@code-yeongyu/senpi-codemode 2026.7.25-2

package/CHANGELOG.md

@@ -0,0 +1,250 @@
+# Changelog
+
+## [Unreleased]
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.25-2] - 2026-07-25
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.25] - 2026-07-25
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.24] - 2026-07-24
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.23] - 2026-07-23
+
+### Breaking Changes
+
+### Added
+
+- Added the GPT-only Code Mode runtime with `exec` and `wait` tools, plus model-aware GPT eval routing ([#301](https://github.com/code-yeongyu/senpi/pull/301)).
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.22-2] - 2026-07-22
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.22] - 2026-07-22
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.20-2] - 2026-07-20
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.20] - 2026-07-20
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+- Fixed the Python kernel's `tool.<name>()` proxy injecting an omp-only `i` ("py prelude") intent field into every bridged tool call. Senpi tool schemas never declare `i`, so strict tools (`additionalProperties: false`, e.g. `web_search`) rejected every eval-bridged call with `Validation failed for tool …: must not have additional properties`. Args now pass through verbatim, matching the JS/Ruby/Julia preludes.
+
+### Removed
+
+## [2026.7.17-5] - 2026-07-17
+
+### Breaking Changes
+
+### Added
+
+### Changed
+- Changed the Kimi K-series eval prompt dialect to make eval-first, whole-step parallel batching the default: strong positive emphasis now directs multi-call work into one `eval` cell, parallelizes independent calls, handles failures in-kernel, and returns distilled facts.
+
+### Fixed
+
+### Removed
+
+## [2026.7.17-4] - 2026-07-17
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.17-3] - 2026-07-17
+
+### Breaking Changes
+
+### Added
+
+### Changed
+
+### Fixed
+
+### Removed
+
+## [2026.7.17-2] - 2026-07-17
+
+### Added
+
+- Added a host-sizing note to the `eval` prompt: the extension now passes a preformatted host line (platform, arch, CPU model, core count) at registration so the prompt tells the model to size `parallel(thunks)` pools to the local cores and keep shell commands platform-appropriate.
+- Added model-aware eval-first batching emphasis: the `eval` tool description and its system-prompt guideline now render in a dialect selected by the active model id (Claude/GLM, OpenAI, Kimi, and a maximum-emphasis default fallback), re-registering on `model_select` so mid-session model switches pick up the matching dialect.
+
+### Changed
+
+### Fixed
+
+## [2026.7.17] - 2026-07-17
+
+### Added
+
+### Changed
+
+### Fixed
+
+- Fixed `eval` tool calls rendering duplicate stacked boxes after a result arrived; the pending, running, and completed states now update in one in-place frame ([#223](https://github.com/code-yeongyu/senpi/pull/223)).
+
+## [2026.7.16-3] - 2026-07-16
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [2026.7.16-2] - 2026-07-16
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [2026.7.16] - 2026-07-16
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [2026.7.14-3] - 2026-07-14
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [2026.7.14-2] - 2026-07-14
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [2026.7.14] - 2026-07-14
+
+### Added
+
+### Changed
+
+- Improved the `eval` prompt instructions and reuse-chain examples to teach persistent-state reuse, batch file processing, and parallel session-tool fan-out within a single cell.
+
+### Fixed
+
+## [2026.7.13] - 2026-07-13
+
+### Added
+
+- Added the source-only `@code-yeongyu/senpi-codemode` workspace package scaffold.
+- Added codemode settings loading, interpreter detection, prompt generation, loopback bridge helpers, and persistent JS/Python/Ruby/Julia kernel building blocks.
+- Added structured kernel status events from the bridge through TUI rendering.
+- Added `agent()` and `output()` bridges that delegate through configured task-tool contracts.
+- Added bounded streaming output with session-adjacent spill files and plain-path notices.
+- Added eval render parity for highlighted cells, status rows, task progress, JSON displays, truncation warnings, and image fallbacks.
+- Added JavaScript import rewriting for persistent eval cells.
+
+### Changed
+
+- Activated the exported extension factory so the bundled package registers and reconfigures the persistent-kernel `eval` tool in Senpi sessions.
