@signalridge/pi-subagents 1.10.0 → 1.10.2

package/CHANGELOG.md

@@ -1,5 +1,66 @@
 # Changelog
 
+## 1.10.2
+### Patch Changes
+
+- 65e2481: Show supervisor questions in subagent input dialogs and identify the asking agent by its actual name.
+
+## 1.10.1
+### Patch Changes
+
+- 1f586f8: Preserve resumable cancellation when a workflow pauses during successful timeout cleanup, and retain finalized subagent output alongside observer failures. Avoid retaining descendant-owned mouse gestures in bordered components. Track successful writes using final rewritten paths, and distinguish uncertain changes observed during failed calls from confirmed agent modifications while keeping both discoverable in file browsing. Preserve exact filesystem paths and lazy-directory scan state when discovering session-observed changes, including changes injected during an active scan.
+- 1f586f8: - Bound workflow cancellation drains and durably fence unconfirmed cleanup, aborting sibling agents and releasing foreground callers.
+  - Await managed timeout cleanup before script continuation and terminal publication.
+  - Keep failed retry text and length responses discarded by overflow-compaction retry out of successful subagent fallbacks, while retaining valid pre-compaction and non-retried length output.
+  - Account for native prompts spanning reporter activation without dropping out-of-order wait debt.
+  - Reconcile settlement after observer-triggered manual compaction with run/session ownership.
+  - Preserve clipped overlay hit ranges and native container ancestry for editor autocomplete focus.
+  - Reconcile errored writes and edits against pre/post mutation evidence, retaining invocation cwd and releasing evidence when blocked or aborted preflight bypasses tool results.
+- d9219d4: Correct shipped documentation that contradicted the code or the manifest.
+  
+  - `pi-subagents`: the summary line still advertised protocol-v3 spawning while
+    the feature list and the managed-spawn section described v4.
+  - `pi-worktree`: the workspace-switching section named `/tree`; the registered
+    command is `/worktree`.
+  - `pi-workflows`: removed an orphaned sentence fragment left behind by an edit
+    to the saved-workflow paragraph.
+  - `pi-plan-mode`, `pi-goal`, `pi-usage-extension`: the stated Pi floor
+    (`0.80.6`, `0.42.4+`) contradicted the declared `^0.84.0` peer range, which
+    resolves to `0.84.x`. `pi-usage-extension` also carried a "Last updated"
+    stamp from a version that predates the workspace baseline.
+  - `pi-welcome`: the example card was a real session capture. It is now a
+    generic sample, and Install moved above the configuration sections instead of
+    sitting below them.
+- 1f586f8: Expand Pi peer support to `^0.84.0 || ^0.85.0`, retaining 0.84 compatibility while admitting 0.85 releases. The previous zero-major caret range excluded Pi 0.85.1. Update existing Pi development dependency pins to 0.85.1.
+- d9219d4: Converge every README on one house style: plain sentence-case headings, no
+  decorative emoji.
+  
+  Ten packages carried an emoji heading scheme inherited from their upstream
+  forks while the other nineteen used plain headings, so the same monorepo
+  rendered as two unrelated projects on npmjs.com. Headings are now emoji-free
+  and titles are sentence case.
+  
+  Also removes the `Keywords` section from those ten. It duplicated each
+  package's `package.json` `keywords` field, which is what npm actually indexes,
+  and no plain-style README carried one. `Installation` is now `Install`
+  everywhere.
+  
+  Headings that were Title Case are sentence case too, so one convention now
+  covers the whole monorepo. Existing in-page anchor links are unaffected:
+  GitHub lowercases heading slugs already.
+  
+  Documentation only — no runtime change.
+- 1f586f8: Keep goal iteration and continuation numbering correct when Pi compacts between tool results and the next response within one run. Preserve existing response limits and manual/post-run compaction handling.
+  
+  Capture subagent invocation results from finalized assistant events with bounded message retention so compaction cannot hide a final error or mix in historical output. Recognize explicitly selected PowerShell when the host provides it, without requiring its export on older Pi versions or expanding default wildcard tools.
+  
+  Use real session contexts for workflow replay and live TUI progress, release widget bindings on shutdown, and allow foreground checkpoint dialogs in RPC mode while retaining print-mode defaults. Cancel unanswered dialogs with their owning execution on tool abort, pause, stop, disposal, or fatal failure without journaling cancelled answers.
+- Updated dependencies [1f586f8]
+- Updated dependencies [1f586f8]
+- Updated dependencies [1f586f8]
+- Updated dependencies [1f586f8]
+  - @signalridge/pi-ui@1.3.1
+
 ## 1.10.0
 ### Minor Changes
 

package/README.md

@@ -1,6 +1,6 @@
 # @signalridge/pi-subagents
 
-Signalridge's managed subagent package for Pi, including protocol-v3 spawning, lifecycle isolation, Fleet UI, persistence, and recovery.
+Signalridge's managed subagent package for Pi, including protocol-v4 spawning, lifecycle isolation, Fleet UI, persistence, and recovery.
 
