@code-yeongyu/senpi-codemode 2026.10.9 → 2026.10.10-10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +158 -0
- package/README.md +161 -11
- package/package.json +6 -5
- package/src/bridge/protocol.ts +8 -3
- package/src/bridge/reserved.ts +2 -0
- package/src/bridges/eval-environment-schemas.ts +48 -0
- package/src/bridges/eval-virtual-schemas.ts +11 -4
- package/src/bridges/reserved-dispatch.ts +13 -1
- package/src/config/executable-settings.json +5 -0
- package/src/config/project-trust.ts +51 -0
- package/src/config/settings-overrides.ts +2 -0
- package/src/environments/js-environments.ts +208 -0
- package/src/environments/js-installer.ts +268 -0
- package/src/environments/js-revision-files.ts +91 -0
- package/src/environments/no-links.ts +132 -0
- package/src/environments/packages-install.ts +98 -0
- package/src/environments/project-resolves.ts +54 -0
- package/src/environments/py-environment.ts +7 -3
- package/src/environments/py-installer.ts +5 -1
- package/src/environments/python-environments.ts +1 -1
- package/src/environments/revision-store.ts +20 -4
- package/src/extension/bridge-tool-call.ts +5 -1
- package/src/extension/eval-notifier.ts +13 -2
- package/src/extension/kernel-registration.ts +7 -2
- package/src/extension/runtime-factory.ts +51 -13
- package/src/extension/runtime-info.ts +6 -2
- package/src/extension/session-manager-contract.ts +3 -0
- package/src/extension/session-manager.ts +16 -2
- package/src/index.ts +3 -2
- package/src/interpreters/detect.ts +34 -1
- package/src/kernels/AGENTS.md +2 -0
- package/src/kernels/js/cell-dispatch.ts +44 -0
- package/src/kernels/js/cell-ownership.js +268 -0
- package/src/kernels/js/cell-run-context.js +89 -0
- package/src/kernels/js/context-manager.ts +46 -51
- package/src/kernels/js/control-frames.ts +24 -0
- package/src/kernels/js/host-entries.ts +62 -0
- package/src/kernels/js/kernel-contract.ts +16 -1
- package/src/kernels/js/kernel-memory-bridge.ts +1 -1
- package/src/kernels/js/kernel-memory.ts +79 -0
- package/src/kernels/js/kernel-tools-registry.js +1 -0
- package/src/kernels/js/local-module-loader.ts +17 -6
- package/src/kernels/js/process-drain-scan.d.ts +8 -0
- package/src/kernels/js/process-drain-scan.js +28 -0
- package/src/kernels/js/process-entry.js +466 -0
- package/src/kernels/js/process-memory-host.ts +83 -0
- package/src/kernels/js/process-worker.ts +298 -0
- package/src/kernels/js/rejection-reports.js +61 -0
- package/src/kernels/js/run-queue.ts +7 -0
- package/src/kernels/js/worker-core.js +53 -8
- package/src/kernels/js/worker-cwd.js +5 -2
- package/src/kernels/js/worker-host.ts +9 -2
- package/src/kernels/js/worker-indirect-eval.js +37 -13
- package/src/kernels/js/worker-package-resolve.js +153 -0
- package/src/kernels/js/worker-require.d.ts +19 -0
- package/src/kernels/js/worker-require.js +52 -0
- package/src/kernels/js/worker-runtime.js +43 -9
- package/src/kernels/js/worker-shadow-guard.js +2 -6
- package/src/kernels/js/worker-shell-capture.js +5 -0
- package/src/kernels/js/worker-slot.ts +9 -0
- package/src/kernels/js/worker-startup.ts +49 -1
- package/src/kernels/js/worker-webview.js +15 -1
- package/src/kernels/py/host-cell.ts +8 -31
- package/src/kernels/py/prelude.py +33 -2
- package/src/kernels/py/process.ts +4 -0
- package/src/kernels/sandbox/sandbox-cell.ts +141 -0
- package/src/kernels/sandbox/vendor/pi-codemode/VENDORED.md +1 -0
- package/src/kernels/sandbox/vendor/pi-codemode/runtime/prelude-source.ts +14 -1
- package/src/kernels/sandbox/vendor/pi-codemode/runtime/worker.ts +4 -1
- package/src/kernels/sandbox/vendor/pi-codemode/types.ts +2 -0
- package/src/kernels/shared/host-cell.ts +57 -0
- package/src/kernels/shared/subprocess-process.ts +14 -0
- package/src/prompt/eval-prompt.ts +8 -1
- package/src/tool/AGENTS.md +1 -0
- package/src/tool/cell-handler.ts +12 -11
- package/src/tool/cell-reserved-dispatch.ts +1 -0
- package/src/tool/cell-runtime.ts +4 -2
- package/src/tool/detached-cell-contract.ts +3 -0
- package/src/tool/detached-cell-manager.ts +4 -8
- package/src/tool/detached-cell-notification.ts +11 -63
- package/src/tool/detached-cell-snapshot.ts +17 -0
- package/src/tool/detached-eval-result.ts +13 -7
- package/src/tool/detached-notification-queue.ts +1 -2
- package/src/tool/eval-request.ts +23 -1
- package/src/tool/eval-tool-options.ts +3 -0
- package/src/tool/eval-tool.ts +17 -4
- package/src/tool/interrupt-note.ts +4 -5
- package/src/tool/live-headline.ts +60 -0
- package/src/tool/load-cell.ts +18 -3
- package/src/tool/magic-cell-host.ts +60 -2
- package/src/tool/magic-cells.ts +12 -2
- package/src/tool/managed-cell.ts +0 -4
- package/src/tool/render.ts +74 -15
- package/src/tool/run-eval-cell.ts +46 -4
- package/src/tool/runtime-label.ts +2 -1
- package/src/tool/types.ts +23 -1
|
@@ -1,61 +1,21 @@
|
|
|
1
|
-
import { mkdir, writeFile } from "node:fs/promises";
|
|
2
|
-
import { dirname, join } from "node:path";
|
|
3
1
|
import type { EvalDetachedCellNotification, EvalDetachedCellSnapshot } from "./detached-cell-manager.ts";
|
|
4
2
|
import { interruptionStateNote, unknownInterruptionStateNote } from "./interrupt-note.ts";
|
|
5
3
|
import type { EvalKernelState, EvalMemoryDetails } from "./types.ts";
|
|
6
4
|
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
export async function buildDetachedCellNotification(
|
|
15
|
-
snapshot: EvalDetachedCellSnapshot,
|
|
16
|
-
spillPath: string | undefined,
|
|
17
|
-
): Promise<EvalDetachedCellNotification> {
|
|
18
|
-
const body = notificationBody(snapshot);
|
|
19
|
-
const overflow = Buffer.byteLength(body, "utf8") > NOTIFICATION_TAIL_BYTES;
|
|
20
|
-
let spillNotice = "";
|
|
21
|
-
if (overflow && spillPath !== undefined) {
|
|
22
|
-
try {
|
|
23
|
-
await mkdir(dirname(spillPath), { recursive: true });
|
|
24
|
-
await writeFile(spillPath, body, "utf8");
|
|
25
|
-
// The agent read tool resolves plain paths only, so the notice must carry
|
|
26
|
-
// the absolute spill path, never the kernel-helper local:// scheme.
