pi-roundtable-coding 0.9.3 → 0.9.5

package/CHANGELOG.md

@@ -2,6 +2,24 @@
 
 ## [Unreleased]
 
+## [0.9.5] - 2026-10-10
+
+0.9.4 was tagged but never published (its release checks failed on the recorded prompt texts); everything listed under it is released in 0.9.5.
+
+### Added
+
+- A coding worker posts its progress into its thread as it works, as the host's turns do: longer or structured text as ordinary messages, and short narration with the tools it called in one small-text progress message edited in place. Its approval cards come after the text written before them, the report is posted once at the end, and a failed post or edit is logged without failing the job. `interimText: "off"` (on `coding` and `CodingDeskOptions`) keeps today's thread of task, cards and report; `interimPrimaryChars` sets the length of primary text (default 400). It needs pi-roundtable 0.9.5's `DispatchThread.interim`.
+- `CodingWorker.run` takes an optional fourth argument, `CodingProgress` (`messageEnd(message)`, `toolStart(name)`), which `PiCodingWorker` feeds from the worker's session over its IPC channel (message text and tool names only, never thinking).
+
+### Changed
+
+- A worker's approval card no longer expires after 30 minutes: the worker waits for the answer, with no countdown on the card, until it is stopped; the waiting does not count against `timeoutMs`.
+
+### Fixed
+
+- A worker's shell gets the host's scratch dir (`env.scratchDir`): TMPDIR points to it, and writes, redirects and removals inside it run without the owner's approval, as in the agents' shell. Before, every scratch-dir command a worker ran asked the owner.
+- A worker whose host has a scratch dir is told it in its system prompt, after the host's own prompt or the default, so a task need not say where temporary files and test homes go.
+
 ## [0.9.3] - 2026-10-10
 
 - Release in lockstep with pi-roundtable 0.9.3; no package-specific behavior changes.

package/README.md

@@ -128,9 +128,13 @@ The parent consults `shellHoldRule`, additional `holds`, and the host's linked h
 When `threads` is configured, progress and cards stay in a thread opened under the resolved run's `origin`; if no thread can open, calls remain held rather than falling back to another channel's approval cards.
 Without `threads`, the resolved report channel's prompts are used.
 `threadText` controls the initial post, held-action post, approval title and final report; the thread is archived before `onResult` is called, and a failed thread post/close does not discard result delivery.
+While it works, the worker posts its progress into its thread, as the host's own turns do: text of `interimPrimaryChars` (400) characters or more, or written with a Markdown heading, list, table or code fence, as ordinary messages, and short narration with the tools it called in one small-text progress message edited in place.
+The thread's cards come after the text written before them, the final report is posted once, and a failed progress post or edit is logged and never fails the job.
+`interimText: "off"` posts only the task, the held-action notices, the cards and the report; a thread whose host cannot edit its posts gets no progress either.
 `promptSlot` tracks these cards and `workTimeout` excludes their waiting time from the worker's budget (one hour by default).
 `timeoutMs` must be positive and finite, and at most 2,147,483,647 ms to fit the host timer.
-An approved call runs; a declined, expired, missing or failed card blocks it and instructs the worker not to retry or work around the refusal.
+A worker's card has no grace period and no deadline: the worker waits for the answer until it is stopped, and the waiting does not count against its time.
+An approved call runs; a declined, missing or failed card blocks it and instructs the worker not to retry or work around the refusal.
 Unapproved actions appear in the report rather than running later automatically.
 By default worker answers are capped at 20,000 characters and the Held list keeps ten entries of at most 1,000 characters each, with explicit truncation and omission notices; `limits` changes each bound.
 A worker failure carries its exit category and numeric code; git and gh stderr and the worker's own error text reach the report only after `scrubDiagnostic` masks credentials, and are cut at `diagnosticChars`.
@@ -181,6 +185,8 @@ The `CODING` service exposes `shelf: RepoShelf` and `desk: CodingDesk` for trust
 | `toolText` | package wording | Trusted description and argument descriptions for each repository tool, as `{ repo_task: { description, parameters: { task: "…" } } }`; it never changes a tool's arguments or approval rules. |
 | `threads` | none | Public `DispatchThreads`-compatible progress and approval thread port. |
 | `threadText` | English package text | Thread introduction, held-action notice, approval title and final report. |
+| `interimText` | `"on"` | `"off"` keeps the worker's progress out of its thread; pass the host config's `interimText` to follow it. |
+| `interimPrimaryChars` | `400` | Length from which a worker's intermediate text is posted as its own message. |
 | `workerWorkspace` | individual clone | Trusted shell-policy write boundary; not an OS sandbox. |
 | `workerPrompt` | generic worker instructions | Trusted standing prompt replacement; it cannot bypass approval or cleanup. |
 | `workerBlockText` | "The owner declined / has not approved this call. Do not retry it or work around it; list it under Held in your report." | Trusted wording of what the worker reads when a call is declined or held, from `(answer, action)`; it never changes who is held, and the desk lists only unanswered calls as held. |

