@songyang0603/dsh-codex-apply-patch-engine

Pinned Codex apply-patch semantics exposed as a DeepSeek Harness Cordis service. The package owns one long-running native sidecar and exports parser, streaming-parser, invocation-recognition, filesystem-verification, and local mutation operations.

This package is not a model-visible apply_patch tool. Installing it does not add a tool to DSH. In particular, it does not substitute a JSON { "patch": "..." } function for Codex's freeform custom-tool contract. Approval, sandbox retry, environment selection, hooks, session events, TurnDiff, UI, and model-provider freeform transport remain separate components.

Status

Version 0.1.0 is parity_verified for the pinned semantic-engine boundary on macOS arm64 source builds. All 96 pinned upstream tests pass; the independent production-path differential matches 23/23 cases; 10 native and 6 TypeScript/native/real-Loader tests pass; and a clean DSH rc.6 profile installs, boots, mutates through, and removes the packed component using its package-local native binary. See conformance/apply-patch-engine/STATUS.md in the source repository for the exact scope and exclusions.

Installation

dsh plugin --profile  add @songyang0603/dsh-codex-apply-patch-engine

The package bundle inserts one service row named dsh-codex-apply-patch-engine. Remove it with the matching DSH plugin command.

Service

The installed service is available as ctx.codexApplyPatch. Its public operations are:

• parse(...) and streamParse(...) for pinned parser projections;
• verifyPatch(...) for filesystem-backed verification and proposed changes;
• verifyInvocation(...) for direct/shell invocation recognition;
• applyPatch(...) for the pinned unsandboxed local filesystem effect;
• shutdown() for explicit, idempotent admission closure and sidecar teardown.

Exact request and response types are exported from the package. Every method uses the native JSONL wire without renaming or weakening semantic fields. The adapter adds no default response-size or pending-request limit below the host runtime's own capacity; deployments may opt into explicit operational caps through service configuration.