mirror of https://github.com/pchuan98/codex.git synced 2026-07-01 00:31:56 +08:00

Files

T

Celia Chen 6042e5810e chore: preserve one more schema layer during large tool compaction (#27084 )

## Summary

Some customer MCP tools expose large input schemas that exceed Codex's
compact schema budget even after description stripping. Today, the final
compaction pass collapses complex schemas starting at depth 2, which can
erase important shallow call structure such as small `anyOf` branches,
required fields, and help-mode entry points. In one reported case, this
degraded a tool schema into `query: any | any`, leaving the model
without enough structure to discover the required help call.

This change raises the deep-schema collapse boundary from depth 2 to
depth 3. That preserves one additional layer of the tool contract while
still collapsing deeper expensive subtrees to `{}` when a schema remains
over budget.

## What Changed

- Increased `MAX_COMPACT_TOOL_SCHEMA_DEPTH` from `2` to `3`.
- Updated the schema compaction traversal test to assert the new
collapse boundary.
- The resulting compacted shape keeps useful shallow structure, for
example:
  - top-level argument names
  - shallow `anyOf` branches
  - required object fields
  - nested property names one level deeper than before

## Validation

- Ran `just test -p codex-tools`: 81 tests passed.
- Ran a golden schema corpus comparison over 214 discovered tool input
schemas under `golden_schemas/*/mcp_tools/*/input_schema.json`.
- Depth 2 and depth 3 had identical percentile token counts across the
corpus.
  - Both ended with `0 / 214` schemas over 1k tokens.
- Both ended with `0 / 214` schemas over the 4,000-byte compact JSON
budget.
- Only one golden schema changed, increasing from 49 to 56 tokens, so
this does not appear to introduce a meaningful corpus-wide regression.

Corpus percentile results:

| Percentile | Depth 2 | Depth 3 |
|---|---:|---:|
| p0 | 9 | 9 |
| p10 | 31 | 31 |
| p25 | 54 | 54 |
| p50 | 81 | 81 |
| p75 | 143 | 143 |
| p90 | 290 | 290 |
| p95 | 431 | 431 |
| p99 | 600 | 600 |
| max | 832 | 832 |

6042e5810e · 2026-06-08 23:07:56 +00:00

History

src

chore: preserve one more schema layer during large tool compaction (#27084 )

2026-06-08 23:07:56 +00:00

tests

feat: support oneOf and allOf in tool input schemas (#24118 )

2026-06-08 21:41:05 +00:00

BUILD.bazel

[codex] Move tool specs into core handlers (#21416 )

2026-05-06 15:40:50 -07:00

Cargo.toml

chore: add JSON schema policy fixture coverage (#24152 )

2026-05-22 16:31:33 -07:00

README.md

Remove ToolsConfig from tool planning (#22835 )

2026-05-19 11:24:09 +02:00

README.md

codex-tools

codex-tools is the shared support crate for building, adapting, and executing model-visible tools outside codex-core.

Today this crate owns the host-facing tool models and helpers that no longer need to live in core/src/tools/spec.rs or core/src/client_common.rs:

aggregate host models such as ToolSpec, ConfiguredToolSpec, LoadableToolSpec, ResponsesApiNamespace, and ResponsesApiNamespaceTool
host discovery models used while assembling tool sets, including discoverable-tool models and request-plugin-install helpers
host adapters such as schema sanitization, MCP/dynamic conversion, code-mode augmentation, and image-detail normalization
shared executable-tool contracts such as ToolExecutor, ToolCall, and ToolOutput

That extraction is the first step in a longer migration. The goal is not to move all of core/src/tools into this crate in one shot. Instead, the plan is to peel off reusable pieces in reviewable increments while keeping compatibility-sensitive orchestration in codex-core until the surrounding boundaries are ready.

Vision

Over time, this crate should hold host-side tool machinery that is shared by multiple consumers, for example:

host-visible aggregate tool models
tool-set planning and discovery helpers
MCP and dynamic-tool adaptation into Responses API shapes
code-mode compatibility shims that do not depend on codex-core
other narrowly scoped host utilities that multiple crates need

The corresponding non-goals are just as important:

do not move codex-core orchestration here prematurely
do not pull Session / TurnContext / approval flow / runtime execution logic into this crate unless those dependencies have first been split into stable shared interfaces
do not turn this crate into a grab-bag for unrelated helper code

Migration approach

The expected migration shape is:

Keep extension-owned executable-tool authoring in codex-extension-api.
Move host-side planning/adaptation helpers here when they no longer need to stay coupled to codex-core.
Leave compatibility-sensitive adapters in codex-core while downstream call sites are updated.
Only extract higher-level host infrastructure after the crate boundaries are clear and independently testable.

Crate conventions

This crate should start with stricter structure than core/src/tools so it stays easy to grow:

src/lib.rs should remain exports-only.
Business logic should live in named module files such as foo.rs.
Unit tests for foo.rs should live in a sibling foo_tests.rs.
The implementation file should wire tests with:

#[cfg(test)]
#[path = "foo_tests.rs"]
mod tests;

If this crate starts accumulating code that needs runtime state from codex-core, that is a sign to revisit the extraction boundary before adding more here.