|
|
27
|
-
spillNotice = `\nBuffered output overflowed; full output: ${spillPath}`;
|
|
28
|
-
} catch (error) {
|
|
29
|
-
const message = error instanceof Error ? error.message : String(error);
|
|
30
|
-
spillNotice = `\nBuffered output overflow could not be spilled: ${message}`;
|
|
31
|
-
}
|
|
32
|
-
}
|
|
33
|
-
return {
|
|
34
|
-
cellId: snapshot.cellId,
|
|
35
|
-
content: `${overflow ? notificationPreview(snapshot) : body}${overflow ? "\n[…notification tail capped…]" : ""}${spillNotice}`,
|
|
36
|
-
};
|
|
37
|
-
}
|
|
38
|
-
|
|
39
|
-
function notificationBody(cell: EvalDetachedCellSnapshot): string {
|
|
40
|
-
const outcome = outcomeOf(cell);
|
|
41
|
-
const resultText = textContent(cell);
|
|
42
|
-
const stateNote = stateNoteOf(cell);
|
|
43
|
-
return [
|
|
44
|
-
`<system-reminder>Detached eval cell ${cell.cellId} (${cell.language}) ${outcome}.`,
|
|
45
|
-
resultText.length === 0 ? "(no output)" : resultText,
|
|
46
|
-
`${stateNote}</system-reminder>`,
|
|
47
|
-
].join("\n");
|
|
5
|
+
/**
|
|
6
|
+
* A detached cell's completion carries the same text its result would have shown in the foreground: the output sink
|
|
7
|
+
* already bounded it (the configured head, the tail, the middle marker and the artifact notice), plus its images.
|
|
8
|
+
*/
|
|
9
|
+
export function buildDetachedCellNotification(snapshot: EvalDetachedCellSnapshot): EvalDetachedCellNotification {
|
|
10
|
+
const images = snapshot.result.content.filter((part) => part.type === "image");
|
|
11
|
+
return { cellId: snapshot.cellId, content: notificationText(snapshot, textContent(snapshot)), images };
|
|
48
12
|
}
|
|
49
13
|
|
|
50
|
-
function
|
|
51
|
-
const outcome = outcomeOf(cell);
|
|
52
|
-
const resultText = textContent(cell);
|
|
53
|
-
const stateNote = stateNoteOf(cell);
|
|
14
|
+
function notificationText(cell: EvalDetachedCellSnapshot, output: string): string {
|
|
54
15
|
return [
|
|
55
|
-
`<system-reminder>Detached eval cell ${cell.cellId} (${cell.language}) ${
|
|
56
|
-
"
|
|
57
|
-
|
|
58
|
-
`${stateNote}</system-reminder>`,
|
|
16
|
+
`<system-reminder>Detached eval cell ${cell.cellId} (${cell.language}) ${outcomeOf(cell)}.`,
|
|
17
|
+
output.length === 0 ? "(no output)" : output,
|
|
18
|
+
`${stateNoteOf(cell)}</system-reminder>`,
|
|
59
19
|
].join("\n");
|
|
60
20
|
}
|
|
61
21
|
|
|
@@ -101,15 +61,3 @@ export function memoryStateNote(memory: EvalMemoryDetails | undefined): string {
|
|
|
101
61
|
return "The kernel was restarted before this cell ran - globals from earlier cells are gone; this cell's variables are available to the next eval cell.";
|
|
102
62
|
return "Kernel state updated - variables are available to the next eval cell.";
|
|
103
63
|
}
|
|
104
|
-
|
|
105
|
-
function safeCellId(cellId: string): string {
|
|
106
|
-
return cellId.replace(/[^a-zA-Z0-9_-]/gu, "_");
|
|
107
|
-
}
|
|
108
|
-
|
|
109
|
-
function truncateTailUtf8(text: string, maxBytes: number): string {
|
|
110
|
-
if (Buffer.byteLength(text, "utf8") <= maxBytes) return text;
|
|
111
|
-
const bytes = Buffer.from(text, "utf8");
|
|
112
|
-
let start = bytes.length - maxBytes;
|
|
113
|
-
while (start < bytes.length && (bytes[start] & 0xc0) === 0x80) start++;
|
|
114
|
-
return bytes.subarray(start).toString("utf8");
|
|
115
|
-
}
|
|
@@ -63,6 +63,23 @@ export function currentDetachedResult(cell: DetachedCellResultSource): AgentTool
|
|
|
63
63
|
return cell.terminalResult ?? cell.liveResult?.() ?? fallbackResult(cell.input);
|
|
64
64
|
}
|
|
65
65
|
|
|
66
|
+
/**
|
|
67
|
+
* The result a cell cancelled before its handler finished settles with: its buffered output, not the live progress
|
|
68
|
+
* frame ("1/1 cells running"), so a stopped cell never reads as still running.
