feat: rebuild the Cursor adapter for OpenCode V2

.github/workflows/release.yml

@@ -39,6 +39,7 @@ jobs:
         shell: bash
         run: |
           set -euo pipefail
+          # Version bumps are reviewed separately; this workflow publishes package.json verbatim.
           version="$(node -p "require('./package.json').version")"
           if [[ ! "$version" =~ ^[0-9]+\.[0-9]+\.[0-9]+-beta\.[0-9]+$ ]]; then
             echo "Expected package version X.Y.Z-beta.N, got: $version" >&2

.gitignore

@@ -5,6 +5,7 @@ dist/
 
 # Local agent notes / environment-specific
 AGENTS.md
+.opencode/
 
 # Graphify knowledge graph output
 graphify-out/

CHANGELOG.md

@@ -1,80 +1,34 @@
 # Changelog
 
-All notable changes to this project are documented in this file.
-
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
-and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
-
-## [3.0.0-beta.2] - 2026-08-24
-
-### Added
-
-- OpenCode V2 `Plugin.define` entrypoint with native integration, catalog, and
-  `LanguageModelV3` registrations
-- Native Cursor AgentService transport for images, cancellation, parallel tool
-  calls, bounded H2 pooling, and live tool-result continuation
-- V2 loader, package, lifecycle, catalog, integration, and language-adapter tests
-- Preserved OpenCode V1 adapter at the `./v1` package export; npm `latest`
-  remains on the `2.x` release line while V2 publishes under npm `beta`
-
-### Changed
-
-- OpenCode V2 now owns transcript reconstruction, credential persistence,
-  permissions, tool execution, and compaction; the V2 path no longer uses the
-  localhost OpenAI-compatible proxy
-- Adapted from CasualDeveloper's V2 implementation in commit
-  `3af03f605243b58b33c2a9e1f9fa638280bca693`
-
-## [2.2.0] - 2026-08-05
-
-### Changed
-
-- Phase-aware stall budget for post-tool resumes: silent tool continuations recover in **90s** instead of 180s
-- H2 bridge workers pre-connect TLS/HTTP-2 at startup (faster first message after restart)
-- Tool-call debounce reduced from 500ms → 250ms
-- Title-gen model probe result persisted to disk (skip ~2.5s Zen probe after restart)
-
-### Added
-
-- `OPENCODE_CURSOR_POST_TOOL_PRE_OUTPUT_STALL_TIMEOUT_MS` (default 90s)
-- Regression coverage for silent post-tool stall recovery
-
-### Performance
-
-- First message (`gpt-5.4-nano`): **8.5s → 4.8s** (−44%)
-- First message (`cursor/default`): **6.1s → 3.7s** (−39%)
-
-## [2.1.0] - 2026-08-05
-
-### Changed
-
-- Phase-aware stall budgets: cold thinking gets **180s** so reasoning models are not discarded mid-thought
-- Recovery limits honor `MAX_STALL_RECOVERIES`
-- `proxyTelemetry` exported for observability and tests
-
-### Added
-
-- `OPENCODE_CURSOR_PRE_OUTPUT_STALL_TIMEOUT_MS` (default 180s)
-
-## [2.0.0] - 2026-08-04
-
-### Fixed
-
-- Root-cause restate loop: tool-result follow-ups no longer look like user interrupts that discard parked bridges
-- Frozen sessions: visible-text stalls surface an error in ~90s instead of holding a step for up to 12 minutes
-- Infinite recovery restarts capped by forward progress
-- Model discovery retries 3× on transient bridge failures
-
-### Removed
-
-- Accumulated loop detectors, loop-break notes, compaction re-framing, and auto-continue nudges (~1,450 lines)
-
-## [1.4.0] - 2026-08-03
-
-See [GitHub Releases](https://github.com/otto-assistant/opencode-cursor/releases) for earlier notes.
-
-[2.2.0]: https://github.com/otto-assistant/opencode-cursor/compare/v2.1.0...v2.2.0
-[3.0.0-beta.2]: https://github.com/otto-assistant/opencode-cursor/compare/v2.2.0...v3.0.0-beta.2
-[2.1.0]: https://github.com/otto-assistant/opencode-cursor/compare/v2.0.0...v2.1.0
-[2.0.0]: https://github.com/otto-assistant/opencode-cursor/compare/v1.4.0...v2.0.0
-[1.4.0]: https://github.com/otto-assistant/opencode-cursor/releases/tag/v1.4.0
+This channel targets OpenCode V2 exclusively.
+
+## Unreleased
+
+Live release validation remains pending for instruction precedence, opaque-reasoning
+replay, and continued work after automatic compaction, as recorded in
+the [acceptance report](docs/opencode-v2-release-acceptance.md).
+
+- Reconstruct authoritative host history through structured Cursor root blobs.
+- Observe host tool, permission, question, and session events while forwarding
+  results only from the next authoritative model invocation.
+- Use bounded, per-Run Node HTTP/2 workers with validated Connect completion.
+- Preserve late reasoning signatures in durable host metadata for fresh replay.
+- Report terminal input, output, cache, and reasoning counters with explicit
+  usage scope and separate checkpoint occupancy.
+- Keep multi-invocation Run totals in durable provider metadata rather than
+  misreporting them as usage for the final host invocation. Verify the correction
+  against the pinned host's automatic compaction behavior.
+- Preserve checkpoint-referenced opaque reasoning in durable host metadata and
+  restore its exact placement during replay. Reject ambiguous or changed new
+  roots rather than guessing their relationship to emitted output.
+- Add deterministic pinned-host acceptance and opt-in synthetic live probes.
+- Project host system instructions into ordered Cursor rules as well as history
+  roots. Add an explicit host contract for precedence and genuine tool outcomes.
+  Instruction delivery remains best effort; live acceptance is pending.
+- Restrict this package to the native V2 plugin API and its runtime dependencies.
+
+## 3.0.0-beta.2
+
+- Native OpenCode V2 plugin, integration, catalog, and language-model APIs.
+- Account-discovered Cursor models with exact model/variant routing.
+- OpenCode-owned OAuth credentials, permissions, sessions, tools, and compaction.

