@tintinweb/pi-subagents 0.3.1 → 0.4.1

package/CHANGELOG.md

@@ -5,6 +5,31 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.4.1] - 2026-03-11
+
+### Fixed
+- **Graceful shutdown in headless mode** — the CLI now waits for all running and queued background agents to complete before exiting (`waitForAll` on `session_shutdown`). Previously, background agents could be silently killed mid-execution when the session ended. Only affects headless/non-interactive mode; interactive sessions already kept the process alive.
+
+### Added
+- `hasRunning()` / `waitForAll()` methods on `AgentManager`.
+- **Cross-package manager access** — agent manager exposed via `Symbol.for("pi-subagents:manager")` on `globalThis` for other extensions to check status or await completion.
+
+## [0.4.0] - 2026-03-11
+
+### Added
+- **XML-delimited prompt sections** — append-mode agents now wrap inherited content in `<inherited_system_prompt>`, `<sub_agent_context>`, and `<agent_instructions>` XML tags, giving the model explicit structure to distinguish inherited rules from sub-agent-specific instructions. Replace mode is unchanged.
+- **Token count in agent results** — foreground agent results, background completion notifications, and `get_subagent_result` now include the token count alongside tool uses and duration (e.g. `Agent completed in 4.2s (12 tool uses, 33.8k token)`).
+- **Widget overflow cap** — the running agents widget now caps at 12 lines. When exceeded, running agents are prioritized over finished ones and an overflow summary line shows hidden counts (e.g. `+3 more (1 running, 2 finished)`).
+
+### Changed - **changing behavior**
+- **General-purpose agent inherits parent prompt** — the default `general-purpose` agent now uses `promptMode: "append"` with an empty system prompt, making it a "parent twin" that inherits the full parent system prompt (including CLAUDE.md rules, project conventions, and safety guardrails). Previously it used a standalone prompt that duplicated a subset of the parent's rules. Explore and Plan are unchanged (standalone prompts). To customize: eject via `/agents` → select `general-purpose` → Eject, then edit the resulting `.md` file. Set `prompt_mode: replace` to go back to a standalone prompt, or keep `prompt_mode: append` and add extra instructions in the body.
+- **Append-mode agents receive parent system prompt** — `buildAgentPrompt` now accepts the parent's system prompt and threads it into append-mode agents (env header + parent prompt + sub-agent context bridge + optional custom instructions). Replace-mode agents are unchanged.
+- **Prompt pipeline simplified** — removed `systemPromptOverride`/`systemPromptAppend` from `SpawnOptions` and `RunOptions`. These were a separate code path where `index.ts` pre-resolved the prompt mode and passed raw strings into the runner, bypassing `buildAgentPrompt`. Now all prompt assembly flows through `buildAgentPrompt` using the agent's `promptMode` config — one code path, no special cases.
+
+### Removed
+- Deprecated backwards-compat aliases: `registerCustomAgents`, `getCustomAgentConfig`, `getCustomAgentNames` (use `registerAgents`, `getAgentConfig`, `getUserAgentNames`).
+- `resolveCustomPrompt()` helper in index.ts — no longer needed now that prompt routing is config-driven.
+
 ## [0.3.1] - 2026-03-09
 
 ### Added
@@ -139,7 +164,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Added
 - **Claude Code-style UI rendering** — `renderCall`/`renderResult`/`onUpdate` for live streaming progress
   - Live activity descriptions: "searching, reading 3 files…"
-  - Token count display: "33.8k tokens"
+  - Token count display: "33.8k token"
   - Per-agent tool use counter
   - Expandable completed results (ctrl+o)
   - Distinct states: running, background, completed, error, aborted
@@ -172,6 +197,9 @@ Initial release.
 - **Thinking level** — per-agent extended thinking control
 - **`/agent` and `/agents` commands**
 
+[0.4.1]: https://github.com/tintinweb/pi-subagents/compare/v0.4.0...v0.4.1
+[0.4.0]: https://github.com/tintinweb/pi-subagents/compare/v0.3.1...v0.4.0
+[0.3.1]: https://github.com/tintinweb/pi-subagents/compare/v0.3.0...v0.3.1
 [0.3.0]: https://github.com/tintinweb/pi-subagents/compare/v0.2.7...v0.3.0
 [0.2.7]: https://github.com/tintinweb/pi-subagents/compare/v0.2.6...v0.2.7
 [0.2.6]: https://github.com/tintinweb/pi-subagents/compare/v0.2.5...v0.2.6