|
|
69
|
+
*/
|
|
70
|
+
export function cancelledDetachedResult(cell: DetachedCellResultSource): AgentToolResult<EvalToolDetails> {
|
|
71
|
+
if (cell.terminalResult !== undefined) return cell.terminalResult;
|
|
72
|
+
const current = currentDetachedResult(cell);
|
|
73
|
+
const output = detachedOutputTail(current);
|
|
74
|
+
return {
|
|
75
|
+
content: [
|
|
76
|
+
{ type: "text", text: output.length > 0 ? output : "(no output)" },
|
|
77
|
+
...current.content.filter((part) => part.type === "image"),
|
|
78
|
+
],
|
|
79
|
+
details: current.details,
|
|
80
|
+
};
|
|
81
|
+
}
|
|
82
|
+
|
|
66
83
|
export function detachedErrorResult(cell: DetachedCellResultSource, error: Error): AgentToolResult<EvalToolDetails> {
|
|
67
84
|
const current = currentDetachedResult(cell);
|
|
68
85
|
const output = detachedOutputTail(current);
|
|
@@ -46,9 +46,7 @@ function createEvalListResult(cellManager: EvalDetachedCellManager): AgentToolRe
|
|
|
46
46
|
language: snapshot.language,
|
|
47
47
|
state: snapshot.state,
|
|
48
48
|
startedAtMs: snapshot.startedAtMs,
|
|
49
|
-
...(snapshot.queuedBehind === undefined
|
|
50
|
-
? {}
|
|
51
|
-
: { queuedBehind: [...snapshot.queuedBehind] }),
|
|
49
|
+
...(snapshot.queuedBehind === undefined ? {} : { queuedBehind: [...snapshot.queuedBehind] }),
|
|
52
50
|
...(summary ? { summary } : {}),
|
|
53
51
|
};
|
|
54
52
|
});
|
|
@@ -60,7 +58,7 @@ function createEvalListResult(cellManager: EvalDetachedCellManager): AgentToolRe
|
|
|
60
58
|
" ",
|
|
61
59
|
);
|
|
62
60
|
const elapsed = Math.floor(snapshot.result.details.durationMs / 1000);
|
|
63
|
-
const queued = cell.queuedBehind === undefined ? "" : `
|
|
61
|
+
const queued = cell.queuedBehind === undefined ? "" : ` ${queuedPhrase(cell.queuedBehind, cell.language)}`;
|
|
64
62
|
return `${cell.cellId} ${cell.language} ${cell.state} ${elapsed}s${queued} - ${preview}`;
|
|
65
63
|
})
|
|
66
64
|
.join("\n");
|
|
@@ -76,11 +74,13 @@ export function resultAfterDetach(
|
|
|
76
74
|
otherLiveCells: number,
|
|
77
75
|
): AgentToolResult<EvalToolDetails> {
|
|
78
76
|
if (snapshot.state !== "detached" && snapshot.state !== "running") return createDetachedControlResult(snapshot);
|
|
79
|
-
const
|
|
77
|
+
const queuedBehind = snapshot.queuedBehind;
|
|
80
78
|
const text =
|
|
81
|
-
|
|
79
|
+
queuedBehind === undefined
|
|
82
80
|
? `Eval cell ${snapshot.cellId} detached and is running in the ${input.language} kernel (${otherLiveCells} other live cells). Completion arrives as a notification; do not re-run it. eval({ action: "peek" | "stop", cell_id }) or eval({ action: "list" }).`
|
|
83
|
-
:
|
|
81
|
+
: queuedBehind.length === 0
|
|
82
|
+
? `Eval cell ${snapshot.cellId} is detached and waiting for the ${input.language} kernel to be ready; it runs first once the kernel is ready and completes as one notification. peek/stop/list with eval({ action, cell_id })`
|
|
83
|
+
: `Eval cell ${snapshot.cellId} is queued behind ${queuedBehind.join(", ")} in the ${input.language} kernel and detached; it runs after ${queuedBehind.join(", ")} and completes as one notification. peek/stop/list with eval({ action, cell_id })`;
|
|
84
84
|
return {
|
|
85
85
|
content: [
|
|
86
86
|
{
|
|
@@ -164,6 +164,12 @@ export function resultForDetachedState(
|
|
|
164
164
|
};
|
|
165
165
|
}
|
|
166
166
|
|
|
167
|
+
function queuedPhrase(queuedBehind: readonly string[], language: string): string {
|
|
168
|
+
return queuedBehind.length === 0
|
|
169
|
+
? `waiting for the ${language} kernel to be ready`
|
|
170
|
+
: `queued behind ${queuedBehind.join(", ")}`;
|
|
171
|
+
}
|
|
172
|
+
|
|
167
173
|
function textContent(result: AgentToolResult<EvalToolDetails>): string {
|
|
168
174
|
return result.content
|
|
169
175
|
.filter((part) => part.type === "text")
|
|
@@ -3,7 +3,6 @@ import { buildDetachedCellNotification } from "./detached-cell-notification.ts";
|
|
|
3
3
|
|
|
4
4
|
export interface PendingDetachedNotification {
|
|
5
5
|
readonly snapshot: () => EvalDetachedCellSnapshot | Promise<EvalDetachedCellSnapshot>;
|
|
6
|
-
readonly spillPath: string | undefined;
|
|
7
6
|
}
|
|
8
7
|
|
|
9
8
|
export class DetachedNotificationQueue {
|
|
@@ -30,7 +29,7 @@ export class DetachedNotificationQueue {
|
|
|
30
29
|
const flush = Promise.resolve().then(async () => {
|
|
31
30
|
const pending = this.#pending.splice(0);
|
|
32
31
|
const notifications = await Promise.all(
|
|
33
|
-
pending.map(async (item) =>
|
|
32
|
+
pending.map(async (item) => buildDetachedCellNotification(await item.snapshot())),
|
|
34
33
|
);
|
|
35
34
|
this.#notifier?.notify(notifications);
|
|
36
35
|
});
|
package/src/tool/eval-request.ts
CHANGED
|
@@ -17,9 +17,19 @@ export function normalizeEvalSummary(value: unknown): string | undefined {