README.md

@@ -8,13 +8,20 @@ Use the models available to your Cursor account from OpenCode 2, including
 Cursor private models, live model variants, image input, streaming, and tool
 continuation.
 
-This is the V2 beta line. OpenCode 1 users remain supported by the `2.x`
-release on npm's `latest` tag; the beta package also exposes the preserved
-legacy adapter as `@otto-assistant/opencode-cursor-oauth/v1` for migration
-testing.
+This package and the `beta` branch target OpenCode V2 exclusively. The channel
+name identifies the OpenCode compatibility line; users of this channel require
+a complete, release-ready plugin.
 
 ## Status
 
+**Live release validation pending.** The local adapter includes the corrected
+usage projection and opaque-reasoning replay. Host instructions use global rules
+with an explicit host contract; their precedence remains best effort. Live
+validation of these changes and sustained automatic compaction is still pending.
+See the [acceptance report](docs/opencode-v2-release-acceptance.md) for evidence
+and the remaining release requirements. The `beta` channel requires production
+readiness for OpenCode V2 users.
+
 - Plugin version: `3.0.0-beta.2`
 - OpenCode channel: V2 beta
 - Plugin API: pinned to the matching OpenCode beta build
@@ -79,16 +86,24 @@ The Cursor desktop application and `cursor-agent` CLI are not required.
 OpenCode 2
   └─ V2 integration and catalog APIs
        └─ native LanguageModelV3 adapter
-            └─ Node HTTP/2 bridge pool
+             └─ request-scoped Node HTTP/2 worker
                  └─ Cursor AgentService
 ```
 
 OpenCode owns provider selection, credentials, permissions, persistence, and
-tool execution. Each user turn starts a Cursor AgentService Run from OpenCode's
-active transcript. If Cursor requests a tool, that Run remains alive only until
-OpenCode returns the matching result; terminal responses discard it. Cursor
-checkpoints are not retained, while private Composer models and native Cursor
-rate limits remain available.
+tool execution. The adapter reconstructs structured history from OpenCode's
+active transcript and can retain a bounded live Run across tool steps. Late
+calls remain queued for delivery. Host events keep the stream open while tools
+run; results come only from the next OpenCode checkpoint. History or account
+changes discard the continuation. Access and billing remain subject to the
+connected Cursor account.
+
+Host system messages also become ordered, always-apply Cursor rules. This
+delivers instructions through Cursor's context mechanism, but does not establish
+the same precedence as an independently controlled system prompt.
+
+See the [adapter decision and acceptance status](docs/opencode-v2-host-owned-adapter.md)
+for the implementation lifecycle, evidence, and remaining live release gates.
 
 ### Model routing
 
@@ -102,13 +117,14 @@ the plugin refreshes discovery and asks OpenCode to reload the catalog.
 
 ### Lifecycle
 
-The plugin starts a bounded HTTP/2 worker pool during V2 `setup()`. Disabling,
-reloading, or shutting down the plugin stops:
+V2 starts a Node worker lazily for each admitted Run. Disabling, reloading, or
+shutting down a plugin instance stops its:
 
 - active AgentService Runs
 - HTTP/2 bridge workers
 - pending OAuth polling
 - catalog event subscriptions
+- tool-observation subscriptions
 
 ## Development
 
@@ -117,6 +133,7 @@ npm test
 npm run test:v2
 npm run typecheck
 npm run build
+npm run test:v2-host
 npm run test:package
 npm run test:v2-loader
 ```
@@ -131,6 +148,21 @@ npm run verify
 dependencies with lifecycle scripts disabled, imports it independently, and
 loads that extracted package through the pinned `opencode2` beta.
 
