@clubnet/seedclub 0.2.36 → 0.2.38

package/README.md

@@ -6,7 +6,7 @@ Requirements: Node.js 22+
 
 ## What it is
 
-`seedclub` is the Seed Club distribution of pi. This repo provides the `seedclub` launcher, installs the Seed Club theme and extensions, and adds Seed Club-specific workflows for connecting your account, using CRM/meeting/media tools, retrieving transcripts, and working with Seed Club-backed app surfaces.
+`seedclub` is the Seed Club distribution of pi. This repo provides the `seedclub` launcher, installs the Seed Club theme and extensions, and adds Seed Club-specific workflows for connecting your account, using CRM and meeting tools, retrieving transcripts, and working with Seed Club-backed app surfaces.
 
 This repo is the source of truth for the published package:
 
@@ -18,15 +18,16 @@ This repo is the source of truth for the published package:
 
 ```bash
 npm install -g @clubnet/seedclub
+seedclub
 ```
 
-Then:
+Alternative:
 
 ```bash
-seedclub
+curl -fsSL https://raw.githubusercontent.com/seedclub/seedclub-agent/main/install.sh | bash
 ```
 
-### First run
+## First run
 
 1. Run `seedclub`
 2. Complete Seed Club sign-in when the browser opens
@@ -34,20 +35,10 @@ seedclub
 4. Run `/model`
 5. Open `/seedclub`
 
-### Alternative: curl | bash
-
-```bash
-curl -fsSL https://raw.githubusercontent.com/seedclub/seedclub-agent/main/install.sh | bash
-```
-
-### Package access
-
-`@clubnet/seedclub` is a public npm package. Install access is open; runtime access is enforced inside the app.
-
 ## Core workflow
 
 1. Start the app with `seedclub`
-2. Complete Seed Club sign-in when the browser opens
+2. Complete Seed Club sign-in when prompted
 3. Complete `/login` and `/model` if this is your first run
 4. Open `/seedclub`
 5. Choose the workflow you need
@@ -55,15 +46,23 @@ curl -fsSL https://raw.githubusercontent.com/seedclub/seedclub-agent/main/instal
 ## Commands
 
 | Command | What it does |
-|---|---|
+| --- | --- |
 | `/login` | Sign in to a model provider for the underlying agent |
 | `/model` | Choose which model to use |
 | `/connect` | Connect your Seed Club account |
 | `/connect-calendar` | Connect a personal Google Calendar to your Seed Club account |
-| `/seedclub` | Main menu — connect, inspect access, and jump into CRM/meetings/media/headlines workflows |
-| `/transcripts` | Export transcript VTT files with filters (date, person, time, output dir) |
+| `/seedclub` | Main menu for Seed Club workflows |
+| `/transcripts` | Export transcript VTT files with filters |
+
+Natural-language transcript retrieval is also supported. Examples: `download vibhu transcripts from 11am`, `i need transcripts for all guests on 11am last week`.
+
+## Repo docs
 
-Natural-language transcript retrieval is also supported (no slash command required). Examples: `download vibhu transcripts from 11am`, `i need transcripts for all guests on 11am last week`. Seed Club will run metadata-first export confirmation and then write VTT files.
+- `README.md` is the public package and repo landing page.
+- `CONTRIBUTING.md` covers local development, architecture boundaries, and release workflow.
+- `AGENTS.md` contains repo instructions for coding agents working in this codebase.
+- `WORKFLOWS.md` contains Seed Club API route-selection rules.
+- `packages/seedclub-runtime/README.md` and `packages/seedclub-tui/README.md` document package-local ownership boundaries.
 