|
|
|
17
17
|
return normalized.length === 0 ? undefined : normalized;
|
|
18
18
|
}
|
|
19
19
|
|
|
20
|
+
export class EvalIsolateInvalidError extends TypeError {
|
|
21
|
+
readonly name = "EvalIsolateInvalidError";
|
|
22
|
+
readonly code = "eval_isolate_invalid";
|
|
23
|
+
|
|
24
|
+
constructor(reason: string) {
|
|
25
|
+
super(`eval_isolate_invalid: ${reason}`);
|
|
26
|
+
}
|
|
27
|
+
}
|
|
28
|
+
|
|
20
29
|
export function parseEvalRequest(
|
|
21
30
|
params: unknown,
|
|
22
31
|
enabledLanguages: readonly EvalLanguage[] = evalLanguageOrder,
|
|
32
|
+
options: { readonly sandbox?: boolean } = {},
|
|
23
33
|
): EvalToolRequest {
|
|
24
34
|
if (!isRecord(params)) throw new TypeError("eval parameters must be an object");
|
|
25
35
|
if (params.action === "list") return { action: "list" };
|
|
@@ -30,7 +40,9 @@ export function parseEvalRequest(
|
|
|
30
40
|
}
|
|
31
41
|
if (params.action !== undefined && params.action !== "run")
|
|
32
42
|
throw new TypeError(`Unknown eval action "${String(params.action)}"`);
|
|
33
|
-
if (
|
|
43
|
+
if (params.language === undefined) throw new TypeError(evalRunRequiresLanguageMessage(enabledLanguages));
|
|
44
|
+
if (!isEvalLanguage(params.language))
|
|
45
|
+
throw new TypeError(`eval run language must be one of: ${enabledLanguages.join(", ")}`);
|
|
34
46
|
if (typeof params.code !== "string")
|
|
35
47
|
throw new TypeError("eval run requires code — the cell body to execute, verbatim");
|
|
36
48
|
const summary = normalizeEvalSummary(params.summary);
|
|
@@ -40,6 +52,15 @@ export function parseEvalRequest(
|
|
|
40
52
|
);
|
|
41
53
|
if (params.on_timeout !== undefined && params.on_timeout !== "detach" && params.on_timeout !== "error")
|
|
42
54
|
throw new TypeError(`Unknown eval on_timeout value "${String(params.on_timeout)}"`);
|
|
55
|
+
if (params.isolate !== undefined && params.isolate !== false) {
|
|
56
|
+
if (params.isolate !== true) throw new EvalIsolateInvalidError("isolate must be true or false");
|
|
57
|
+
if (options.sandbox !== true) {
|
|
58
|
+
throw new EvalIsolateInvalidError("sandbox cells are turned off (set sandbox.enabled in codemode settings)");
|
|
59
|
+
}
|
|
60
|
+
if (params.language !== "js")
|
|
61
|
+
throw new EvalIsolateInvalidError('sandbox cells run JavaScript only (language: "js")');
|
|
62
|
+
if (params.reset === true) throw new EvalIsolateInvalidError("a sandbox cell starts fresh; reset does not apply");
|
|
63
|
+
}
|
|
43
64
|
return {
|
|
44
65
|
language: params.language,
|
|
45
66
|
code: params.code,
|
|
@@ -48,6 +69,7 @@ export function parseEvalRequest(
|
|
|
48
69
|
...(typeof params.timeout === "number" ? { timeout: params.timeout } : {}),
|
|
49
70
|
...(params.on_timeout === "detach" || params.on_timeout === "error" ? { on_timeout: params.on_timeout } : {}),
|
|
50
71
|
...(typeof params.reset === "boolean" ? { reset: params.reset } : {}),
|
|
72
|
+
...(params.isolate === true ? { isolate: true } : {}),
|
|
51
73
|
};
|
|
52
74
|
}
|
|
53
75
|
|
|
@@ -8,6 +8,7 @@ import type {
|
|
|
8
8
|
import type { EvalSchemaToolInfo } from "../bridges/schema-bridge.ts";
|
|
9
9
|
import type { CompletionRequest, CompletionResult } from "../completion/handler.ts";
|
|
10
10
|
import type { ResolvedCodemodeSettings } from "../config/settings.ts";
|
|
11
|
+
import type { JsEnvironments } from "../environments/js-environments.ts";
|
|
11
12
|
import type { PythonEnvironments } from "../environments/python-environments.ts";
|
|
12
13
|
import type { EvalExecutionTracker } from "../extension/session-manager.ts";
|
|
13
14
|
import type { HandleRegistry } from "../handles/handle-registry.ts";
|
|
@@ -72,6 +73,8 @@ export interface CreateEvalToolOptions {
|
|
|
72
73
|
readonly kernelPreludes?: () => readonly KernelPreludeContribution[];
|
|
73
74
|
/** The session's Python environments: `%pip` / `%environment` cells and the import root of every Python cell. */
|
|
74
75
|
readonly pythonEnvironments?: PythonEnvironments;
|
|
76
|
+
/** The session's managed JavaScript packages: `%bun add` / `%npm add` cells and the bare-import fallback root. */
|
|
77
|
+
readonly jsEnvironments?: JsEnvironments;
|
|
75
78
|
/** Contributions whose documentation lines the description lists; snapshot taken when the tool is (re)registered. */
|
|
76
79
|
readonly promptKernelPreludes?: readonly KernelPreludeContribution[];
|
|
77
80
|
}
|
package/src/tool/eval-tool.ts
CHANGED
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import type { AgentToolResult, ToolDefinition } from "@code-yeongyu/senpi";
|
|
2
2
|
import type { TUnsafe } from "typebox";
|
|
3
|
+
import { resolveAdvertiseHelpers, resolveSandbox } from "../config/feature-settings.ts";
|
|
3
4
|
import {
|
|
4
5
|