+- Improved `eval` TUI rendering with streaming status and timing, bounded expandable previews, width-safe ANSI/CJK/emoji reflow, nested tool-call state, and terminal-aware image fallbacks.
+- Re-register the eval prompt and schema at session start after settings, interpreter availability, and active task-tool names resolve.
+- Recorded the completed oh-my-pi eval-port provenance for this extension; task delegation and artifact handling follow Senpi extension boundaries.
+
+### Fixed
+
+- Prevented image MIME labels from injecting terminal control sequences through eval text fallbacks.
+- Fixed eval cancellation and timeout handling across JavaScript, Python, Ruby, and Julia kernels: aborts now interrupt active work, unresponsive subprocesses escalate to bounded hard termination, queued Python cells cannot execute after cancellation, persistent Python state survives graceful interrupts, timeout/death durations remain truthful, and late bridge or retired-process output cannot keep an eval hung or contaminate the next cell.
+- Fixed the bundled `eval` extension failing to load in packaged installs: `completion/handler.ts` imported peer symbols via the monorepo source path `../../../ai/src/*`, which only resolves inside the workspace and threw `Cannot find module` once packed. It now imports from the `@earendil-works/pi-ai/compat` package entry, so `eval` loads in the shipped Node package.
+- Fixed a temporal-dead-zone crash in the `eval` tool: subprocess kernels (py/rb/jl) emit their `ready` frame synchronously during kernel startup, which invoked the message handler before the `kernel` binding initialized and crashed the whole agent process. The self-referential binding is now hoisted so startup frames no longer throw.
+- Fixed cell-output misattribution on reused persistent kernels: `getKernel` now rebinds the per-cell `onMessage` on every call, so a second (and later) cell's streamed `text`/`display`/`log` output is delivered to that cell instead of the previous one.
+- Fixed the Ruby kernel corrupting its JSONL protocol channel: user `puts`/`print` output is now captured via a redirected `$stdout` and emitted as `text` frames instead of being written directly onto the shared stdout stream.
+- Fixed the Ruby kernel raising `ArgumentError: unknown keywords` on Ruby 3.0+ (e.g. CI's Ruby 3.x, while local Ruby 2.6 masked it): `env()`/`read()`/`write()` passed braceless string-keyed hashes to `__senpi_emit_status`, which Ruby 3 parses as keyword arguments against its `force:` keyword parameter instead of the positional `fields` hash. The field hashes are now wrapped in explicit braces so status emission and final-expression auto-display work identically across Ruby 2.6–3.4.

package/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) 2025 Mario Zechner (upstream pi-mono)
+Copyright (c) 2026 Yeongyu Kim and senpi contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,161 @@
+# @code-yeongyu/senpi-codemode
+
+`@code-yeongyu/senpi-codemode` is Senpi's source-only Code Mode extension. It
+registers persistent-kernel `eval` for every eligible model and GPT-only
+`exec`/`wait` cells for bounded JavaScript orchestration. `eval` owns one
+persistent kernel per enabled language and re-registers at session start after
+configuration, interpreter availability, and active task-tool names are known.
+
+## Capabilities
+
+- Persistent JavaScript, Python, Ruby, and Julia cells. State survives later
+  cells in the same language until reset, restart, or session disposal.
+- Loopback, bearer-authenticated kernel bridge with bounded JSONL frames.
+- Structured status events for file operations, environment access, phases,
+  bridge activity, and delegated task progress.
+- Bounded streaming output with head/tail previews, column clamping, and
+  session-adjacent spill files for large streams.
+- TUI and HTML-export rendering for syntax-highlighted cells, status rows,
+  task progress, structured display values, truncation warnings, and image
+  fallbacks.
+- JavaScript import rewriting for supported local modules and package imports
+  in the persistent Node.js worker.
+- GPT Code Mode `exec`/`wait` cells use fresh JavaScript workers, can compose
+  active tools through `tools.<name>(args)`, and never replace `eval`.