 A [pi](https://pi.dev) extension that brings **Claude Code-style autonomous sub-agents** to pi. Spawn specialized agents that run in isolated sessions — each with its own tools, system prompt, model, and thinking level. Run them in foreground or background, steer them mid-run, resume completed sessions, and define your own custom agent types.
 
@@ -50,7 +50,7 @@ pi -e ./packages/pi-subagents
 
 Formatter checks remain disabled for this package, while legacy `noExplicitAny`, control-character regex, and empty-interface rules remain off. Signalridge additions follow the package AGENTS rules and the strict `--error-on-warnings` lint gate.
 
-## Quick Start
+## Quick start
 
 The parent agent spawns sub-agents using the `Agent` tool:
 
@@ -161,7 +161,7 @@ Find auth files · completed
 
 Group completions render each agent as a separate block. The LLM receives structured `<task-notification>` XML for parsing, while the user sees the themed visual.
 
-## Default Agent Types
+## Default agent types
 
 | Type | Tools | Model | Prompt Mode | Description |
 |------|-------|-------|-------------|-------------|
@@ -173,7 +173,7 @@ The `general-purpose` agent is a **parent twin** — it receives the parent's en
 
 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
+## Custom agents
 
 Define custom agent types by creating `.md` files. The filename becomes the agent type name. Any name is allowed — using a default agent's name overrides it.
 