DEFAULT_FOREGROUND_WINDOW_SECONDS,
|
|
5
6
|
defaultCodemodeSettings,
|
|
@@ -8,7 +9,12 @@ import {
|
|
|
8
9
|
import { buildEvalPrompt } from "../prompt/eval-prompt.ts";
|
|
9
10
|
import { EvalDetachedCellManager } from "./detached-cell-manager.ts";
|
|
10
11
|
import { executeEvalControl } from "./detached-eval-result.ts";
|
|
11
|
-
import {
|
|
12
|
+
import {
|
|
13
|
+
EvalIsolateInvalidError,
|
|
14
|
+
isEvalControlRequest,
|
|
15
|
+
normalizeEvalSummary,
|
|
16
|
+
parseEvalRequest,
|
|
17
|
+
} from "./eval-request.ts";
|
|
12
18
|
import type { CreateEvalToolOptions } from "./eval-tool-options.ts";
|
|
13
19
|
import { runEvalCell } from "./run-eval-cell.ts";
|
|
14
20
|
import {
|
|
@@ -43,9 +49,11 @@ export function createEvalTool(options: CreateEvalToolOptions) {
|
|
|
43
49
|
foregroundWindowSeconds,
|
|
44
50
|
hardLimitSeconds: options.hardLimitSeconds ?? defaultEvalDeadlineSeconds.hardLimitSeconds,
|
|
45
51
|
};
|
|
46
|
-
const
|
|
52
|
+
const sandbox = resolveSandbox(options.settings ?? defaultCodemodeSettings).enabled;
|
|
53
|
+
const parameters = createEvalInputSchema(options.enabledLanguages, deadlines, { sandbox });
|
|
47
54
|
const prompt = buildEvalPrompt(options.enabledLanguages, {
|
|
48
55
|
spawns: options.spawns ?? false,
|
|
56
|
+
...(resolveAdvertiseHelpers(options.settings ?? defaultCodemodeSettings) ? { advertiseHelpers: true } : {}),
|
|
49
57
|
monitor: options.monitor,
|
|
50
58
|
maxDetachedCells,
|
|
51
59
|
runBudgetSeconds: deadlines.runBudgetSeconds,
|
|
@@ -74,9 +82,14 @@ export function createEvalTool(options: CreateEvalToolOptions) {
|
|
|
74
82
|
async function execute(
|
|
75
83
|
...[toolCallId, params, signal, onUpdate, ctx]: EvalExecuteArgs<EvalToolRequest>
|
|
76
84
|
): Promise<AgentToolResult<EvalResultDetails>> {
|
|
77
|
-
const request = parseEvalRequest(params, languages);
|
|
85
|
+
const request = parseEvalRequest(params, languages, { sandbox });
|
|
78
86
|
if (isEvalControlRequest(request)) return await executeEvalControl(cellManager, request);
|
|
79
|
-
if (options.proxyExecutor)
|
|
87
|
+
if (options.proxyExecutor) {
|
|
88
|
+
// A proxy runs cells elsewhere and knows nothing of sandbox cells: an isolated cell must never run unisolated.
|
|
89
|
+
if ("isolate" in request && request.isolate === true)
|
|
90
|
+
throw new EvalIsolateInvalidError("isolate: true is not available through this eval proxy");
|
|
91
|
+
return await options.proxyExecutor(request, signal);
|
|
92
|
+
}
|
|
80
93
|
if (!languages.includes(request.language))
|
|
81
94
|
throw new RangeError(
|
|
82
95
|
`Unsupported eval language "${request.language}". Enabled languages: ${languages.join(", ")}`,
|
|
@@ -26,9 +26,8 @@ export async function describeTimeoutState(
|
|
|
26
26
|
);
|
|
27
27
|
if (outcome === undefined) error.message = fallbackTimeoutMessage(error.message);
|
|
28
28
|
else if (outcome.retained)
|
|
29
|
-
error.message = `${error.message} The kernel
|
|
30
|
-
else
|
|
31
|
-
error.message = `${error.message} The kernel was unresponsive and restarted; variables from earlier cells are lost.`;
|
|
29
|
+
error.message = `${error.message} The kernel was not restarted; variables from earlier cells are kept.`;
|
|
30
|
+
else error.message = `${error.message} The kernel was restarted; variables from earlier cells are lost.`;
|
|
32
31
|
if (outcome?.note !== undefined) error.message = `${error.message} ${outcome.note.trim()}`;
|
|
33
32
|
return error;
|
|
34
33
|
}
|
|
@@ -63,8 +62,8 @@ const LANGUAGE_LABEL: Record<EvalLanguage, string> = {
|
|
|
63
62
|
export function interruptionStateNote(language: EvalLanguage, stateRetained: boolean | undefined): string | undefined {
|
|
64
63
|
if (stateRetained === undefined) return undefined;
|
|
65
64
|
const label = LANGUAGE_LABEL[language];
|
|
66
|
-
if (stateRetained) return
|
|
67
|
-
return
|
|
65
|
+
if (stateRetained) return `The ${label} was not restarted; variables from earlier cells are kept.`;
|
|
66
|
+
return `The ${label} was restarted; variables from earlier cells are lost.`;
|
|
68
67
|
}
|
|
69
68
|
|
|
70
69
|
export function unknownInterruptionStateNote(language: EvalLanguage): string {
|
|
@@ -0,0 +1,60 @@
|
|
|
1
|
+
import { sanitizeTerminalLabel, visibleWidth } from "@code-yeongyu/senpi";
|
|
2
|
+
import { type EvalCellResult, type EvalLanguage, evalLanguageOrder } from "./types.ts";
|
|
3
|
+
|
|
4
|
+
const HEADLINED_STATUSES: ReadonlySet<EvalCellResult["status"]> = new Set(["pending", "queued", "running", "detached"]);
|
|
5
|
+
|
|
6
|
+
/**
|
|
7
|
+
* A collapsed headline never wraps the frame: it gets what the language, state and timing leave on the line. This
|
|
8
|
+
* floor (in screen cells) only keeps a word or two visible in very narrow terminals.
