.spec.ts, 489 passed. New coverage:
- opting into tolerant matching and replace_all only when negotiated;
- refusing replace_all before dispatch on a legacy worker;
- diagnostic, changed-file and legacy 409 messages;
- skill-file replace_all;
- request and result validation for matching, replaceAll and matches;
- parsing negotiated features (unknown names dropped) and carrying them into the workspace context.
api: npx jest server/services/Files/Code/process.spec.js, 166 passed, including forwarding matching and replaceAll on edits and previews.
npx tsc --noEmit in packages/api; eslint on the changed files; the pre-commit circular-dependency check.

Screenshots / recordings

Not applicable. No UI change; the tool result text changes only when a worker reports non-exact matches.

… Through

danny-avila · 2026-09-29T18:39:18Z

@codex review

chatgpt-codex-connector · 2026-09-29T18:39:39Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-29T20:25:41.494447Z	`db43838`	Manual request

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f9220dd621

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".

…om Parsed Facts, Bound replace_all

danny-avila · 2026-09-29T19:57:24Z

@codex review the latest head

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6302f23157

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".

… replace_all

danny-avila · 2026-09-29T20:05:37Z

@codex review the latest head

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ba17a0534c

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".

…Both Matching Modes

danny-avila · 2026-09-29T20:13:08Z

@codex review the latest head

chatgpt-codex-connector

💡 Codex Review

LibreChat/packages/api/src/agents/tools.ts

Line 958 in a456ff0

parameters: attachedFileAuthoringParameters(CODE_EDIT_FILE_PARAMETERS, false),

Hide replace_all from legacy attached workers

When the selected attached worker does not advertise replace_all, this definition still exposes and recommends the option because the attached tool unconditionally reuses CODE_EDIT_FILE_PARAMETERS; any model call that follows that schema is then guaranteed to fail at the capability guard in handleAttachedWorkspaceEditFileCall. Build the attached definition from codeWorkspace.editFileFeatures (or explicitly describe the option as unavailable) so mixed-version deployments do not advertise an operation they reject.

AGENTS.md reference: AGENTS.md:L44-L47

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".

…of CJS

danny-avila · 2026-09-29T20:22:31Z

@codex review the latest head, final review. no further reviews will be requested, last chance

chatgpt-codex-connector · 2026-09-29T20:25:39Z

Codex Review: Didn't find any major issues. Keep it up!

Reviewed commit: db43838471

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".

…ics (LibreChat-AI#16520) * 🎯 feat: Opt Into Tolerant Workspace Edits and Pass Worker Diagnostics Through * 🛡️ fix: Gate Tolerant Edits Behind Config, Render Worker Conflicts From Parsed Facts, Bound replace_all * ⚖️ fix: Default Workspace Edits to Tolerant Matching and Stream Exact replace_all * 🧱 fix: Build Exact replace_all Output in Bounded Chunks and Describe Both Matching Modes * 🧹 fix: Keep Worker Conflict Bodies Out of Logs and Edit Matching Out of CJS

🎯 feat: Opt Into Tolerant Workspace Edits and Pass Worker Diagnostics…

f9220dd

… Through

chatgpt-codex-connector Bot reviewed Sep 29, 2026

View reviewed changes

Comment thread packages/api/src/agents/handlers.ts Outdated

Comment thread packages/api/src/agents/handlers.ts Outdated

Comment thread packages/api/src/agents/handlers.ts

🛡️ fix: Gate Tolerant Edits Behind Config, Render Worker Conflicts Fr…

6302f23

…om Parsed Facts, Bound replace_all

chatgpt-codex-connector Bot reviewed Sep 29, 2026

View reviewed changes

Comment thread packages/api/src/agents/handlers.ts

Comment thread packages/api/src/agents/handlers.ts

⚖️ fix: Default Workspace Edits to Tolerant Matching and Stream Exact…

ba17a05

… replace_all

chatgpt-codex-connector Bot reviewed Sep 29, 2026

View reviewed changes

Comment thread packages/api/src/agents/handlers.ts Outdated

Comment thread packages/api/src/agents/tools.ts Outdated

🧱 fix: Build Exact replace_all Output in Bounded Chunks and Describe …

a456ff0

…Both Matching Modes

chatgpt-codex-connector Bot reviewed Sep 29, 2026

View reviewed changes

Comment thread packages/api/src/agents/handlers.ts Outdated

Comment thread api/server/services/Files/Code/process.js Outdated

🧹 fix: Keep Worker Conflict Bodies Out of Logs and Edit Matching Out …

db43838

…of CJS

danny-avila changed the title ~~🎯 feat: Opt Into Tolerant Workspace Edits and Pass Worker Diagnostics Through~~ 🎯 feat: Tolerant Workspace Edits with Host-Rendered Conflict Diagnostics Sep 29, 2026

danny-avila merged commit 3e19b18 into dev Sep 29, 2026
44 checks passed

danny-avila deleted the danny-avila/workspace-edit-matching branch September 29, 2026 20:55

lia-by-librechat Bot mentioned this pull request Sep 30, 2026

🪮 perf: Match Tolerant Edit Lines Without Rebuilding Windows #16562

Merged

8 of 10 tasks

Uh oh!

Conversation

danny-avila commented Sep 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

How it works

Type of change

Testing

Screenshots / recordings

Uh oh!

danny-avila commented Sep 29, 2026

Uh oh!

chatgpt-codex-connector Bot commented Sep 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

danny-avila commented Sep 29, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

danny-avila commented Sep 29, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

danny-avila commented Sep 29, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

danny-avila commented Sep 29, 2026

Uh oh!

chatgpt-codex-connector Bot commented Sep 29, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

danny-avila commented Sep 29, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Sep 29, 2026 •

edited

Loading