@@ -216,7 +216,7 @@ Then spawn it like any built-in type:
 Agent({ subagent_type: "auditor", prompt: "Review the auth module", description: "Security audit" })
 ```
 
-### Frontmatter Fields
+### Frontmatter fields
 
 All fields are optional — sensible defaults for everything.
 
@@ -376,7 +376,7 @@ Settings
 - **Model tiers** — create, edit and delete the [tier](#model-tiers) profiles. Each row shows what the tier resolves to on this machine (`small — claude-haiku-4-5 · thinking max`), and a tier dropped as malformed is listed as `blocked` so it can be fixed rather than staying invisible. The model picker offers this machine's available models (narrowed to your scope when **Scope models** is on) plus `inherit` and a typed escape hatch; the thinking picker offers only the levels the chosen model actually supports, since the rest would be silently clamped at spawn. Deleting the tier that `defaultTier` names clears the default in the same step
 - **Settings** — configure max concurrency, default max turns, grace turns, default model, default tier, and join mode at runtime
 
-## Graceful Max Turns
+## Graceful max turns
 
 Instead of hard-aborting at the turn limit, agents get a graceful shutdown:
 
@@ -397,7 +397,7 @@ Background agents are subject to a configurable concurrency limit (default: 4).
 
 Foreground agents bypass the queue — they block the parent anyway.
 
-## Join Strategies
+## Join strategies
 
 When background agents complete, they notify the main agent. The **join mode** controls how these notifications are delivered. It applies only to background agents.
 
@@ -640,7 +640,7 @@ then `defaultModel`, or the parent's model when none is set.
 Programmatic callers and the legacy RPC may still pass `model`/`thinking`
 directly. That is the escape hatch for code, not a way to configure an agent.
 
-## Model Scope
+## Model scope
 
 **Opt-in:** off by default. Enable via `/agents → Settings → Scope models`.
 
@@ -662,7 +662,7 @@ When on, each subagent spawn's effective model is validated against pi's own `en
 
 **No-op safety:** if `enabledModels` is missing or empty in pi's settings, scope check skips entirely — no false positives, no spurious errors.
 
-## Persistent Settings
+## Persistent settings
 
 Runtime tuning values set via `/agents` → Settings (max concurrency, default max turns, grace turns, nested depth, fallback agent, default model, default tier, default join mode, scheduling on/off, scope models on/off, strict agent files on/off, disable defaults on/off, output transcript on/off, tool description full/compact/custom, fleet view on/off) persist across pi restarts. Two files, merged on load:
 
@@ -839,7 +839,7 @@ pi.events.emit("subagents:rpc:stop", { requestId, agentId: "agent-id-here" });
 Reply channels are scoped per `requestId`, so concurrent requests don't interfere.
 Workflow callers should pass a foreground signal only to their own wait. Managed child cancellation is explicit through the owner-scoped lifecycle RPC; aborting an RPC wait does not abort the child.
 
-## Persistent Agent Memory
+## Persistent agent memory
 
 Agents can have persistent memory across sessions. Set `memory` in frontmatter to enable:
 
@@ -861,7 +861,7 @@ Memory uses a `MEMORY.md` index file and individual memory files with frontmatte
 
 The `disallowed_tools` field is respected when determining write capability — an agent with `tools: write` + `disallowed_tools: write` correctly gets read-only memory.
 
-## Worktree Isolation
+## Worktree isolation
 
 Set `isolation: worktree` to run an agent in a temporary git worktree:
 
@@ -881,7 +881,7 @@ If cleanup cannot be confirmed, the result keeps the worktree path and includes
 
 The child prompt also states that the temporary worktree is the only writable checkout and names the original base repository as off-limits. This prevents inherited parent instructions from sending the agent back into the shared tree.
 
-## Skill Preloading
+## Skill preloading
 
 Skills can be preloaded by name and injected into the agent's system prompt:
 
@@ -911,7 +911,7 @@ Recursion skips dotfile directories and `node_modules`. A directory that itself
 
 **Security:** symlinks are rejected at every layer (root, flat file, skill directory, `SKILL.md` inside a skill directory) — intentional deviation from Pi, which follows symlinks. Skill names with path-traversal characters (`..`, `/`, `\`, spaces, leading dot, >128 chars) are rejected.
 
-## Tool Denylist
+## Tool denylist
 
 Block specific tools from an agent even if extensions provide them:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@signalridge/pi-subagents",
-  "version": "1.10.0",
+  "version": "1.10.2",
   "description": "Signalridge's managed subagent runtime with workflow-owned orchestration RPC.",
   "author": "tintinweb and signalridge contributors",
   "license": "MIT",
@@ -25,15 +25,15 @@
     "autonomous"
   ],
   "peerDependencies": {
-    "@earendil-works/pi-ai": "^0.84.0",
-    "@earendil-works/pi-coding-agent": "^0.84.0",
-    "@earendil-works/pi-tui": "^0.84.0"
+    "@earendil-works/pi-ai": "^0.84.0 || ^0.85.0",
+    "@earendil-works/pi-coding-agent": "^0.84.0 || ^0.85.0",
+    "@earendil-works/pi-tui": "^0.84.0 || ^0.85.0"
   },
   "dependencies": {
     "@sinclair/typebox": "^0.34.50",
     "croner": "^10.0.1",
     "@signalridge/pi-subagents-protocol": "^1.5.0",
-    "@signalridge/pi-ui": "^1.3.0",
+    "@signalridge/pi-ui": "^1.3.1",
     "nanoid": "^5.0.0"
   },
   "scripts": {

package/src/agent-runner.ts

@@ -7,6 +7,7 @@ import { homedir } from "node:os";