|
|
9
|
+
*/
|
|
10
|
+
const MIN_HEADLINE_CELLS = 8;
|
|
11
|
+
|
|
12
|
+
export function leadsWithHeadline(status: EvalCellResult["status"]): boolean {
|
|
13
|
+
return HEADLINED_STATUSES.has(status);
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
/**
|
|
17
|
+
* What an in-progress cell is doing, in one line: its summary, or else the first line of its code that has content once sanitized (the
|
|
18
|
+
* oh-my-pi fallback), or an ellipsis while the arguments are still streaming in. Collapsed rows cut it to
|
|
19
|
+
* `maxCells` screen cells (a wide character takes two), so the headline stays one line. Whichever is chosen is
|
|
20
|
+
* sanitized before it is measured, so escape and control characters never reach the row (senpi#2831, senpi#2839).
|
|
21
|
+
*/
|
|
22
|
+
export function liveHeadline(
|
|
23
|
+
summary: string | undefined,
|
|
24
|
+
code: string | undefined,
|
|
25
|
+
maxCells: number | undefined,
|
|
26
|
+
): string {
|
|
27
|
+
const text = (summary === undefined ? "" : sanitizeTerminalLabel(summary)) || firstCodeLine(code) || "…";
|
|
28
|
+
const limit = maxCells === undefined ? undefined : Math.max(MIN_HEADLINE_CELLS, maxCells);
|
|
29
|
+
if (limit === undefined || visibleWidth(text) <= limit) return text;
|
|
30
|
+
return `${cellPrefix(text, limit - 1)}…`;
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
// A plain cut by screen cells: the headline is styled by the caller, so no reset codes may be inserted here.
|
|
34
|
+
function cellPrefix(text: string, cells: number): string {
|
|
35
|
+
let kept = "";
|
|
36
|
+
let used = 0;
|
|
37
|
+
for (const { segment } of new Intl.Segmenter(undefined, { granularity: "grapheme" }).segment(text)) {
|
|
38
|
+
const width = visibleWidth(segment);
|
|
39
|
+
if (used + width > cells) break;
|
|
40
|
+
kept += segment;
|
|
41
|
+
used += width;
|
|
42
|
+
}
|
|
43
|
+
return kept;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
export function knownLanguage(language: unknown): EvalLanguage | undefined {
|
|
47
|
+
return evalLanguageOrder.find((known) => known === language);
|
|
48
|
+
}
|
|
49
|
+
|
|
50
|
+
// The first line with something to show once it is sanitized: a line of only escape or control characters is skipped
|
|
51
|
+
// (senpi#2850), and the later lines never reach the row.
|
|
52
|
+
function firstCodeLine(code: string | undefined): string | undefined {
|
|
53
|
+
if (code === undefined) return undefined;
|
|
54
|
+
// Stops at the first line with content, so a long cell is not sanitized line by line on every frame.
|
|
55
|
+
for (const line of code.split("\n")) {
|
|
56
|
+
const clean = sanitizeTerminalLabel(line);
|
|
57
|
+
if (clean.length > 0) return clean;
|
|
58
|
+
}
|
|
59
|
+
return undefined;
|
|
60
|
+
}
|
package/src/tool/load-cell.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { closeSync, constants, fstatSync, openSync,
|
|
1
|
+
import { closeSync, constants, fstatSync, openSync, readSync } from "node:fs";
|
|
2
2
|
import { isAbsolute, join, relative, resolve } from "node:path";
|
|
3
3
|
import { fileURLToPath } from "node:url";
|
|
4
4
|
|
|
@@ -50,6 +50,18 @@ function resolveTarget(target: string, options: LoadCellOptions): { path: string
|
|
|
50
50
|
* Reads a `%load` target when its cell's turn comes in the kernel's queue, so a file the previous cell wrote is
|
|
51
51
|
* there. Every failure is a refusal that names the target as the user wrote it, never an absolute path.