package/README.md

@@ -57,9 +57,9 @@ The extension renders a persistent widget above the editor showing all active ag
 
 ```
 ● Agents
-├─ ⠹ Agent  Refactor auth module · 5 tool uses · 33.8k tokens · 12.3s
+├─ ⠹ Agent  Refactor auth module · 5 tool uses · 33.8k token · 12.3s
 │    ⎿  editing 2 files…
-├─ ⠹ Explore  Find auth files · 3 tool uses · 12.4k tokens · 4.1s
+├─ ⠹ Explore  Find auth files · 3 tool uses · 12.4k token · 4.1s
 │    ⎿  searching…
 └─ 2 queued
 ```
@@ -68,24 +68,26 @@ Individual agent results render Claude Code-style in the conversation:
 
 | State | Example |
 |-------|---------|
-| **Running** | `⠹ 3 tool uses · 12.4k tokens` / `⎿ searching, reading 3 files…` |
-| **Completed** | `✓ 5 tool uses · 33.8k tokens · 12.3s` / `⎿ Done` |
-| **Wrapped up** | `✓ 50 tool uses · 89.1k tokens · 45.2s` / `⎿ Wrapped up (turn limit)` |
-| **Stopped** | `■ 3 tool uses · 12.4k tokens` / `⎿ Stopped` |
-| **Error** | `✗ 3 tool uses · 12.4k tokens` / `⎿ Error: timeout` |
-| **Aborted** | `✗ 55 tool uses · 102.3k tokens` / `⎿ Aborted (max turns exceeded)` |
+| **Running** | `⠹ 3 tool uses · 12.4k token` / `⎿ searching, reading 3 files…` |
+| **Completed** | `✓ 5 tool uses · 33.8k token · 12.3s` / `⎿ Done` |
+| **Wrapped up** | `✓ 50 tool uses · 89.1k token · 45.2s` / `⎿ Wrapped up (turn limit)` |
+| **Stopped** | `■ 3 tool uses · 12.4k token` / `⎿ Stopped` |
+| **Error** | `✗ 3 tool uses · 12.4k token` / `⎿ Error: timeout` |
+| **Aborted** | `✗ 55 tool uses · 102.3k token` / `⎿ Aborted (max turns exceeded)` |
 
 Completed results can be expanded (ctrl+o in pi) to show the full agent output inline.
 
 ## Default Agent Types
 
-| Type | Tools | Model | Description |
-|------|-------|-------|-------------|
-| `general-purpose` | all 7 | inherit | Full read/write access for complex multi-step tasks |
-| `Explore` | read, bash, grep, find, ls | haiku (falls back to inherit) | Fast codebase exploration (read-only) |
-| `Plan` | read, bash, grep, find, ls | inherit | Software architect for implementation planning (read-only) |
+| Type | Tools | Model | Prompt Mode | Description |
+|------|-------|-------|-------------|-------------|
+| `general-purpose` | all 7 | inherit | `append` (parent twin) | Inherits the parent's full system prompt — same rules, CLAUDE.md, project conventions |
+| `Explore` | read, bash, grep, find, ls | haiku (falls back to inherit) | `replace` (standalone) | Fast codebase exploration (read-only) |
+| `Plan` | read, bash, grep, find, ls | inherit | `replace` (standalone) | Software architect for implementation planning (read-only) |
 
-Default agents can be **overridden** by creating a `.md` file with the same name (e.g. `.pi/agents/Explore.md`), or **disabled** per-project by creating a `.md` file with `enabled: false` frontmatter.
+The `general-purpose` agent is a **parent twin** — it receives the parent's entire system prompt plus a sub-agent context bridge, so it follows the same rules the parent does. Explore and Plan use standalone prompts tailored to their read-only roles.
+
+Default agents can be **ejected** (`/agents` → select agent → Eject) to export them as `.md` files for customization, **overridden** by creating a `.md` file with the same name (e.g. `.pi/agents/general-purpose.md`), or **disabled** per-project with `enabled: false` frontmatter.
 
 ## Custom Agents
 
@@ -140,7 +142,7 @@ All fields are optional — sensible defaults for everything.
 | `model` | inherit parent | Model — `provider/modelId` or fuzzy name (`"haiku"`, `"sonnet"`) |
 | `thinking` | inherit | off, minimal, low, medium, high, xhigh |
 | `max_turns` | 50 | Max agentic turns before graceful shutdown |
-| `prompt_mode` | `replace` | `replace`: body is the full system prompt. `append`: body appended to default prompt |
+| `prompt_mode` | `replace` | `replace`: body is the full system prompt. `append`: body appended to parent's prompt (agent acts as a "parent twin" with optional extra instructions) |
 | `inherit_context` | `false` | Fork parent conversation into agent |
 | `run_in_background` | `false` | Run in background by default |
 | `isolated` | `false` | No extension/MCP tools, only built-in |