 import { basename, dirname, isAbsolute, join, resolve } from "node:path";
 import type { Model } from "@earendil-works/pi-ai";
 import type { ExtensionContext, LoadExtensionsResult } from "@earendil-works/pi-coding-agent";
+import * as PiCodingAgent from "@earendil-works/pi-coding-agent";
 import {
   type AgentSession,
   type AgentSessionEvent,
@@ -641,6 +642,11 @@ export interface RunResult {
  */
 function collectResponseText(session: AgentSession) {
   let text = "";
+  // Invocation-owned events survive compaction replacing session history.
+  // Retain only the final message and latest nonempty text, not a transcript.
+  let lastMessage: AgentSession["messages"][number] | undefined;
+  let lastNonemptyText = "";
+  let beforeLastMessageText = "";
   const unsubscribe = session.subscribe((event: AgentSessionEvent) => {
     // message_start also fires for user and toolResult messages — resetting on
     // those would wipe assistant text already collected. Reset only when a new
@@ -651,25 +657,33 @@ function collectResponseText(session: AgentSession) {
     if (event.type === "message_update" && event.assistantMessageEvent.type === "text_delta") {
       text += event.assistantMessageEvent.delta;
     }
+    if (event.type === "message_end" && event.message.role === "assistant") {
+      beforeLastMessageText = lastNonemptyText;
+      lastMessage = event.message;
+      text = extractText(event.message.content);
+      // Failed attempts remain visible as the current result, but must never
+      // replace the valid fallback: Pi removes them when retrying. Compaction
+      // may already have removed the earlier valid message from history.
+      if (text.trim() && event.message.stopReason !== "error" && event.message.stopReason !== "aborted") {
+        lastNonemptyText = text.trim();
+      }
+    }
+    if (event.type === "compaction_end" && event.reason === "overflow" && !event.aborted && event.willRetry &&
+        lastMessage?.role === "assistant" && lastMessage.stopReason === "length") {
+      // A length result is legitimate until the SDK actually retries it. On
+      // overflow recovery Pi removes this candidate, including any occurrence
+      // restored by compaction AFTER this notification. Retract its fallback
+      // contribution from invocation state rather than rescanning that history.
+      lastNonemptyText = beforeLastMessageText;
+      text = "";
+      lastMessage = undefined;
+    }
   });
-  return { getText: () => text, unsubscribe };
-}
-
-/**
- * Get the last non-empty assistant text produced during THIS invocation.
- * `startIndex` is the message count captured before the prompt, so the walk-back
- * never crosses into a previous turn: on a resume whose new turn failed empty,
- * this returns "" instead of the prior turn's answer (#144). Defaults to 0 (a
- * fresh spawn, where the whole history belongs to this run).
- */
-function getLastAssistantText(session: AgentSession, startIndex = 0): string {
-  for (let i = session.messages.length - 1; i >= startIndex; i--) {
-    const msg = session.messages[i];
-    if (msg.role !== "assistant") continue;
-    const text = extractText(msg.content).trim();
-    if (text) return text;
-  }
-  return "";
+  return {
+    getText: () => text.trim() || lastNonemptyText,
+    getFailure: () => finalTurnError(lastMessage ? [lastMessage] : []),
+    unsubscribe,
+  };
 }
 
 /**
@@ -683,12 +697,11 @@ function getLastAssistantText(session: AgentSession, startIndex = 0): string {
  * Everything else completes: a clean "stop"/"toolUse" final, and — crucially — a
  * "length" stop that DID produce text (a legitimate truncated-but-useful answer).
  * "aborted" is handled by the manager's abort flag / "stopped" guard, not here.
- * Bounded by `startIndex` (like the text fallback) so a resume that produced no
- * assistant message of its own never inherits a PRIOR turn's stop reason.
+ * Invocation-owned messages exclude historical failures, even after compaction.
  */
-function finalTurnError(session: AgentSession, startIndex = 0): string | undefined {
-  for (let i = session.messages.length - 1; i >= startIndex; i--) {
-    const msg = session.messages[i];
+function finalTurnError(messages: AgentSession["messages"]): string | undefined {
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const msg = messages[i];
     if (msg.role !== "assistant") continue;
     if (msg.stopReason === "error") {
       return (msg as { errorMessage?: string }).errorMessage?.trim() || "provider error with no output";
@@ -984,6 +997,10 @@ export async function runAgent(
   // into the allowlist, then dropped it at registration with no signal back.
   if (agentConfig?.builtinToolNames?.length) {
     const knownBuiltins = new Set(BUILTIN_TOOL_NAMES);
+    // Optional public capability: do not require a new named export on Pi 0.84.
+    if ("createPowerShellTool" in PiCodingAgent && typeof Reflect.get(PiCodingAgent, "createPowerShellTool") === "function") {
+      knownBuiltins.add("powershell");
+    }
     for (const name of agentConfig.builtinToolNames) {
       if (!knownBuiltins.has(name)) {
         options.onToolActivity?.({
@@ -1109,7 +1126,9 @@ export async function runAgent(
   const supervisorTools =
     ctx.hasUI && !isolated && options.supervisorQuestions !== false
       ? createSupervisorTool({
-          agentLabel: agentConfig?.displayName ?? type,
+          // Use the canonical agent type, not display_name: the question must
+          // identify the actual subagent that requested the decision.
