[t3-code]

what changed

• isolates the direct cordis implementation from spike: compare plugin runtime models #7448 into a standalone poc
• uses real cordis contexts, fibers, injection, providers, effects, and isolation
• keeps the shared manifest and runtime contract
• adds a single-variant demo for activation, replacement, cleanup, and a 250-plugin chain

why

this lets reviewers evaluate cordis without the effect-scope or framework-free implementations in the same diff. the branch preserves the adapter and its measured tradeoffs, including the extra runtime dependency and structural type shim.

validation

• 31 tests passed with node 24.13.1
• package typecheck passed
• lint with warnings denied passed
• formatting and frozen lockfile install passed
• demo completed successfully

limits

this is an isolated runtime poc. it does not load packages, add a sandbox, expose plugin rpc, or wire product ui.

checklist

• this pr is focused on one implementation
• i explained what changed and why
• no ui changes

request provenance

• requested by: @UtkarshUsername

---

Note

Low Risk
New private POC package with no product wiring; changes are additive and test-heavy, with no auth, data, or UI surface.

Overview
Adds a standalone @t3tools/plugin-runtime-cordis-poc package that pulls the Cordis-backed plugin runtime out of the broader three-way experiment so reviewers can evaluate it alone.

The POC implements createCordisRuntime against a shared PluginRuntime contract: dependency ordering and blocking, cycle detection, differential reconcile (reuse unchanged fibers), staged isolate labels for transactional provider overlap, frozen contribution snapshots, queued exclusive operations with reentrancy guards via AsyncLocalStorage, and rollback/retire cleanup with aggregated errors. A structural shim works around Cordis rc.8 declaration issues under this repo’s ESM setup.

Shared contract tests (defineRuntimeContract, 31 cases) and Effect Schema manifest validation (semver, path escape, mobile declarative) ship with the package; a demo script benchmarks activation, noop reconcile, and a 250-plugin chain. cordis is added to the lockfile. Scope is explicitly POC-only—no package loading, sandbox, RPC, or product UI.

^{Reviewed by Cursor Bugbot for commit 50112cd. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add Cordis-backed plugin runtime POC with transactional reconciliation

• Introduces a new @t3tools/plugin-runtime-cordis-poc package implementing a plugin runtime backed by Cordis, with full lifecycle management in cordisRuntime.ts.
• reconcile() performs dependency analysis (cycle detection, topological ordering, duplicate id/capability rejection), restarts only changed plugins and their dependents, rolls back staged plugins on failure, and returns deeply frozen snapshots.
• A reusable contract test suite in runtimeContract.ts validates behaviors including dependency-ordered activation, reentrancy rejection, rollback correctness, lifecycle event ordering, and cleanup error aggregation.
• An Effect Schema-based manifest validator in manifest.ts enforces namespaced ids, semver, capability ids, and entrypoint path constraints.
• A demo.ts benchmarks a 250-plugin dependency chain and exercises activation, upgrade, and teardown flows.

^{Macroscope summarized 50112cd.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	12.6 KiB	12.6 KiB	−21 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.3 KiB	6.3 KiB	−1 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.3 KiB	−20 B (−0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.8 KiB	51.8 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	12.6 KiB	12.6 KiB	−20 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.3 KiB	6.3 KiB	−8 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.3 KiB	6.3 KiB	−12 B (−0.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.7 KiB	52.7 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: bcfd485 · PR result: 50112cd · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 99.9 KiB
• Claude decoded thread snapshot: 100.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR introduces a substantial new plugin runtime POC (~1500 lines of new code) with complex dependency management, reconciliation logic, and lifecycle handling. While isolated from production code, the scope and complexity of this new capability warrants human review to validate the design approach.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3-code]

addressed the review findings in 0b502a819:

• staged cordis fibers emit lifecycle events only after the composition commits
• rollback preserves the activation error while cleanup failures go to onCleanupError
• retirement cleanup failures are reported without rejecting an already committed snapshot
• definition equality is order-insensitive and includes activation callback identity
• entrypoints reject .. escape segments, semver validation is strict, and snapshot maps are typed as sparse records

22 tests pass with node 24.13.1, plus typecheck, lint, formatting, frozen install, and the demo.

[t3-code]

addressed the latest findings in fb81809c5:

• plugin activation and finalizers run in an async callback context
• reentrant reconcile or dispose calls now fail immediately instead of deadlocking the serialized queue
• lifecycle observer failures are isolated and reported without interrupting commit or cleanup

25 tests, typecheck, lint with warnings denied, formatting, and frozen install pass on node 24.13.1.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit fb81809. Configure here.}

[t3-code]

addressed the latest review findings in e3039b23d:

• synchronous callback microtasks can use the runtime after the callback returns
• direct reentrant calls from active callbacks still reject immediately
• late activation-context calls now reject
• reconcile snapshots its input before queueing

29 tests, typecheck, lint with warnings denied, formatting, frozen install, and the demo pass on node 24.13.1.

Dismissing prior approval to re-evaluate c42beab

[t3-code]

synced the plain-function callback return regression test and fix in c42beabec. callback lifetime tracking now treats only thenables as asynchronous.

30 tests and the full local validation pass on node 24.13.1.

[t3-code]

Consolidated into #8007 as one PR targeting main.