 ## Update
 
@@ -78,16 +77,6 @@ npm uninstall -g @clubnet/seedclub
 rm -rf ~/.seedclub
 ```
 
-### Coming from the old version (curl | bash)
-
-The previous version installed into `~/.seedclub/bin/` and modified your PATH. The npm package cleans this up automatically, but you can also remove the old PATH line from your shell profile (`~/.zshrc`, `~/.bashrc`, etc.) manually.
-
-```bash
-npm install -g @clubnet/seedclub
-```
-
-If you have pi installed globally (`npm install -g @mariozechner/pi-coding-agent`), that's fine — seedclub and pi are completely independent.
-
 ## License
 
 MIT

package/assets/SYSTEM.md

@@ -0,0 +1,45 @@
+You are a general-purpose computer and coding agent operating inside the Seed Club terminal application.
+
+Your job is to help users get real work done across code, files, research, and Seed Club platform workflows. Act directly, stay concise, and use the best available interface for the task.
+
+Identity:
+
+- In user-facing replies, answer directly and naturally. Mention Seed Club only when the user is asking about the product, the platform, auth, installation, or branded workflows.
+- Use exact upstream names such as `pi-coding-agent` and `pi-tui` only when technical precision matters, for example when discussing package names, file paths, vendored code, or repository history.
+
+Operating model:
+
+- Be a useful general coding and computer agent first, not just a platform assistant.
+- Inspect the real environment before making assumptions.
+- Prefer doing the work directly over describing what you would do.
+- Show file paths clearly when discussing code or local files.
+- Stay within the user's actual access and permissions.
+- If the user is asking about upstream pi internals or SDK behavior, inspect the installed upstream docs and code before answering.
+
+Interface selection:
+
+- Use local machine tools for code changes, file retrieval, file transforms, shell tasks, exports, and other computer-local workflows.
+- Use Seed Club platform tools when the task is about Seed Club records, workflows, people, meetings, transcripts, media, or network operations.
+- Use external web research only when the task is outside Seed Club's own records or when current external verification is needed.
+- Choose the narrowest tool or workflow that can answer the question or complete the task reliably.
+
+Seed Club platform policy:
+
+- Prefer Seed Club tools first when the user is asking about Seed Club data or workflows.
+- For Seed Club records questions, treat Seed Club platform tools as the primary source of truth, not the local repo or filesystem.
+- Do not inspect local repo code, extension source, or API client helpers to rediscover Seed Club routes when a registered Seed Club tool already matches the task.
+- Resolve program names, slugs, and shorthand against the user's accessible Seed Club programs when that context is available.
+- Prefer metadata-first exploration before loading large transcript or media payloads.
+- For prior-conversation questions such as "what did we talk about?" or "what were the main topics?", prefer `full_conversation` media assets when available, then use meeting transcripts only as a fallback or when the canonical meeting transcript is specifically needed.
+- Treat local export, download, upload, and publish actions as workflow endpoints: use the purpose-built tools for those actions instead of recreating them indirectly.
+- When the user wants files on disk or assets uploaded, keep the flow anchored in the local machine plus the relevant Seed Club platform tool.
+- Stay scoped to the user's platform permissions and do not imply access the user does not have.
+- Do not claim Seed Club platform data is unavailable unless a relevant Seed Club tool actually returns an auth, permission, or not-found failure.
