Replace pager-first large-output handling with server-managed output bundles

.github/workflows/ci.yml

@@ -43,7 +43,12 @@ jobs:
         run: cargo clippy --all-targets --all-features
 
       - name: cargo test (skip client integrations)
+        if: matrix.os != 'windows-latest'
         run: cargo test -- --skip codex_tui_initial_sandbox_state --skip codex_tui_initial_sandbox_state_windows_stub
 
+      - name: cargo test (windows serial, skip client integrations)
+        if: matrix.os == 'windows-latest'
+        run: cargo test -j 1 -- --test-threads=1 --skip codex_tui_initial_sandbox_state --skip codex_tui_initial_sandbox_state_windows_stub
+
       - name: cargo +nightly fmt
         run: cargo +nightly fmt --all -- --check

AGENTS.md

@@ -1,22 +1,47 @@
+# Agent Map
 
-- If you modified code, before responding to the user, always run all tests, lints, and checks and make sure everything succeeds and returns cleanly:
-  - cargo check
-  - cargo build
-  - cargo clippy
-  - cargo test
-  - cargo +nightly fmt
-
-- Project convention: never pass `--vanilla` to `R`/`Rscript` commands unless the user explicitly asks for it.
-
-- Snapshot tests (insta):
-  - Preferred local loop:
-    - `cargo insta test`
-    - `cargo insta pending-snapshots`
-    - `cargo insta review` (interactive) or `cargo insta accept` / `cargo insta reject` (non-interactive)
-  - CI-style validation: `cargo insta test --check --unreferenced=reject`
-  - For intentional snapshot format/metadata migrations: `cargo insta test --force-update-snapshots --accept`
-  - Bulk rewrite fallback (for intentional broad refreshes): `INSTA_UPDATE=always cargo test`
-  - Do not manually delete `tests/snapshots/*.snap.new`; use `cargo insta reject` to clean pending snapshots canonically.
-  - `cargo insta ...` requires the `cargo-insta` subcommand.
-
-At any time, you may consult the R source code at `~/github/wch/r-source` and the Python source code at `~/github/python/cpython` to investigate and resolve questions.
+Keep this file short. It is a table of contents, not the full manual.
+
+## Immediate Rules
+
+- If you modified code, run all required checks before replying:
+  - `cargo check`
+  - `cargo build`
+  - `cargo clippy --all-targets --all-features -- -D warnings`
+  - `cargo test`
+  - `cargo +nightly fmt`
+- Treat all clippy warnings as failures. Do not leave warning cleanup for later.
+- Never pass `--vanilla` to `R` or `Rscript` unless the user explicitly asks for it.
+
+## Start Here
+
+- `docs/index.md`: source-of-truth map for repository docs.
+- `docs/architecture.md`: subsystem map for the binary, worker, sandbox, and eval surfaces.
+- `docs/testing.md`: public verification surface and snapshot workflow.
+- `docs/debugging.md`: debug logs, `--debug-repl`, and stdio tracing.
+- `docs/sandbox.md`: sandbox modes and writable-root policy.
+- `docs/plans/AGENTS.md`: when to create checked-in execution plans.
+
+## Snapshot Workflow
+
+- Preferred loop:
+  - `cargo insta test`
+  - `cargo insta pending-snapshots`
+  - `cargo insta review` or `cargo insta accept` / `cargo insta reject`
+- CI-style validation: `cargo insta test --check --unreferenced=reject`
+- For broad intentional snapshot migrations: `cargo insta test --force-update-snapshots --accept`
+- Do not delete `tests/snapshots/*.snap.new` manually. Use `cargo insta reject`.
+
+## Planning Rule
+
+- For multi-phase refactors, redesigns, or other work that spans discovery, iteration, and implementation, keep a living plan under `docs/plans/active/` until the initiative is complete.
+- Use the plan to capture design decisions, rejected options, phase boundaries, unresolved questions, and the next safe slice of work so a later agent does not need to rediscover them.
+- If you pause or hand off work mid-task, update the plan before stopping.
+- Do not create plan files for routine, obvious, or low-risk changes. Keep the plans area useful, not noisy.
+- Move completed plans to `docs/plans/completed/`.
+- Treat `docs/notes/` and `docs/futurework/` as exploratory, not normative.
+
+## External References
+
+- Consult `~/github/wch/r-source` for R behavior details.
+- Consult `~/github/python/cpython` for Python behavior details.