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-roundtable-coding",
-	"version": "0.9.3",
+	"version": "0.9.5",
 	"description": "Repository shelves and owner-approved Pi coding workers for pi-roundtable",
 	"type": "module",
 	"license": "MIT",
@@ -44,7 +44,7 @@
 		"@earendil-works/pi-ai": ">=1.0.0 <2",
 		"@biomejs/biome": "2.5.15",
 		"@types/bun": "1.4.2",
-		"pi-roundtable": "0.9.3",
+		"pi-roundtable": "0.9.5",
 		"typescript": "7.0.2"
 	}
 }

package/src/coding-desk.ts

@@ -1,6 +1,7 @@
 import type {
 	ChannelKey,
 	HeldCall,
+	InterimTextMode,
 	Logger,
 	OwnerPrompts,
 	ThinkingLevel,
@@ -10,6 +11,7 @@ import {
 	approvalCard,
 	type DispatchThread,
 	type DispatchThreads,
+	InterimPoster,
 	promptSlot,
 	workTimeout,
 } from "pi-roundtable/kit";
@@ -34,11 +36,24 @@ export interface CodingJob {
 	startedAt: Date;
 	startHead: string;
 }
+/** What the worker writes and runs as it goes, for its thread's progress posts. */
+export interface CodingProgress {
+	/** A finished message of the worker's session: its role, stop reason, and text. */
+	messageEnd(message: {
+		role: string;
+		content?: unknown;
+		stopReason?: string;
+	}): void;
+	/** A tool the worker started. */
+	toolStart(name: string): void;
+}
 export interface CodingWorker {
+	/** `progress`, when given, hears the session's messages and tools as the worker goes. */
 	run(
 		job: CodingJob & { dir: string },
 		signal: AbortSignal,
 		review: (call: HeldCall) => Promise<HeldCallAnswer>,
+		progress?: CodingProgress,
 	): Promise<string>;
 }
 export interface CodingResult {
@@ -65,6 +80,13 @@ export interface CodingDeskOptions {
 	logger: Logger;
 	timeoutMs?: number;
 	limits?: CodingLimits;
+	/**
+	 * Whether the worker posts the text it writes before its report into its thread as it goes,
+	 * as the host's turns do; "on" by default. A thread whose host cannot edit its posts gets none.
+	 */
+	interimText?: InterimTextMode;
+	/** Length from which an intermediate text is posted as its own message; default 400. */
+	interimPrimaryChars?: number;
 }
 /** What a coding report keeps of a long run; each is a count of characters or entries, or `Infinity` for no bound. */
 export interface CodingLimits {
@@ -212,6 +234,7 @@ export class CodingDesk {
 			this.#options.limits?.heldEntries ?? MAX_HELD_ENTRIES;
 		const maxHeldChars = this.#options.limits?.heldChars ?? MAX_HELD_CHARS;
 		let cancel = () => {};
+		let interim: InterimPoster | undefined;
 		let timedOut = false;
 		let outcome: CodingResult["outcome"];
 		try {
@@ -235,11 +258,22 @@ export class CodingDesk {
 			}
 			if (controller.signal.aborted)
 				throw new AgentError("the worker was stopped");
+			// The worker's progress goes to its thread as it works, under the report's name.
+			if (job.thread?.interim && this.#options.interimText !== "off")
+				interim = new InterimPoster(job.thread.interim, {
+					logger: this.#options.logger,
+					channel: job.thread.channel,
+					...(this.#options.interimPrimaryChars
+						? { primaryChars: this.#options.interimPrimaryChars }
+						: {}),
+				});
+			const poster = interim;
 			slot.bind(
 				threads
 					? job.thread && prompts?.(job.thread.channel)
 					: prompts?.(job.channel),
 				`Coding worker #${job.id}`,
+				poster ? () => poster.flush() : undefined,
 			);
 			cancel = workTimeout(timeoutMs, slot, () => {
 				timedOut = true;
@@ -268,6 +302,7 @@ export class CodingDesk {
 						);
 						held.push(entry);
 						try {
+							await interim?.flush();
 							await job.thread?.post(
 								threadText?.held?.(entry, job) ?? `Held: ${entry}`,
 							);
@@ -282,6 +317,7 @@ export class CodingDesk {
 				{ ...job, dir },
 				controller.signal,
 				review,
+				interim,
 			);
 			if (controller.signal.aborted)
 				throw new AgentError("the worker was stopped");
@@ -304,6 +340,8 @@ export class CodingDesk {
 		} finally {
 			cancel();
 			slot.unbind();
+			// What the worker wrote lands before its report.
+			await interim?.flush();
 		}
 		let state: RepoState | undefined;
 		let commits: string[] = [];

