Repository navigation
Conversation
ehrvs
force-pushed
the
pr/14-codec
branch
4 times, most recently
from
October 16, 2026 20:24
5028532 to
e0185c4
Compare
Flatten (LM Studio) / DottedInbound (Ollama) codec in codex-local-models, OnceLock hooks in model-provider, client.rs and stream_events_utils.rs. router.rs/router_tests.rs reverted to rust-v0.162.1. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RPCzgwU247gMtjybR5v4iR
…lise OutputItemAdded Alias is a pure function of (ns,name) shared by tool defs and history; literal top-level names are never reverse-mapped. Document known limits. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RPCzgwU247gMtjybR5v4iR
Ollama -> DottedInbound; LM Studio and llama.cpp -> Flatten (llama.cpp namespace handling is unverified; flatten is always safe). Action::Run carries the engine so the single registration point in main covers exec, tui and resume. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RPCzgwU247gMtjybR5v4iR
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Collaborator
Author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem: LM Studio silently drops
namespacetools (MCP servers, multi-agent) and top-levelcustomtools (freeform apply_patch). It answers 200 with an empty tool list. Ollama returns namespaced calls as dottedns.toolnames. H8 patched upstreamrouter.rsto cope with the latter.Fix: H9, a local wire codec that is registered only by codex-local.
client.rs): namespace and custom tools become plainfunctiontools, and the whole input history is translated the same way, so resumed sessions work.stream_events_utils.rsandsession/turn.rs): items are canonicalised before they are recorded or dispatched, so rollouts and history only ever hold real names.(namespace, name). Plain aliases and hashed aliases (…_h<8hex>) live in disjoint name spaces, so mapping is injective; the proof is incodec.rs. Literal top-level names are never reverse-mapped.router.rsandrouter_tests.rsare back to pure upstream. Stock codex is unchanged: with nothing registered, the hooks are no-ops.Fork surface: 131 lines over 12 files, within the 135/12 budget.
FORK_SURFACE.mdis updated.Verified live:
qwen3.8-27b-mlx) via--engine lmstudio: an MCP namespace tool (smoke/ping) dispatches and returnspong, and the record keepsserver:smoke, tool:ping. Without the codec, LM Studio drops that tool.Known limitations (documented):
Review: three Codex gpt-6.1-sol rounds. Final verdict: APPROVE.
🤖 Generated with Claude Code
https://claude.ai/code/session_01RPCzgwU247gMtjybR5v4iR