|
|
52
52
|
*/
|
|
53
|
+
/** Reads from `fd` until EOF or `limit` bytes, whichever comes first; a file still growing cannot exceed it. */
|
|
54
|
+
export function readAtMost(fd: number, limit: number): Buffer {
|
|
55
|
+
const buffer = Buffer.alloc(limit);
|
|
56
|
+
let filled = 0;
|
|
57
|
+
while (filled < limit) {
|
|
58
|
+
const read = readSync(fd, buffer, filled, limit - filled, null);
|
|
59
|
+
if (read === 0) break;
|
|
60
|
+
filled += read;
|
|
61
|
+
}
|
|
62
|
+
return buffer.subarray(0, filled);
|
|
63
|
+
}
|
|
64
|
+
|
|
53
65
|
export function loadCell(target: string, options: LoadCellOptions): LoadedCell {
|
|
54
66
|
const resolved = resolveTarget(target, options);
|
|
55
67
|
if ("message" in resolved) return { ok: false, message: resolved.message };
|
|
@@ -60,8 +72,11 @@ export function loadCell(target: string, options: LoadCellOptions): LoadedCell {
|
|
|
60
72
|
fd = openSync(resolved.path, constants.O_RDONLY | constants.O_NONBLOCK);
|
|
61
73
|
const info = fstatSync(fd);
|
|
62
74
|
if (!info.isFile()) return { ok: false, message: `not a file: ${target}` };
|
|
63
|
-
|
|
64
|
-
|
|
75
|
+
const tooLarge = { ok: false as const, message: `%load reads files up to 8 MiB: ${target}` };
|
|
76
|
+
if (info.size > MAX_LOAD_BYTES) return tooLarge;
|
|
77
|
+
const bytes = readAtMost(fd, MAX_LOAD_BYTES + 1);
|
|
78
|
+
if (bytes.length > MAX_LOAD_BYTES) return tooLarge;
|
|
79
|
+
return { ok: true, code: bytes.toString("utf8"), sourceFile: resolved.path };
|
|
65
80
|
} catch (error) {
|
|
66
81
|
const code = errorCode(error);
|
|
67
82
|
if (code === "ENOENT" || code === "ENOTDIR") return { ok: false, message: `file not found: ${target}` };
|
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import type { JsEnvironments } from "../environments/js-environments.ts";
|
|
1
2
|
import type { PythonEnvironments } from "../environments/python-environments.ts";
|
|
2
3
|
import { TIMEOUT_PAUSE_OP, TIMEOUT_RESUME_OP } from "../timeouts/bridge-timeout.ts";
|
|
3
4
|
import { MagicCellError, parseMagicCell } from "./magic-cells.ts";
|
|
@@ -13,6 +14,7 @@ export function planMagicCell(
|
|
|
13
14
|
language: EvalLanguage,
|
|
14
15
|
code: string,
|
|
15
16
|
environments: PythonEnvironments | undefined,
|
|
17
|
+
jsEnvironments?: JsEnvironments,
|
|
16
18
|
): MagicCellPlan {
|
|
17
19
|
let magic: ReturnType<typeof parseMagicCell>;
|
|
18
20
|
try {
|
|
@@ -23,6 +25,7 @@ export function planMagicCell(
|
|
|
23
25
|
}
|
|
24
26
|
if (magic === undefined) return { kind: "ordinary" };
|
|
25
27
|
if (magic.kind === "load") return { kind: "load", target: magic.target };
|
|
28
|
+
if (language === "js") return planJsMagic(magic, jsEnvironments);
|
|
26
29
|
if (environments === undefined) {
|
|
27
30
|
return {
|
|
28
31
|
kind: "refused",
|
|
@@ -34,8 +37,9 @@ export function planMagicCell(
|
|
|
34
37
|
return {
|
|
35
38
|
kind: "host",
|
|
36
39
|
executor: async () => {
|
|
37
|
-
|
|
38
|
-
|
|
40
|
+
await environments.setMode(mode);
|
|
41
|
+
const where = mode === "project" ? "the session directory" : "this session's managed revisions";
|
|
42
|
+
return { ok: true, valueRepr: `environment: ${mode} (${where})` };
|
|
39
43
|
},
|
|
40
44
|
};
|
|
41
45
|
}
|
|
@@ -63,3 +67,57 @@ export function planMagicCell(
|
|
|
63
67
|
},
|
|
64
68
|
};
|
|
65
69
|
}
|
|
70
|
+
|
|
71
|
+
function planJsMagic(
|
|
72
|
+
magic: Exclude<NonNullable<ReturnType<typeof parseMagicCell>>, { kind: "load" }>,
|
|
73
|
+
environments: JsEnvironments | undefined,
|
|
74
|
+
): MagicCellPlan {
|
|
75
|
+
if (environments === undefined) {
|
|
76
|
+
return {
|
|
77
|
+
kind: "refused",
|
|
78
|
+
message: "environment_installer_unavailable: this session has no JavaScript package environment",
|
|
79
|
+
};
|
|
80
|
+
}
|
|
81
|
+
if (magic.kind === "environment") {
|
|
82
|
+
const mode = magic.mode;
|
|
83
|
+
return {
|
|
84
|
+
kind: "host",
|
|
85
|
+
executor: async () => {
|
|
86
|
+
await environments.setMode(mode);
|
|
87
|
+
const where = mode === "project" ? "the session directory" : "this session's managed revisions";
|
|
88
|
+
return { ok: true, valueRepr: `environment: ${mode} (${where})` };
|
|
89
|
+
},
|
|
90
|
+
};
|
|
91
|
+
}
|
|
92
|
+
if (magic.kind !== "js-add") return { kind: "refused", message: "this magic is not available in JavaScript" };
|
|
93
|
+
const requested = magic.args;
|
|
94
|
+
return {
|
|
95
|
+
kind: "host",
|
|
96
|
+
executor: async ({ signal, emit }) => {
|
|
97
|
+
emit({ type: "status", event: { op: TIMEOUT_PAUSE_OP } });
|
|
98
|
+
try {
|
|
99
|
+
const receipt = await environments.install(
|
|
100
|
+
requested,
|
|
101
|
+
signal,
|
|
102
|
+
(stream, data) => emit({ type: "text", stream, data }),
|
|
103
|
+
magic.installer,
|
|
104
|
+
);
|
|
105
|
+
const added = receipt.added.length > 0 ? receipt.added.join(", ") : "nothing new";
|
|
106
|
+
const where =
|
|
107
|
+
receipt.revision === undefined ? receipt.mode : `${receipt.mode} (revision ${receipt.revision})`;