+
+External research policy:
+
+- If the user asks an open-ended research question that is not explicitly scoped to Seed Club records, do a fast external research pass before answering.
+- Use `search_web` first for external search and `fetch_web_page` for follow-up reads.
+- Prefer built-in web tools over bash, curl, Python, or ad hoc scraping for external research.
+- Gather at least 2 reputable sources, then answer with a concise synthesis.
+- Include source URLs inline so the user can verify.

package/assets/extensions/seedclub/commands/seedclub.ts

@@ -18,6 +18,7 @@ interface SeedclubDeps {
 	connect: (args: string | undefined, ctx: any) => Promise<boolean>;
 	connectCalendar: (ctx: any) => Promise<void>;
 	disconnect: (ctx: any) => Promise<void>;
+	showMemoryMenu?: (ctx: any) => Promise<void>;
 }
 
 async function compactSeedclubContext(ctx: any) {
@@ -136,6 +137,7 @@ export function registerSeedclubCommand(pi: ExtensionAPI, deps: SeedclubDeps) {
 				"Use local API/Auth",
 				"Use prod API/Auth",
 				"Connect personal calendar",
+				"Memory",
 				"Open CRM prompt",
 				"Open meetings prompt",
 				"Open transcripts prompt",
@@ -178,6 +180,10 @@ export function registerSeedclubCommand(pi: ExtensionAPI, deps: SeedclubDeps) {
 				case "Connect personal calendar":
 					await deps.connectCalendar(ctx);
 					break;
+				case "Memory":
+					if (deps.showMemoryMenu) await deps.showMemoryMenu(ctx);
+					else ctx.ui.notify("Memory controls are unavailable.", "warning");
+					break;
 				case "Open CRM prompt":
 					await new Promise((r) => setTimeout(r, 300));
 					ctx.ui.setEditorText("List CRM records for my workspace.");

package/assets/extensions/seedclub/commands/transcript-export.ts

@@ -4,6 +4,7 @@ import { join, resolve } from "node:path";
 import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
 import { api } from "../api-client.js";
 import { getSessionContext } from "../tools/utility.js";
+import { setWorkingMessage } from "../tool-utils.js";
 import {
 	exactNormalizedTextMatch,
 	isLikelyVtt,
@@ -148,7 +149,15 @@ function extractTranscriptText(obj: any): string | null {
 	return null;
 }
 
-function resolveVttContent(source: any): { vtt: string; mode: "native" | "generated_from_text" } | null {
+function allowSyntheticVttExport() {
+	return /^(1|true|yes)$/i.test(process.env.SEEDCLUB_ALLOW_SYNTHETIC_VTT_EXPORT ?? "");
+}
+
+function resolveVttContent(
+	source: any,
+	options?: { allowGeneratedFromText?: boolean },
+): { vtt: string; mode: "native" | "generated_from_text" } | null {
+	const allowGeneratedFromText = options?.allowGeneratedFromText === true;
 	const explicitVtt = typeof source?.transcript_vtt === "string" ? source.transcript_vtt.trim() : "";
 	if (explicitVtt) return { vtt: explicitVtt, mode: "native" };
 
@@ -158,15 +167,17 @@ function resolveVttContent(source: any): { vtt: string; mode: "native" | "genera
 	const maybeVtt = extractTranscriptText(source);
 	if (typeof maybeVtt === "string" && maybeVtt.trim()) {
 		if (isLikelyVtt(maybeVtt)) return { vtt: maybeVtt.trim(), mode: "native" };
-		const generated = textToSyntheticVtt(maybeVtt);
-		if (generated) return { vtt: generated, mode: "generated_from_text" };
+		if (allowGeneratedFromText) {
+			const generated = textToSyntheticVtt(maybeVtt);
+			if (generated) return { vtt: generated, mode: "generated_from_text" };
+		}
 	}
 
 	return null;
 }
 
-function ensureVtt(asset: any): string | null {
-	return resolveVttContent(asset)?.vtt ?? null;
+function ensureVtt(asset: any, options?: { allowGeneratedFromText?: boolean }): string | null {
+	return resolveVttContent(asset, options)?.vtt ?? null;
 }
 
 async function listCandidateAssets(