+
+## Kernels
+
+| Language | Default | Runtime | Notes |
+| --- | --- | --- | --- |
+| `js` | enabled | Node.js worker | Requires Node.js 24 or newer; supports top-level `await` and `return`. |
+| `py` | enabled | `python3` or `python` | Optional interpreter detected at session start. |
+| `rb` | disabled | `ruby` | Optional interpreter detected at session start. |
+| `jl` | disabled | `julia` | Optional interpreter detected at session start. |
+
+A missing optional interpreter removes that language from the session's `eval`
+schema; it is not an installation failure.
+
+## Settings
+
+Configuration is loaded in this order:
+
+1. `.senpi/codemode.json` in the session working directory
+2. `~/.senpi/agent/codemode.json`
+3. Built-in defaults
+
+```json
+{
+  "languages": {
+    "py": true,
+    "js": true,
+    "rb": false,
+    "jl": false
+  },
+  "cellTimeoutSeconds": 30,
+  "parallelPoolWidth": 4,
+  "taskTools": {
+    "task": "task",
+    "output": "task_output"
+  },
+  "outputSink": {
+    "headBytes": 20480,
+    "maxColumns": 768
+  },
+  "statusEvents": true
+}
+```
+
+| Key | Default | Effect |
+| --- | --- | --- |
+| `languages` | `py`/`js` enabled; `rb`/`jl` disabled | Selects desired languages before interpreter detection. |
+| `cellTimeoutSeconds` | `30` | Idle timeout for one cell unless the call supplies `timeout`. |
+| `parallelPoolWidth` | `4` | Maximum concurrent `parallel()` thunks. |
+| `taskTools.task` | `"task"` | Registered tool name used by `agent()`. |
+| `taskTools.output` | `"task_output"` | Registered tool name used by `output()`. |
+| `outputSink.headBytes` | `20480` | Bytes retained from the beginning of a middle-truncated preview; `0` disables it. |
+| `outputSink.maxColumns` | `768` | Maximum rendered output columns; `0` disables column clamping. |
+| `statusEvents` | `true` | Enables kernel status-event forwarding and rendering. |
+
+`SENPI_CODEMODE_PY`, `SENPI_CODEMODE_JS`, `SENPI_CODEMODE_RB`, and
+`SENPI_CODEMODE_JL` override the corresponding file setting. `1` or `true`
+enables; `0` or `false` disables. Any other value leaves the file setting in
+effect.
+
+Malformed JSON or invalid settings fall back to defaults with a warning.
+
+## Cell helpers
+
+Python, JavaScript, Ruby, and Julia expose the same conceptual helpers. Python,
+Ruby, and Julia use trailing keyword options; JavaScript uses one trailing
+options object and asynchronous helpers are `await`-able.
+
+| Helper | Contract |
+| --- | --- |
+| `display(value)` | Emits text, structured JSON, markdown, or supported image display data. |
+| `print(value, ...)` | Emits text output. |
+| `read(path, offset?, limit?)` | Reads text with 1-indexed line slicing. `local://` paths resolve under the session artifact root. |
+| `write(path, content)` | Creates parent directories and writes text. `local://` paths persist in the session artifact root. |
+| `env(key?, value?)` | Reads all kernel environment values, one value, or sets one value. |
+| `tool.<name>(args)` / `tools.<name>(args)` | Invokes an active Senpi tool through the normal `pi.executeTool` pipeline. `tools` is the GPT Code Mode spelling. |
+| `completion(prompt, model?, system?, schema?)` | Requests a one-shot host completion; `schema` asks the host to parse structured output. |
+| `agent(prompt, ...)` | Delegates to the configured active `taskTools.task` tool. Supports background handles and structured JSON results. |
+| `output(ids, format?, offset?, limit?)` | Delegates transcript retrieval to the configured active `taskTools.output` tool. |
+| `parallel(thunks)` | Runs thunks through the configured bounded pool while preserving input order. |
+| `pipeline(items, ...stages)` | Applies stages left to right with a barrier between stages. |
+| `log(message)` / `phase(title)` | Emits progress text and starts a status phase. |
+
+`agent()` is available only when the configured task tool is active in the
+session. `output()` similarly requires the configured task-output tool. Missing
+tools produce a clear availability error instead of importing an orchestration
+package. `agent()` delegates through the tool contract, so task-engine
+permissions, progress updates, and transcripts remain owned by that engine.