package/dist/agent-manager.d.ts

@@ -0,0 +1,70 @@
+/**
+ * agent-manager.ts — Tracks agents, background execution, resume support.
+ *
+ * Background agents are subject to a configurable concurrency limit (default: 4).
+ * Excess agents are queued and auto-started as running agents complete.
+ * Foreground agents bypass the queue (they block the parent anyway).
+ */
+import type { ExtensionContext, ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { Model } from "@mariozechner/pi-ai";
+import type { AgentSession } from "@mariozechner/pi-coding-agent";
+import { type ToolActivity } from "./agent-runner.js";
+import type { SubagentType, AgentRecord, ThinkingLevel } from "./types.js";
+export type OnAgentComplete = (record: AgentRecord) => void;
+interface SpawnOptions {
+    description: string;
+    model?: Model<any>;
+    maxTurns?: number;
+    isolated?: boolean;
+    inheritContext?: boolean;
+    thinkingLevel?: ThinkingLevel;
+    isBackground?: boolean;
+    /** Called on tool start/end with activity info (for streaming progress to UI). */
+    onToolActivity?: (activity: ToolActivity) => void;
+    /** Called on streaming text deltas from the assistant response. */
+    onTextDelta?: (delta: string, fullText: string) => void;
+    /** Called when the agent session is created (for accessing session stats). */
+    onSessionCreated?: (session: AgentSession) => void;
+}
+export declare class AgentManager {
+    private agents;
+    private cleanupInterval;
+    private onComplete?;
+    private maxConcurrent;
+    /** Queue of background agents waiting to start. */
+    private queue;
+    /** Number of currently running background agents. */
+    private runningBackground;
+    constructor(onComplete?: OnAgentComplete, maxConcurrent?: number);
+    /** Update the max concurrent background agents limit. */
+    setMaxConcurrent(n: number): void;
+    getMaxConcurrent(): number;
+    /**
+     * Spawn an agent and return its ID immediately (for background use).
+     * If the concurrency limit is reached, the agent is queued.
+     */
+    spawn(pi: ExtensionAPI, ctx: ExtensionContext, type: SubagentType, prompt: string, options: SpawnOptions): string;
+    /** Actually start an agent (called immediately or from queue drain). */
+    private startAgent;
+    /** Start queued agents up to the concurrency limit. */
+    private drainQueue;
+    /**
+     * Spawn an agent and wait for completion (foreground use).
+     * Foreground agents bypass the concurrency queue.
+     */
+    spawnAndWait(pi: ExtensionAPI, ctx: ExtensionContext, type: SubagentType, prompt: string, options: Omit<SpawnOptions, "isBackground">): Promise<AgentRecord>;
+    /**
+     * Resume an existing agent session with a new prompt.
+     */
+    resume(id: string, prompt: string, signal?: AbortSignal): Promise<AgentRecord | undefined>;
+    getRecord(id: string): AgentRecord | undefined;
+    listAgents(): AgentRecord[];
+    abort(id: string): boolean;
+    private cleanup;
+    /** Whether any agents are still running or queued. */
+    hasRunning(): boolean;
+    /** Wait for all running and queued agents to complete (including queued ones). */
+    waitForAll(): Promise<void>;
+    dispose(): void;
+}
+export {};