@@ -238,8 +249,12 @@ function personFromMeetingTitle(title?: string | null): string | null {
 	return cleaned || null;
 }
 
-function toCandidateFromAsset(asset: any, fallbackPerson: string): TranscriptCandidate | null {
-	const resolvedVtt = resolveVttContent(asset);
+function toCandidateFromAsset(
+	asset: any,
+	fallbackPerson: string,
+	options?: { allowGeneratedFromText?: boolean },
+): TranscriptCandidate | null {
+	const resolvedVtt = resolveVttContent(asset, options);
 	if (!resolvedVtt) return null;
 	const eventDate = pickDate(asset);
 	return {
@@ -257,7 +272,11 @@ function toCandidateFromAsset(asset: any, fallbackPerson: string): TranscriptCan
 	};
 }
 
-async function listFallbackTranscriptCandidates(programSlug: string, intent: TranscriptIntent): Promise<TranscriptCandidate[]> {
+async function listFallbackTranscriptCandidates(
+	programSlug: string,
+	intent: TranscriptIntent,
+	options?: { allowGeneratedFromText?: boolean },
+): Promise<TranscriptCandidate[]> {
 	const response = await api.get<any>("/meetings/transcripts", {
 		program_slug: programSlug,
 		limit: 20,
@@ -300,7 +319,7 @@ async function listFallbackTranscriptCandidates(programSlug: string, intent: Tra
 
 	const candidates: TranscriptCandidate[] = [];
 	for (const row of rows) {
-		const resolvedVtt = resolveVttContent(row?.transcript ?? {});
+		const resolvedVtt = resolveVttContent(row?.transcript ?? {}, options);
 		if (!resolvedVtt) continue;
 		const eventDate = typeof row?.transcript?.event_date === "string" ? row.transcript.event_date : new Date().toISOString().slice(0, 10);
 		const person = pickPersonFromTranscriptRow(row, intent.person?.trim() || "unassigned");
@@ -467,190 +486,194 @@ async function offerOpenDownloadedPath(agent: ExtensionAPI, ctx: any, outDir: st
 }
 
 export async function exportTranscripts(agent: ExtensionAPI, ctx: any, args: TranscriptExportArgs) {
-	const request = args.request?.trim();
-	if (!request) {
-		return {
-			error: "Provide the original transcript request.",
-			status: 400,
+	try {
+		const request = args.request?.trim();
+		if (!request) {
+			return {
+				error: "Provide the original transcript request.",
+				status: 400,
+			};
+		}
+
+		const intent = {
+			...parseIntent(request),
+			...(args.person ? { person: args.person } : {}),
+			...(args.date ? { date: args.date } : {}),
+			...(args.time ? { time: args.time } : {}),
+			...(args.outDir ? { outDir: args.outDir } : {}),
+			...(args.latestCount ? { latestCount: Math.max(1, Math.min(20, Math.trunc(args.latestCount))) } : {}),
 		};
-	}
+		const allowGeneratedFromText = allowSyntheticVttExport();
 
-	const intent = {
-		...parseIntent(request),
-		...(args.person ? { person: args.person } : {}),
-		...(args.date ? { date: args.date } : {}),
-		...(args.time ? { time: args.time } : {}),
-		...(args.outDir ? { outDir: args.outDir } : {}),
-		...(args.latestCount ? { latestCount: Math.max(1, Math.min(20, Math.trunc(args.latestCount))) } : {}),
-	};
+		setWorkingMessage(ctx, "Loading Seed Club session...");
+		const session = await getSessionContext();
+		if ("error" in session) {
+			ctx.ui.notify(`Unable to load Seed Club session: ${session.error}`, "error");
+			return {
+				error: session.error,
+				status: session.status ?? 500,
+			};
+		}
 
-	const session = await getSessionContext();
-	if ("error" in session) {
-		ctx.ui.notify(`Unable to load Seed Club session: ${session.error}`, "error");
-		return {
-			error: session.error,
-			status: session.status ?? 500,
-		};
-	}
+		const programs = pickPrograms(session);
+		setWorkingMessage(ctx, "Resolving transcript program...");
+		const program = args.programSlug
+			? programs.find((candidate) => candidate.slug === args.programSlug) ?? null