+The [Cursor capability probe](docs/opencode-v2-cursor-capability-probe.md)
+compares tool restrictions and structured history on synthetic conversations.
+Its offline tests run in `verify`; live Runs require a separate invocation.
+
+### Beta release
+
+Before dispatching the **Release V2 Beta** workflow, land a reviewed commit that
+updates `package.json` to the next unused `X.Y.Z-beta.N` version and updates any
+lockfiles changed by the package manager. Run the workflow from `beta` with
+`dry_run` enabled first, then rerun the same commit with `dry_run` disabled to
+publish.
+
+The workflow does not bump versions or create release commits. It validates,
+packs, and publishes the exact reviewed version to npm's `beta` dist-tag.
+
 ## Debugging
 
 Enable plugin logs:
@@ -142,16 +174,14 @@ OPENCODE_CURSOR_DEBUG=1 opencode2
 Optional AgentService controls:
 
 - `OPENCODE_CURSOR_NATIVE_TOOL_SETTLE_MS`
+- `OPENCODE_CURSOR_NATIVE_TOOL_WAIT_MS`
 - `OPENCODE_CURSOR_PRE_OUTPUT_STALL_TIMEOUT_MS`
 - `OPENCODE_CURSOR_POST_TOOL_PRE_OUTPUT_STALL_TIMEOUT_MS`
 - `OPENCODE_CURSOR_STALL_TIMEOUT_MS`
 - `OPENCODE_CURSOR_MAX_ACTIVE_RUNS`
 - `OPENCODE_CURSOR_NATIVE_PARK_TTL_MS`
 - `OPENCODE_CURSOR_DEFAULT_CONTEXT_WINDOW`
 - `OPENCODE_CURSOR_DEFAULT_MAX_TOKENS`
-- `OPENCODE_CURSOR_BRIDGE_POOL_MIN`
-- `OPENCODE_CURSOR_BRIDGE_POOL_MAX`
-- `OPENCODE_CURSOR_BRIDGE_POOL_DISABLED`
 
 ## Known beta constraints
 
@@ -162,6 +192,21 @@ Optional AgentService controls:
 - Models without explicit Cursor context metadata use the configurable default
   context and output limits listed above.
 - The plugin relies on Cursor's private API, which can change without notice.
+- The replacement has offline host acceptance and live Composer/Auto/Opus,
+  signed-reasoning restart, 452k mixed-history, and billing-comparison evidence.
+  Retained Runs reported cache reads; fresh long Runs did not show a warm-replay
+  saving. See the [release acceptance report](docs/opencode-v2-release-acceptance.md).
+- Terminal inference and cache counters are retained in `cursor.turnUsage`
+  provider metadata. A Run spanning multiple host invocations has unknown
+  per-invocation usage, so OpenCode's built-in token and cost totals are incomplete
+  for those turns. Checkpoint occupancy and progress deltas remain separate.
+- Signed and opaque reasoning have offline-verified persistence and restart
+  replay. Opaque blocks retain their exact data and placement, anchored to one
+  unchanged assistant message. Unmatched or ambiguous new blocks fail explicitly.
+  Signed reasoning also has live replay evidence; opaque reasoning does not yet.
+- Host instructions are sent as ordered global rules and genuine system-history
+  roots, with an explicit rule for instruction priority and real host-tool use.
+  This is best-effort delivery, not verified replacement of Cursor's system prompt.
 
 ## License