+          agentLabel: agentConfig?.name ?? type,
           ask: {
             input: (title, placeholder) => ctx.ui.input(title, placeholder),
             select: (title, choices) => ctx.ui.select(title, choices),
@@ -1167,6 +1186,10 @@ export async function runAgent(
     const denyTools = new Set<string>(
       EXCLUDED_TOOL_NAMES.filter((t) => !nestedToolNames.has(t)),
     );
+    // Keep the historical wildcard defaults; PowerShell is explicit opt-in.
+    if ("createPowerShellTool" in PiCodingAgent && typeof Reflect.get(PiCodingAgent, "createPowerShellTool") === "function" && !builtinToolNameSet.has("powershell")) {
+      denyTools.add("powershell");
+    }
     // Keep only the built-ins the agent asked for — deny the rest.
     for (const name of BUILTIN_TOOL_NAMES) {
       if (!builtinToolNameSet.has(name)) denyTools.add(name);
@@ -1392,6 +1415,9 @@ export async function runAgent(
     handle.unref?.();
     pendingToolTimeouts.set(toolCallId, handle);
   };
+  // Own finalized output before observer dispatch: a synchronous callback
+  // failure must remain visible without erasing the answer it interrupted.
+  const collector = collectResponseText(session);
   const unsubTurns = session.subscribe((event: AgentSessionEvent) => {
     if (event.type === "turn_end") {
       turnCount++;
@@ -1442,8 +1468,6 @@ export async function runAgent(
     }
   });
 
-  const collector = collectResponseText(session);
-
   // Build the effective prompt: optionally prepend parent context
   let effectivePrompt = prompt;
   const inheritContext = internalOverride?.inheritContext ?? options.inheritContext;
@@ -1454,9 +1478,6 @@ export async function runAgent(
     }
   }
 
-  // Boundary for the history fallback: only assistant text produced from here
-  // on counts as this run's output (a fresh session, so usually 0).
-  const startLen = session.messages.length;
   try {
     throwIfAborted(options.signal);
     await session.prompt(effectivePrompt);
@@ -1467,7 +1488,7 @@ export async function runAgent(
     pendingToolTimeouts.clear();
   }
 
-  const baseText = collector.getText().trim() || getLastAssistantText(session, startLen);
+  const baseText = collector.getText();
   // The acceptance gate runs AFTER the agent is done and its verdict is
   // appended to the result, so the parent reads the check and the claim it is
   // checking side by side. It deliberately does not steer the agent to fix
@@ -1476,7 +1497,7 @@ export async function runAgent(
     ? `${baseText}${await runConfiguredGate(agentConfig.gate, effectiveCwd, options.pi)}`
     : baseText;
   completed = true;
-  return { responseText, session, aborted, steered: softLimitReached, failure: finalTurnError(session, startLen) };
+  return { responseText, session, aborted, steered: softLimitReached, failure: collector.getFailure() };
   } finally {
     cleanupAbort();
     // The manager takes ownership only after onSessionCreated is invoked. If
@@ -1505,11 +1526,7 @@ export async function resumeAgent(
     signal?: AbortSignal;
   } = {},
 ): Promise<{ text: string; failure?: string }> {
-  // Boundary for the history fallback: the session already holds prior turns,
-  // so only assistant text produced by THIS resume prompt counts as its output
-  // — a failed resume must not surface the previous turn's answer (#144).
   throwIfAborted(options.signal);
-  const startLen = session.messages.length;
   const collector = collectResponseText(session);
   const cleanupAbort = forwardAbortSignal(session, options.signal);
 
@@ -1574,8 +1591,8 @@ export async function resumeAgent(
   }
 
   return {
-    text: collector.getText().trim() || getLastAssistantText(session, startLen),
-    failure: finalTurnError(session, startLen),
+    text: collector.getText(),
+    failure: collector.getFailure(),
   };
 }
 

package/src/supervisor.ts

@@ -36,7 +36,7 @@ export interface SupervisorAsk {
 export interface SupervisorToolContext {
   /** The parent's UI. Omitted when there is no human to ask. */
   ask?: SupervisorAsk;
-  /** Display name of the asking agent, for the prompt title. */
+  /** Actual type/name of the asking agent, for the prompt title. */
   agentLabel: string;
 }
 
@@ -82,7 +82,10 @@ export function createSupervisorTool(context: SupervisorToolContext) {
         const question = truncateCodePoints(sanitizeDisplayText(params.question), MAX_QUESTION, "…").trim();
         if (!question) return textResult("Ask a non-empty question.", true);
 
-        const title = `${context.agentLabel} asks`;
+        const title = `${sanitizeDisplayText(context.agentLabel)} asks`;
+        // Pi 0.85.1 ignores ui.input()'s placeholder argument. Keep the
+        // question in the title so the human sees it in both old and new hosts.
+        const promptTitle = `${title}: ${question}`;
         const options = (params.options ?? [])
           .map((option) => truncateCodePoints(sanitizeDisplayText(option), MAX_OPTION, "…").trim())
           .filter((option) => option.length > 0)
@@ -92,8 +95,8 @@ export function createSupervisorTool(context: SupervisorToolContext) {
         try {
           answer =
             options.length > 0
-              ? await ask.select(`${title}: ${question}`, options)
-              : await ask.input(title, question);
+              ? await ask.select(promptTitle, options)
+              : await ask.input(promptTitle);
         } catch (error: unknown) {
           // A dialog that cannot open must not fail the child's whole run; it
           // is told nobody answered and carries on under its own judgement.