+			: await resolveProgram(request, programs, ctx);
+		if (!program) {
+			ctx.ui.notify("No program selected. Include a program name like 11AM or use /transcripts.", "warning");
+			return {
+				error: args.programSlug
+					? `Program ${args.programSlug} is not accessible for the current user.`
+					: "No program selected.",
+				status: 404,
+			};
+		}
 
-	const programs = pickPrograms(session);
-	const program = args.programSlug
-		? programs.find((candidate) => candidate.slug === args.programSlug) ?? null
-		: await resolveProgram(request, programs, ctx);
-	if (!program) {
-		ctx.ui.notify("No program selected. Include a program name like 11AM or use /transcripts.", "warning");
-		return {
-			error: args.programSlug
-				? `Program ${args.programSlug} is not accessible for the current user.`
-				: "No program selected.",
-			status: 404,
+		const filterAssetsForIntent = async (inputAssets: any[]) => {
+			let filtered = await filterByMeetingTime(inputAssets, intent.time);
+			filtered = filtered.filter((asset) => !!ensureVtt(asset, { allowGeneratedFromText }));
+			if (!args.partyId) {
+				const peopleQueries = Array.from(new Set([...(intent.people ?? []), ...(intent.person ? [intent.person] : [])].filter(Boolean)));
+				if (peopleQueries.length) {
+					filtered = filtered.filter((asset) => {
+						const haystack = [asset?.title, asset?.key_question, asset?.file_name]
+							.filter((v) => typeof v === "string" && v.trim())
+							.join(" ");
+						return peopleQueries.some((query) => exactNormalizedTextMatch(query, haystack));
+					});
+				}
+			}
+			return filtered;
 		};
-	}
 
-	const filterAssetsForIntent = async (inputAssets: any[]) => {
-		let filtered = await filterByMeetingTime(inputAssets, intent.time);
-		filtered = filtered.filter((asset) => !!ensureVtt(asset));
-		if (!args.partyId) {
-			const peopleQueries = Array.from(new Set([...(intent.people ?? []), ...(intent.person ? [intent.person] : [])].filter(Boolean)));
-			if (peopleQueries.length) {
-				filtered = filtered.filter((asset) => {
-					const haystack = [asset?.title, asset?.key_question, asset?.file_name]
-						.filter((v) => typeof v === "string" && v.trim())
-						.join(" ");
-					return peopleQueries.some((query) => exactNormalizedTextMatch(query, haystack));
-				});
-			}
+		setWorkingMessage(ctx, `Checking full-conversation VTT files in ${program.slug}...`);
+		let assets = await filterAssetsForIntent(await listCandidateAssets(program.slug, args.partyId, intent.date, "full_conversation"));
+		if (!assets.length) {
+			setWorkingMessage(ctx, `Checking clip transcript text in ${program.slug}...`);
+			assets = await filterAssetsForIntent(await listCandidateAssets(program.slug, args.partyId, intent.date, "clip"));
 		}
-		return filtered;
-	};
 
-	ctx.ui.notify(`Looking up full-conversation VTT files in ${program.slug}...`, "info");
-	let assets = await filterAssetsForIntent(await listCandidateAssets(program.slug, args.partyId, intent.date, "full_conversation"));
-	if (!assets.length) {
-		ctx.ui.notify(`No full-conversation VTT found. Checking clip transcript text to generate VTT in ${program.slug}...`, "info");
-		assets = await filterAssetsForIntent(await listCandidateAssets(program.slug, args.partyId, intent.date, "clip"));
-	}
+		let candidates = assets
+			.map((asset) => toCandidateFromAsset(asset, intent.person || "unassigned", { allowGeneratedFromText }))
+			.filter((item): item is TranscriptCandidate => !!item);
 
-	let candidates = assets
-		.map((asset) => toCandidateFromAsset(asset, intent.person || "unassigned"))
-		.filter((item): item is TranscriptCandidate => !!item);
+		candidates = applyLatestFilter(candidates, intent.latestCount, (row) => row.sortTimestamp);
 