package/dist/agent-manager.js

@@ -0,0 +1,236 @@
+/**
+ * agent-manager.ts — Tracks agents, background execution, resume support.
+ *
+ * Background agents are subject to a configurable concurrency limit (default: 4).
+ * Excess agents are queued and auto-started as running agents complete.
+ * Foreground agents bypass the queue (they block the parent anyway).
+ */
+import { randomUUID } from "node:crypto";
+import { runAgent, resumeAgent } from "./agent-runner.js";
+/** Default max concurrent background agents. */
+const DEFAULT_MAX_CONCURRENT = 4;
+export class AgentManager {
+    agents = new Map();
+    cleanupInterval;
+    onComplete;
+    maxConcurrent;
+    /** Queue of background agents waiting to start. */
+    queue = [];
+    /** Number of currently running background agents. */
+    runningBackground = 0;
+    constructor(onComplete, maxConcurrent = DEFAULT_MAX_CONCURRENT) {
+        this.onComplete = onComplete;
+        this.maxConcurrent = maxConcurrent;
+        // Cleanup completed agents after 10 minutes (but keep sessions for resume)
+        this.cleanupInterval = setInterval(() => this.cleanup(), 60_000);
+    }
+    /** Update the max concurrent background agents limit. */
+    setMaxConcurrent(n) {
+        this.maxConcurrent = Math.max(1, n);
+        // Start queued agents if the new limit allows
+        this.drainQueue();
+    }
+    getMaxConcurrent() {
+        return this.maxConcurrent;
+    }
+    /**
+     * Spawn an agent and return its ID immediately (for background use).
+     * If the concurrency limit is reached, the agent is queued.
+     */
+    spawn(pi, ctx, type, prompt, options) {
+        const id = randomUUID().slice(0, 17);
+        const abortController = new AbortController();
+        const record = {
+            id,
+            type,
+            description: options.description,
+            status: options.isBackground ? "queued" : "running",
+            toolUses: 0,
+            startedAt: Date.now(),
+            abortController,
+        };
+        this.agents.set(id, record);
+        const args = { pi, ctx, type, prompt, options };
+        if (options.isBackground && this.runningBackground >= this.maxConcurrent) {
+            // Queue it — will be started when a running agent completes
+            this.queue.push({ id, args });
+            return id;
+        }
+        this.startAgent(id, record, args);
+        return id;
+    }
+    /** Actually start an agent (called immediately or from queue drain). */
+    startAgent(id, record, { pi, ctx, type, prompt, options }) {
+        record.status = "running";
+        record.startedAt = Date.now();
+        if (options.isBackground)
+            this.runningBackground++;
+        const promise = runAgent(ctx, type, prompt, {
+            pi,
+            model: options.model,
+            maxTurns: options.maxTurns,
+            isolated: options.isolated,
+            inheritContext: options.inheritContext,
+            thinkingLevel: options.thinkingLevel,
+            signal: record.abortController.signal,
+            onToolActivity: (activity) => {
+                if (activity.type === "end")
+                    record.toolUses++;
+                options.onToolActivity?.(activity);
+            },
+            onTextDelta: options.onTextDelta,
+            onSessionCreated: (session) => {
+                record.session = session;
+                options.onSessionCreated?.(session);
+            },
+        })
+            .then(({ responseText, session, aborted, steered }) => {
+            // Don't overwrite status if externally stopped via abort()
+            if (record.status !== "stopped") {
+                record.status = aborted ? "aborted" : steered ? "steered" : "completed";
+            }
+            record.result = responseText;
+            record.session = session;
+            record.completedAt ??= Date.now();
+            if (options.isBackground) {
+                this.runningBackground--;
+                this.onComplete?.(record);
+                this.drainQueue();
+            }
+            return responseText;
+        })
+            .catch((err) => {
+            // Don't overwrite status if externally stopped via abort()
+            if (record.status !== "stopped") {
+                record.status = "error";
+            }
+            record.error = err instanceof Error ? err.message : String(err);
+            record.completedAt ??= Date.now();
+            if (options.isBackground) {
+                this.runningBackground--;
+                this.onComplete?.(record);
+                this.drainQueue();
+            }
+            return "";
+        });
+        record.promise = promise;
+    }
+    /** Start queued agents up to the concurrency limit. */
+    drainQueue() {
+        while (this.queue.length > 0 && this.runningBackground < this.maxConcurrent) {
+            const next = this.queue.shift();
+            const record = this.agents.get(next.id);
+            if (!record || record.status !== "queued")
+                continue;
+            this.startAgent(next.id, record, next.args);
+        }
+    }
+    /**
+     * Spawn an agent and wait for completion (foreground use).
+     * Foreground agents bypass the concurrency queue.
+     */
+    async spawnAndWait(pi, ctx, type, prompt, options) {
+        const id = this.spawn(pi, ctx, type, prompt, { ...options, isBackground: false });
+        const record = this.agents.get(id);
+        await record.promise;
+        return record;
+    }
+    /**
+     * Resume an existing agent session with a new prompt.
+     */
+    async resume(id, prompt, signal) {
+        const record = this.agents.get(id);
+        if (!record?.session)
+            return undefined;
+        record.status = "running";
+        record.startedAt = Date.now();
+        record.completedAt = undefined;
+        record.result = undefined;
+        record.error = undefined;
+        try {
+            const responseText = await resumeAgent(record.session, prompt, {
+                onToolActivity: (activity) => {
+                    if (activity.type === "end")
+                        record.toolUses++;
+                },
+                signal,
+            });
+            record.status = "completed";
+            record.result = responseText;
+            record.completedAt = Date.now();
+        }
+        catch (err) {
+            record.status = "error";
+            record.error = err instanceof Error ? err.message : String(err);
+            record.completedAt = Date.now();
+        }
+        return record;
+    }
+    getRecord(id) {
+        return this.agents.get(id);
+    }
+    listAgents() {
+        return [...this.agents.values()].sort((a, b) => b.startedAt - a.startedAt);
+    }
+    abort(id) {
+        const record = this.agents.get(id);
+        if (!record)
+            return false;
+        // Remove from queue if queued
+        if (record.status === "queued") {
+            this.queue = this.queue.filter(q => q.id !== id);
+            record.status = "stopped";
+            record.completedAt = Date.now();
+            return true;
+        }
+        if (record.status !== "running")
+            return false;
+        record.abortController?.abort();
+        record.status = "stopped";
+        record.completedAt = Date.now();
+        return true;
+    }
+    cleanup() {
+        const cutoff = Date.now() - 10 * 60_000;
+        for (const [id, record] of this.agents) {
+            if (record.status === "running" || record.status === "queued")
+                continue;
+            if ((record.completedAt ?? 0) >= cutoff)
+                continue;
+            // Dispose and clear session so memory can be reclaimed
+            if (record.session) {
+                record.session.dispose();
+                record.session = undefined;
+            }
+            this.agents.delete(id);
+        }
+    }
+    /** Whether any agents are still running or queued. */
+    hasRunning() {
+        return [...this.agents.values()].some(r => r.status === "running" || r.status === "queued");
+    }
+    /** Wait for all running and queued agents to complete (including queued ones). */
+    async waitForAll() {
+        // Loop because drainQueue respects the concurrency limit — as running
+        // agents finish they start queued ones, which need awaiting too.
+        while (true) {
+            this.drainQueue();
+            const pending = [...this.agents.values()]
+                .filter(r => r.status === "running" || r.status === "queued")
+                .map(r => r.promise)
+                .filter(Boolean);
+            if (pending.length === 0)
+                break;
+            await Promise.allSettled(pending);
+        }
+    }
+    dispose() {
+        clearInterval(this.cleanupInterval);
+        // Clear queue
+        this.queue = [];
+        for (const record of this.agents.values()) {
+            record.session?.dispose();
+        }
+        this.agents.clear();
+    }
+}