|
|
108
|
+
const shadow =
|
|
109
|
+
receipt.shadowed.length > 0
|
|
110
|
+
? `; environment_resolution_conflict: ${receipt.shadowed.join(", ")} still ${receipt.shadowed.length === 1 ? "resolves" : "resolve"} from the project's node_modules first`
|
|
111
|
+
: "";
|
|
112
|
+
return {
|
|
113
|
+
ok: true,
|
|
114
|
+
valueRepr: `added ${added} with ${receipt.installer} into ${where}${shadow}; already-imported modules stay cached until reset`,
|
|
115
|
+
};
|
|
116
|
+
} catch (error) {
|
|
117
|
+
return { ok: false, error: { message: error instanceof Error ? error.message : String(error) } };
|
|
118
|
+
} finally {
|
|
119
|
+
emit({ type: "status", event: { op: TIMEOUT_RESUME_OP } });
|
|
120
|
+
}
|
|
121
|
+
},
|
|
122
|
+
};
|
|
123
|
+
}
|
package/src/tool/magic-cells.ts
CHANGED
|
@@ -3,9 +3,10 @@ import type { EvalLanguage } from "./types.ts";
|
|
|
3
3
|
export type MagicCell =
|
|
4
4
|
| { readonly kind: "pip"; readonly args: string }
|
|
5
5
|
| { readonly kind: "environment"; readonly mode: "managed" | "project" }
|
|
6
|
-
| { readonly kind: "load"; readonly target: string }
|
|
6
|
+
| { readonly kind: "load"; readonly target: string }
|
|
7
|
+
| { readonly kind: "js-add"; readonly installer: "bun" | "npm"; readonly args: string };
|
|
7
8
|
|
|
8
|
-
const HOST_MAGICS = ["pip", "environment", "load"] as const;
|
|
9
|
+
const HOST_MAGICS = ["pip", "environment", "load", "bun", "npm"] as const;
|
|
9
10
|
const LOAD_LANGUAGES: ReadonlySet<EvalLanguage> = new Set(["py", "js"]);
|
|
10
11
|
const COMMENT_PREFIX: Partial<Record<EvalLanguage, string>> = { py: "#", js: "//" };
|
|
11
12
|
type HostMagic = (typeof HOST_MAGICS)[number];
|
|
@@ -34,6 +35,8 @@ function hostMagicOf(language: EvalLanguage, line: string): HostMagic | undefine
|
|
|
34
35
|
const name = match?.[1];
|
|
35
36
|
const magic = HOST_MAGICS.find((candidate) => candidate === name);
|
|
36
37
|
if (magic === "load") return LOAD_LANGUAGES.has(language) ? magic : undefined;
|
|
38
|
+
if (magic === "bun" || magic === "npm") return language === "js" ? magic : undefined;
|
|
39
|
+
if (magic === "environment") return language === "py" || language === "js" ? magic : undefined;
|
|
37
40
|
return language === "py" ? magic : undefined;
|
|
38
41
|
}
|
|
39
42
|
|
|
@@ -56,6 +59,13 @@ export function parseMagicCell(language: EvalLanguage, code: string): MagicCell
|
|
|
56
59
|
throw new MagicCellError(`put %${magic} on its own cell, then run the code that uses it in the next cell`);
|
|
57
60
|
const args = first.trim().slice(`%${magic}`.length).trim();
|
|
58
61
|
if (magic === "pip") return { kind: "pip", args };
|
|
62
|
+
if (magic === "bun" || magic === "npm") {
|
|
63
|
+
const [verb, ...rest] = args.split(/\s+/);
|
|
64
|
+
if (verb !== "add" && verb !== "install") {
|
|
65
|
+
throw new MagicCellError(`%${magic} supports only add: %${magic} add <package ...>`);
|
|
66
|
+
}
|
|
67
|
+
return { kind: "js-add", installer: magic, args: rest.join(" ") };
|
|
68
|
+
}
|
|
59
69
|
if (magic === "load") {
|
|
60
70
|
if (args === "") throw new MagicCellError("%load takes one argument: the path of the file to run");
|
|
61
71
|
return { kind: "load", target: args };
|
package/src/tool/managed-cell.ts
CHANGED
|
@@ -1,7 +1,6 @@
|
|
|
1
1
|
import type { AgentToolResult } from "@code-yeongyu/senpi";
|
|
2
2
|
import { type CellDeadlineExpiry, CellDeadlines } from "./cell-deadlines.ts";
|
|
3
3
|
import type { EvalDetachedCellSnapshot, EvalDetachedCellState } from "./detached-cell-contract.ts";
|
|
4
|
-
import { detachedNotificationSpillPath } from "./detached-cell-notification.ts";
|
|
5
4
|
import type { EvalKernel, EvalToolDetails, EvalToolInput } from "./types.ts";
|
|
6
5
|
|
|
7
6
|
export type LiveResultProvider = () => AgentToolResult<EvalToolDetails>;
|
|
@@ -9,7 +8,6 @@ export type LiveResultProvider = () => AgentToolResult<EvalToolDetails>;
|
|
|
9
8
|
export type ManagedCell = {
|
|
10
9
|
readonly cellId: string;
|
|
11
10
|
readonly input: EvalToolInput;
|
|
12
|
-
readonly spillPath: string | undefined;
|
|
13
11
|
readonly startedAtMs: number;
|
|
14
12
|
readonly terminal: PromiseWithResolvers<EvalDetachedCellSnapshot>;
|
|
15
13
|
state: Exclude<EvalDetachedCellState, "detached">;
|
|
@@ -37,7 +35,6 @@ export type ManagedCell = {
|
|
|
37
35
|
export interface ManagedCellInit {
|
|
38
36
|
readonly cellId: string;
|
|
39
37
|
readonly input: EvalToolInput;
|
|
40
|
-
readonly artifactsDir: string | undefined;
|
|
41
38
|
readonly now: () => number;
|
|
42
39
|
readonly defaultHardLimitSeconds: number;
|
|
43
40
|
readonly defaultRunBudgetSeconds: number;
|
|
@@ -60,7 +57,6 @@ export function createManagedCell(init: ManagedCellInit): ManagedCell {
|
|
|
60
57
|
return {
|
|
61
58
|
cellId: init.cellId,
|
|
62
59
|
input: init.input,
|
|
63
|
-
spillPath: detachedNotificationSpillPath(init.artifactsDir, init.cellId),
|
|
64
60
|
startedAtMs: init.now(),
|
|
65
61
|
state: "queued",
|
|
66
62
|
runStartedAtMs: undefined,
|