-	candidates = applyLatestFilter(candidates, intent.latestCount, (row) => row.sortTimestamp);
+		if (!candidates.length) {
+			setWorkingMessage(ctx, "Checking meeting transcript rows...");
+			candidates = await listFallbackTranscriptCandidates(program.slug, intent, { allowGeneratedFromText });
+		}
 
-	if (!candidates.length) {
-		ctx.ui.notify("No matching asset transcript content found. Checking meeting transcript rows...", "info");
-		candidates = await listFallbackTranscriptCandidates(program.slug, intent);
-	}
+		if (!candidates.length) {
+			return {
+				program: program.slug,
+				request,
+				count: 0,
+				files: [],
+				message: "No matching transcript VTT files found.",
+			};
+		}
 
-	if (!candidates.length) {
-		ctx.ui.notify("No matching transcript VTT files found.", "info");
-		return {
-			program: program.slug,
-			request,
-			count: 0,
-			files: [],
-			message: "No matching transcript VTT files found.",
-		};
-	}
+		const unresolvedSource = candidates.filter((candidate) => !hasTranscriptSourceRef(candidate));
+		const unresolvedPerson = candidates.filter((candidate) => !hasNamedPerson(candidate.person));
+
+		if (unresolvedSource.length || unresolvedPerson.length) {
+			const unresolvedRows = candidates
+				.filter((candidate) => !hasTranscriptSourceRef(candidate) || !hasNamedPerson(candidate.person))
+				.slice(0, 5)
+				.map((candidate) => {
+					const sourceRef = hasTranscriptSourceRef(candidate) ? "ok" : "missing-source";
+					const personRef = hasNamedPerson(candidate.person) ? "ok" : "missing-person";
+					return `- ${candidate.eventDate} :: ${candidate.person || "unknown"} [${sourceRef}; ${personRef}]`;
+				})
+				.join("\n");
+
+			const message = `Strict transcript export guard blocked this download. ${
+				unresolvedSource.length
+			} row(s) are missing source reference (assetId/meetingId), and ${unresolvedPerson.length} row(s) are missing a named person.`;
+			ctx.ui.notify(`${message}${unresolvedRows ? `\n${unresolvedRows}` : ""}`, "warning");
+			return {
+				program: program.slug,
+				request,
+				count: 0,
+				files: [],
+				message,
+			};
+		}
 
-	const unresolvedSource = candidates.filter((candidate) => !hasTranscriptSourceRef(candidate));
-	const unresolvedPerson = candidates.filter((candidate) => !hasNamedPerson(candidate.person));
-
-	if (unresolvedSource.length || unresolvedPerson.length) {
-		const unresolvedRows = candidates
-			.filter((candidate) => !hasTranscriptSourceRef(candidate) || !hasNamedPerson(candidate.person))
-			.slice(0, 5)
-			.map((candidate) => {
-				const sourceRef = hasTranscriptSourceRef(candidate) ? "ok" : "missing-source";
-				const personRef = hasNamedPerson(candidate.person) ? "ok" : "missing-person";
-				return `- ${candidate.eventDate} :: ${candidate.person || "unknown"} [${sourceRef}; ${personRef}]`;
-			})
-			.join("\n");
-
-		const message = `Strict transcript export guard blocked this download. ${
-			unresolvedSource.length
-		} row(s) are missing source reference (assetId/meetingId), and ${unresolvedPerson.length} row(s) are missing a named person.`;
-		ctx.ui.notify(`${message}${unresolvedRows ? `\n${unresolvedRows}` : ""}`, "warning");
-		return {
-			program: program.slug,
-			request,
-			count: 0,
-			files: [],
-			message,
-		};
-	}
+		const baseOutDir = resolveOutputDir(program.slug, intent.outDir);
+		const planned = planCandidateFileNames(candidates);
+		const outDir = intent.outDir ? baseOutDir : join(baseOutDir, buildExportBatchFolderName(planned));
+		if (!intent.outDir && ctx.hasUI !== false) {
+			const choice = await ctx.ui.select(
+				buildTranscriptCandidateSummary(planned, outDir),
+				["Download", "Cancel"],
+			);
+			if (choice !== "Download") {
+				return {
+					program: program.slug,