package/src/coding-plugin.ts

@@ -4,6 +4,7 @@ import {
 	defineTool,
 	type HoldCheck,
 	IDENTITY,
+	type InterimTextMode,
 	PluginError,
 	SKILLS,
 	SYSTEM_PRINCIPAL,
@@ -99,6 +100,14 @@ export interface CodingOptions {
 	toolText?: Partial<Record<RepoToolName, CodingToolText>>;
 	threads?: Pick<DispatchThreads, "open">;
 	threadText?: CodingThreadText;
+	/**
+	 * Whether a worker posts its progress into its thread as it goes: longer text as messages, short
+	 * narration and its tools in one edited small-text message. Default "on"; "off" posts only the
+	 * task and the report. Pass the host config's `interimText` to follow it.
+	 */
+	interimText?: InterimTextMode;
+	/** Length from which a worker's intermediate text is its own message; default 400. */
+	interimPrimaryChars?: number;
 	workerWorkspace?: string;
 	workerPrompt?: (dir: string) => string;
 	/** How much of a long run a report keeps; see `CodingLimits`. */
@@ -168,6 +177,9 @@ export function coding(options: CodingOptions) {
 					packages: options.workerPackages,
 					agentDir: options.agentDir,
 					workspace: options.workerWorkspace,
+					...(context.env.scratchDir
+						? { scratchDir: context.env.scratchDir }
+						: {}),
 					prompt: options.workerPrompt,
 					blockText: options.workerBlockText,
 					diagnosticChars: options.diagnosticChars,
@@ -183,6 +195,10 @@ export function coding(options: CodingOptions) {
 				logger: context.logger,
 				threads: options.threads,
 				threadText: options.threadText,
+				...(options.interimText ? { interimText: options.interimText } : {}),
+				...(options.interimPrimaryChars
+					? { interimPrimaryChars: options.interimPrimaryChars }
+					: {}),
 				prompts: (channel) => context.surfaces.prompts(channel),
 				deliver:
 					options.onResult ??

package/src/index.ts

@@ -2,6 +2,7 @@ export type {
 	CodingDeskOptions,
 	CodingJob,
 	CodingLimits,
+	CodingProgress,
 	CodingResult,
 	CodingThreadText,
 	CodingWorker,

package/src/pi-coding-worker.ts

@@ -2,7 +2,12 @@ import { fileURLToPath } from "node:url";
 import { getAgentDir } from "@earendil-works/pi-coding-agent";
 import type { HeldCall, HoldCheck } from "pi-roundtable";
 import { AgentError, canonicalJson, shellHoldRule } from "pi-roundtable/kit";
-import type { CodingJob, CodingWorker, HeldCallAnswer } from "./coding-desk.ts";
+import type {
+	CodingJob,
+	CodingProgress,
+	CodingWorker,
+	HeldCallAnswer,
+} from "./coding-desk.ts";
 import { CodingWorkerFailure } from "./worker-failure.ts";
 