package/dist/agent-runner.d.ts

@@ -0,0 +1,60 @@
+/**
+ * agent-runner.ts — Core execution engine: creates sessions, runs agents, collects results.
+ */
+import { type AgentSession, type ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
+import type { Model } from "@mariozechner/pi-ai";
+import type { SubagentType, ThinkingLevel } from "./types.js";
+/** Get the default max turns value. */
+export declare function getDefaultMaxTurns(): number;
+/** Set the default max turns value (minimum 1). */
+export declare function setDefaultMaxTurns(n: number): void;
+/** Get the grace turns value. */
+export declare function getGraceTurns(): number;
+/** Set the grace turns value (minimum 1). */
+export declare function setGraceTurns(n: number): void;
+/** Info about a tool event in the subagent. */
+export interface ToolActivity {
+    type: "start" | "end";
+    toolName: string;
+}
+export interface RunOptions {
+    /** ExtensionAPI instance — used for pi.exec() instead of execSync. */
+    pi: ExtensionAPI;
+    model?: Model<any>;
+    maxTurns?: number;
+    signal?: AbortSignal;
+    isolated?: boolean;
+    inheritContext?: boolean;
+    thinkingLevel?: ThinkingLevel;
+    /** Called on tool start/end with activity info. */
+    onToolActivity?: (activity: ToolActivity) => void;
+    /** Called on streaming text deltas from the assistant response. */
+    onTextDelta?: (delta: string, fullText: string) => void;
+    onSessionCreated?: (session: AgentSession) => void;
+}
+export interface RunResult {
+    responseText: string;
+    session: AgentSession;
+    /** True if the agent was hard-aborted (max_turns + grace exceeded). */
+    aborted: boolean;
+    /** True if the agent was steered to wrap up (hit soft turn limit) but finished in time. */
+    steered: boolean;
+}
+export declare function runAgent(ctx: ExtensionContext, type: SubagentType, prompt: string, options: RunOptions): Promise<RunResult>;
+/**
+ * Send a new prompt to an existing session (resume).
+ */
+export declare function resumeAgent(session: AgentSession, prompt: string, options?: {
+    onToolActivity?: (activity: ToolActivity) => void;
+    signal?: AbortSignal;
+}): Promise<string>;
+/**
+ * Send a steering message to a running subagent.
+ * The message will interrupt the agent after its current tool execution.
+ */
+export declare function steerAgent(session: AgentSession, message: string): Promise<void>;
+/**
+ * Get the subagent's conversation messages as formatted text.
+ */
+export declare function getAgentConversation(session: AgentSession): string;