+					request,
+					cancelled: true,
+					count: planned.length,
+					files: [],
+					message: "Transcript download cancelled.",
+				};
+			}
+		}
+
+		setWorkingMessage(ctx, "Writing transcript files...");
+		await mkdir(outDir, { recursive: true });
+
+		const manifest: Array<Record<string, unknown>> = [];
+
+		for (const { candidate, fileName } of planned) {
+			await writeFile(join(outDir, fileName), candidate.vtt, "utf8");
+			manifest.push({
+				file: fileName,
+				source: candidate.source,
+				assetId: candidate.assetId ?? null,
+				meetingId: candidate.meetingId ?? null,
+				eventDate: candidate.eventDate,
+				person: candidate.person,
+				vttMode: candidate.vttMode ?? "native",
+				videoUrl: candidate.videoUrl ?? null,
+				audioUrl: candidate.audioUrl ?? null,
+			});
+		}
 
-	const baseOutDir = resolveOutputDir(program.slug, intent.outDir);
-	const planned = planCandidateFileNames(candidates);
-	const outDir = intent.outDir ? baseOutDir : join(baseOutDir, buildExportBatchFolderName(planned));
-	if (!intent.outDir && ctx.hasUI !== false) {
-		const choice = await ctx.ui.select(
-			buildTranscriptCandidateSummary(planned, outDir),
-			["Download", "Cancel"],
-		);
-		if (choice !== "Download") {
+		if (!manifest.length) {
 			return {
 				program: program.slug,
 				request,
-				cancelled: true,
-				count: planned.length,
+				count: 0,
 				files: [],
-				message: "Transcript download cancelled.",
+				message: "Matched assets, but no usable VTT content was found.",
 			};
 		}
-	}
 
-	await mkdir(outDir, { recursive: true });
-
-	const manifest: Array<Record<string, unknown>> = [];
-
-	for (const { candidate, fileName } of planned) {
-		await writeFile(join(outDir, fileName), candidate.vtt, "utf8");
-		manifest.push({
-			file: fileName,
-			source: candidate.source,
-			assetId: candidate.assetId ?? null,
-			meetingId: candidate.meetingId ?? null,
-			eventDate: candidate.eventDate,
-			person: candidate.person,
-			vttMode: candidate.vttMode ?? "native",
-			videoUrl: candidate.videoUrl ?? null,
-			audioUrl: candidate.audioUrl ?? null,
-		});
-	}
+		const generatedCount = manifest.filter((row: any) => row?.vttMode === "generated_from_text").length;
+		const summary = buildTranscriptDownloadSummary(manifest, outDir);
+		const message =
+			generatedCount > 0
+				? `${summary}\n\nGenerated ${generatedCount} VTT file${generatedCount === 1 ? "" : "s"} from transcript text.`
+				: summary;
+
+		if (ctx.hasUI !== false) await offerOpenDownloadedPath(agent, ctx, outDir);
+		const downloadableVideoCount = candidates.filter((item) => typeof item.videoUrl === "string" && item.videoUrl.trim()).length;
 
-	if (!manifest.length) {
-		ctx.ui.notify("Matched assets, but no usable VTT content was found.", "warning");
 		return {
 			program: program.slug,
 			request,
-			count: 0,
-			files: [],
-			message: "Matched assets, but no usable VTT content was found.",
+			outDir,
+			manifestPath: null,
+			videoDir: null,
+			downloadableVideoCount,
+			count: manifest.length,
+			files: manifest,
+			message,
 		};
+	} finally {
+		setWorkingMessage(ctx, undefined);
 	}
-
-	const generatedCount = manifest.filter((row: any) => row?.vttMode === "generated_from_text").length;
-	const summary = buildTranscriptDownloadSummary(manifest, outDir);
-	ctx.ui.notify(
-		generatedCount > 0
-			? `${summary}\n\nGenerated ${generatedCount} VTT file${generatedCount === 1 ? "" : "s"} from transcript text.`
-			: summary,
-		"info",
-	);
-
-	if (ctx.hasUI !== false) await offerOpenDownloadedPath(agent, ctx, outDir);
-	const downloadableVideoCount = candidates.filter((item) => typeof item.videoUrl === "string" && item.videoUrl.trim()).length;
-
-	return {
-		program: program.slug,
-		request,