 export interface PiCodingWorkerOptions {
@@ -14,6 +19,11 @@ export interface PiCodingWorkerOptions {
 	holds?: HoldCheck;
 	/** Trusted host policy boundary for writes; defaults to the individual clone. */
 	workspace?: string;
+	/**
+	 * The host's scratch dir: the worker's shell runs with TMPDIR pointing to it, and writes and
+	 * removals inside it run without a hold, as in the agents' shell.
+	 */
+	scratchDir?: string;
 	/** Trusted host standing prompt, replacing the generic worker instructions. */
 	prompt?: (dir: string) => string;
 	/**
@@ -31,6 +41,10 @@ interface WorkerMessage {
 	input?: Record<string, unknown>;
 	report?: string;
 	message?: string;
+	role?: string;
+	stopReason?: string;
+	text?: string;
+	name?: string;
 }
 function isMessage(value: unknown): value is WorkerMessage {
 	return (
@@ -51,6 +65,7 @@ export class PiCodingWorker implements CodingWorker {
 		job: CodingJob & { dir: string },
 		signal: AbortSignal,
 		review: (call: HeldCall) => Promise<HeldCallAnswer>,
+		progress?: CodingProgress,
 	): Promise<string> {
 		if (process.platform === "win32")
 			throw new AgentError("Coding workers require a POSIX host.");
@@ -65,6 +80,9 @@ export class PiCodingWorker implements CodingWorker {
 			],
 			{
 				cwd: job.dir,
+				...(this.#options.scratchDir
+					? { env: { ...process.env, TMPDIR: this.#options.scratchDir } }
+					: {}),
 				stdin: "ignore",
 				stdout: "ignore",
 				stderr: "ignore",
@@ -79,12 +97,30 @@ export class PiCodingWorker implements CodingWorker {
 							packages: this.#options.packages ?? [],
 							agentDir: this.#options.agentDir ?? getAgentDir(),
 							prompt,
+							progress: progress !== undefined,
+							...(this.#options.scratchDir
+								? { scratchDir: this.#options.scratchDir }
+								: {}),
 						});
 					} else if (
 						value.type === "report" &&
 						typeof value.report === "string"
 					) {
 						report = value.report;
+					} else if (
+						value.type === "message" &&
+						typeof value.role === "string" &&
+						typeof value.text === "string"
+					) {
+						progress?.messageEnd({
+							role: value.role,
+							content: value.text,
+							...(typeof value.stopReason === "string"
+								? { stopReason: value.stopReason }
+								: {}),
+						});
+					} else if (value.type === "tool" && typeof value.name === "string") {
+						progress?.toolStart(value.name);
 					} else if (
 						value.type === "failure" &&
 						typeof value.message === "string"
@@ -109,6 +145,9 @@ export class PiCodingWorker implements CodingWorker {
 							try {
 								const context = {
 									workspace: this.#options.workspace ?? job.dir,
+									...(this.#options.scratchDir
+										? { scratchDir: this.#options.scratchDir }
+										: {}),
 								};
 								const action =
 									shellHoldRule.describe(tool, input, context) ??

package/src/worker-entry.ts

@@ -11,6 +11,7 @@ import {
 	activeToolsExtension,
 	runWorkerTask,
 	SHELL_TOOLS,
+	textOf,
 } from "pi-roundtable/kit";
 import type { CodingJob, HeldCallAnswer } from "./coding-desk.ts";
 
@@ -20,6 +21,14 @@ interface StartMessage {
 	packages: string[];
 	agentDir: string;
 	prompt?: string;
+	/** The host posts the worker's progress, so the session's messages and tools go to it. */
+	progress?: boolean;
+	/** The host's scratch dir, the worker's TMPDIR, where writes and removals need no approval. */
+	scratchDir?: string;
+}
+/** Said to every worker whose host has a scratch dir, after the host's own prompt or the default. */
+export function scratchDirPrompt(scratchDir: string): string {
+	return `Temporary files, test homes and throwaway clones go under the scratch dir ${scratchDir} ($TMPDIR): writes and removals there, and inside the repository, run without approval; elsewhere they wait for the owner.`;
 }
 export function codingWorkerPrompt(dir: string): string {
 	return [
@@ -42,6 +51,8 @@ async function run({
 	agentDir,
 	packages,
 	prompt,
+	progress,
+	scratchDir,
 }: StartMessage): Promise<string> {
 	const modelRuntime = await ModelRuntime.create({
 		authPath: join(agentDir, "auth.json"),
@@ -103,7 +114,10 @@ async function run({
 				},
 			},
 		],
-		appendSystemPrompt: [prompt ?? codingWorkerPrompt(job.dir)],
+		appendSystemPrompt: [
+			prompt ?? codingWorkerPrompt(job.dir),
+			...(scratchDir ? [scratchDirPrompt(scratchDir)] : []),
+		],
 	});
 	await loader.reload();
 	const { session } = await createAgentSession({
@@ -115,6 +129,19 @@ async function run({
 		sessionManager: SessionManager.inMemory(job.dir),
 		settingsManager: SettingsManager.inMemory({}),
 	});
+	// Only what the host's progress posts read crosses: a message's text, never its thinking.
+	if (progress)
+		session.subscribe((event) => {
+			if (event.type === "tool_execution_start")
+				process.send?.({ type: "tool", name: event.toolName });
+			if (event.type === "message_end" && event.message.role === "assistant")
+				process.send?.({
+					type: "message",
+					role: event.message.role,
+					stopReason: event.message.stopReason,
+					text: textOf(event.message.content),
+				});
+		});
 	return runWorkerTask(session, {
 		modelRuntime,
 		model: job.model,