+`isolated`, `apply`, and `merge` are accepted for compatibility but emit a
+warning because this task-engine integration has no isolation model.
+
+## Output and artifacts
+
+Cell output is streamed while the cell runs. Large streams spill to an absolute
+file after the default 50 KiB threshold. With a session file such as
+`/path/session.jsonl`, artifacts live in `/path/session-artifacts/`; sessions
+without a file use a unique temporary directory. Truncated results include a
+plain-path notice such as `[Full output: /absolute/path/eval-….log]`.
+
+## Deliberate differences from oh-my-pi
+
+- There is no `budget` helper.
+- There is no `artifact://` protocol. Spill references are ordinary absolute
+  file paths.
+- `agent()` and `output()` compose registered task tools through
+  `pi.executeTool`; this package does not import a task-engine workspace
+  package.
+- Task transcript formats are limited to full (`raw`) and trailing (`tail`)
+  output. Query, JSON, and stripped metadata formats are task-engine concerns.
+
+## Security and lifecycle
+
+Kernels run locally with the invoking user's permissions. The bridge listens on
+loopback only and authenticates each session with a random bearer token.
+Session generations fence retired kernels and callbacks; each cell settles once
+across completion, errors, cancellation, timeout, bridge failure, or a kernel
+crash.
+
+GPT Code Mode is not an additional sandbox: it uses the same Node worker trust
+boundary as JavaScript `eval`. Its separate cell lifecycle exists so a yielded
+`exec` can be observed or terminated with `wait` without affecting persistent
+`eval` kernels. `eval`, `exec`, and `wait` are excluded from the nested tool
+namespace to prevent recursive Code Mode execution.
+
+## Validation
+
+```bash
+cd packages/senpi-codemode
+npm test
+
+cd ../..
+npm run check
+```
+
+Direct real-surface QA drivers live in `scripts/qa-*.ts`: kernel cells
+(`qa-py-cell.ts`, `qa-js-cell.ts`, `qa-rb-cell.ts`, `qa-jl-cell.ts`), end-to-end
+extension execution (`qa-e2e-eval.ts`, `qa-gpt-code-mode.ts`), and renderer output
+(`qa-render-dump.ts`).

package/package.json

@@ -0,0 +1,58 @@
+{
+	"name": "@code-yeongyu/senpi-codemode",
+	"version": "2026.7.25-2",
+	"description": "Source-only senpi extension package for codemode evaluation tools",
+	"type": "module",
+	"main": "./src/index.ts",
+	"types": "./src/index.ts",
+	"exports": {
+		".": "./src/index.ts",
+		"./package.json": "./package.json"
+	},
+	"files": [
+		"src",
+		"README.md",
+		"CHANGELOG.md",
+		"LICENSE"
+	],
+	"scripts": {
+		"test": "npx tsx ../../node_modules/vitest/dist/cli.js --run test/",
+		"clean": "node -e \"\"",
+		"build": "node -e \"\""
+	},
+	"pi": {
+		"extensions": [
+			"./src/index.ts"
+		]
+	},
+	"publishConfig": {
+		"access": "public"
+	},
+	"dependencies": {
+		"@babel/parser": "7.29.7",
+		"@earendil-works/pi-ai": "npm:@code-yeongyu/senpi-ai@^2026.7.25-2",
+		"typebox": "1.1.38"
+	},
+	"peerDependencies": {
+		"@code-yeongyu/senpi": "*"
+	},
+	"devDependencies": {
+		"@code-yeongyu/senpi": "2026.7.25-2"
+	},
+	"keywords": [
+		"senpi",
+		"pi",
+		"extension",
+		"codemode"
+	],
+	"author": "Yeongyu Kim",
+	"license": "MIT",
+	"repository": {
+		"type": "git",
+		"url": "git+https://github.com/code-yeongyu/senpi.git",
+		"directory": "packages/senpi-codemode"
+	},
+	"engines": {
+		"node": ">=24.0.0"
+	}
+}