-		outDir,
-		manifestPath: null,
-		videoDir: null,
-		downloadableVideoCount,
-		count: manifest.length,
-		files: manifest,
-		message: summary,
-	};
 }

package/assets/extensions/seedclub/commands/transcript-intent.ts

@@ -102,7 +102,7 @@ export function registerTranscriptIntentInterceptor(pi: ExtensionAPI) {
 		name: "seedclub_export_transcripts",
 		label: "Export Transcripts",
 		description:
-			"Download matching Seed Club transcript VTT files locally using exact constraints already established in chat. Use this after the user asks to get, pull, export, download, save, or review transcript/VTT/caption files locally, including short prompts like \"today's transcripts\" or \"2pm VTTs\". Do not use this tool for fuzzy person disambiguation or contact search; ask the user to confirm, or use metadata/contact tools first, then pass exact date/time/person/partyId here. Do not use for transcript inventory questions such as \"what was the last day we have transcripts for 11am?\"; answer those with seedclub_list_meeting_transcripts metadata instead. Always pass the user's original request string in request.",
+			"Download matching Seed Club transcript VTT files locally using exact constraints already established in chat. Use this after the user asks to get, pull, export, download, save, or review transcript/VTT/caption files locally, including short prompts like \"today's transcripts\" or \"2pm VTTs\". Keep the user's original request string in request. Do not use this tool for fuzzy person disambiguation or contact search; ask the user to confirm, or use metadata/contact tools first, then pass exact date/time/person/partyId here. Do not use for transcript inventory questions such as \"what was the last day we have transcripts for 11am?\"; answer those with seedclub_list_meeting_transcripts metadata instead. This tool is for local file export results, not for pasting transcript text or raw JSON into chat.",
 		parameters: Type.Object({
 			request: Type.String({ description: "The user's original transcript request, verbatim when possible." }),
 			programSlug: Type.Optional(Type.String({ description: "Optional program slug, e.g. 11am." })),
@@ -217,12 +217,15 @@ export function registerTranscriptIntentInterceptor(pi: ExtensionAPI) {
 IMPORTANT SEED CLUB TRANSCRIPT EXPORT ROUTING:
 - The user's prompt appears to be a transcript file export/retrieval request.
 - Keep the user's prompt in normal conversation context.
+- Short artifact prompts like "today's transcripts", "Tuesday VTTs", or "2pm captions" are export asks, not inline transcript display asks.
 - If person/date/program constraints are not clear, ask or use metadata tools first; do not let seedclub_export_transcripts perform fuzzy disambiguation.
 - Prefer calling seedclub_export_transcripts only after the constraints are exact, with request set to the user's original prompt.
+- If the user explicitly wants transcript files, keep the flow in transcript export once constraints are exact instead of chaining broad media or transcript-reading tools first.
 - Avoid chaining high-payload media/transcript tools for export asks when seedclub_export_transcripts can satisfy the request.
+- If the user asks what transcripts exist, use transcript metadata tools instead of export.
 - For transcript export requests, download transcripts only in the tool. If videos are available, ask as a follow-up in chat instead of prompting inside the tool.
 - Once a transcript export succeeds for the identified person in this turn, stop calling transcript export again.
-- Do not answer by dumping transcript text or raw JSON into chat.`,
+- Report export results, paths, and follow-up options in chat instead of dumping transcript text or raw JSON.`,
 		};
 	});
 }