@gotgenes/pi-subagents 18.0.1 → 18.0.3

package/docs/architecture/history/phase-19-implement-ui-decisions.md

@@ -0,0 +1,282 @@
+# Phase 19: Implement the ADR-0004 UI decisions
+
+## Summary
+
+Phase 19 implements the per-component UI decisions recorded in [ADR-0004]: shrink the widget to background-only, replace the bespoke conversation viewer with native session navigation, dissolve the monolithic `/agents` menu, and keep the surviving UI in-core.
+
+The sequencing follows Kent Beck's "make the change easy, then make the easy change."
+The end state deletes `agent-menu.ts` — the god-command that bundles four unrelated jobs — and everything reachable only from it.
+Rather than surgically mutate that doomed module (and the #1 churn hotspot `index.ts`) once per option, Phase 19 first stood up the replacement surfaces additively, then removed the now-orphaned subtree in a single terminal cut.
+This kept every responsibility's old surface live until its replacement existed (ADR-0004's no-interim-regression invariant), turned the three replacement steps into genuinely parallel work (none touched `agent-menu.ts`), and reduced `index.ts` edits from four surgical removals to one deregistration.
+
+Seven numbered steps in three phases, plus two follow-ups (Steps 4a–4b) carved from the #445 slice:
+
+- **Phase A — stand up replacements (additive):** spike, settings command, background widget, native session navigation and its renderer/source follow-ups (Steps 1–4, 4a–4b).
+- **Phase B — dissolve `/agents` (terminal cut):** delete the orphaned subtree in two deletion commits, one per subtree (Steps 5–6).
+- **Phase C — test health:** consolidate the test clones that survive the cut (Step 7).
+
+All nine steps are closed: [#446], [#447], [#444], [#445], [#462], [#463], [#442], [#441], [#443].
+A follow-on issue, [#470] (README staleness — the terminal cut removed `/agents` but left the README describing it), was filed after Steps 5–6 shipped and closed independently; see "Follow-on issues" below.
+
+## Health metrics
+
+| Metric                 | Phase 18 (start)         | Phase 19 target      | Phase 19 (delivered)                 |
+| ---------------------- | ------------------------ | -------------------- | ------------------------------------ |
+| Health score           | 78/100 (B)               | 83/100 (B+)          | 78/100 (B) — unchanged               |
+| Source LOC             | 7,650 (61 files)         | ~6,780 (~55 files)   | 7,068 (57 files)                     |
+| Production duplication | 11 lines (1 group)       | 0 lines              | 0 lines ✅                           |
+| Test clone groups      | 16                       | ≤ 10                 | 9 ✅                                 |
+| Top churn hotspot      | `index.ts` (103 commits) | `index.ts` (cooling) | `index.ts` (109 commits, cooling) ✅ |
+
+The health score held flat at 78/100 rather than reaching the 83/100 target — the score's `hotspots` and `unit size` deductions are dominated by long-lived test-suite characteristics (large test functions, `index.ts`'s cumulative churn history) that this phase's scope did not target.
+Every metric the phase's steps directly controlled — production duplication, test clone groups, and the churn trend on `index.ts` — hit or beat its target.
+
+## Steps
+
+### Step 1 — Spike: resolve ADR-0004 entry criteria ([#446])
+
+Smell: Category C (coupling boundary) — four open decisions block the session-navigation implementation.
+Target: `docs/decisions/0004-reconsider-ui-direction.md` addendum.
+
+The four entry criteria from ADR-0004:
+
+1. **Root-continuity:** Does the root's in-flight turn survive `ctx.switchSession()` and a return gesture?
+2. **View-only vs interactive:** `switchSession` (full interactive takeover) or `loadEntriesFromFile` (read-only transcript built from JSONL)?
+3. **Parallel-agent navigation:** Operator gesture to select which of N background agents to view (from the widget, a command, or both).
+4. **Settings command name:** `/subagents-settings`, `/agents-settings`, or another form consistent with sibling packages?
+
+Produce a minimal spike (observed test or PoC against a real session) that answers each question, then record the answers as an addendum to ADR-0004.
+No production source files change; the spike closes when the ADR addendum is merged.
+
+Outcome: ADR-0004 updated with all four entry-criteria answers; Step 4 unblocked.
+
+`Release: independent`
+
+### Step 2 — Extract settings to a focused `/subagents-settings` command ([#447])
+
+Smell: Category E (naming/organization) — settings are buried inside the monolithic `/agents` command per ADR-0004 Decision C. This step is purely additive: it stands up the new surface without touching `agent-menu.ts`.
+Target files:
+
+- New `src/ui/subagents-settings.ts` — `SubagentsSettingsHandler` lifted from `AgentsMenuHandler.showSettings`, carrying its own narrow `SubagentsSettingsManager` interface (the three `apply*` methods and three readonly accessors only).
+- `src/index.ts` — register the new command (name confirmed by Step 1); pass `settings` directly.
+- New `test/ui/subagents-settings.test.ts` — unit tests for the extracted handler.
+
+`showSettings` depends only on `this.settings` (the self-contained `AgentMenuSettings` shape), so the extraction copies that logic into a new file with zero coupling to the wizard, editor, or viewer.
+The old in-menu Settings option keeps working until the terminal cut deletes `agent-menu.ts` wholesale — there is no surgical removal of `showSettings` or `AgentMenuSettings` from the doomed file.
+
+Outcome: new `subagents-settings.ts` (~80 LOC) and focused command registered; `agent-menu.ts` untouched.
+
+`Release: independent`
+
+### Step 3 — Shrink widget to background agents only ([#444])
+
+Smell: Category C (coupling) — the widget shows all agents including foreground ones, duplicating the `subagent` tool's inline `onUpdate` stream for foreground runs.
+Target files:
+
+- `src/ui/agent-widget.ts` — funnel both `manager.listAgents()` call sites (`update()` and `renderWidget()`) through a single private accessor, then flip that accessor to background-only via `record.invocation?.runInBackground === true`.
+- `src/ui/widget-renderer.ts` — verify no foreground-specific rendering path survives.
+- `test/ui/agent-widget.test.ts` — add background-only filtering tests; update assertions.
+
+The widget calls `listAgents()` at two sites today — `update()` (feeding `seedFinishedAgents`, `assembleWidgetState`, and `clearWidget`) and `renderWidget()` (the tree map).
+Filtering at only one site leaves the other rendering foreground agents, so the enabling move is to route both through one accessor and apply the predicate once at the source.
+`Subagent.invocation.runInBackground` is the reliable signal: set by `spawn-config.ts` → `AgentInvocation.runInBackground` → stored on `Subagent.invocation`.
+ADR-0004 Decision A: foreground runs suppress the widget; the inline `onUpdate` stream is authoritative there.
+
+Outcome: widget shows only background agents; foreground/widget duplication eliminated; the background predicate lives at a single funnel.
+
+`Release: independent`
+
+### Step 4 — Implement native session navigation ([#445])
+
+Smell: Category C (coupling) — the bespoke `ConversationViewer` re-implements session-transcript rendering when Pi's own machinery targets the already-persisted child session JSONL.
+This step adds the new surface alongside the existing viewer; it does not touch `agent-menu.ts`.
+Target files:
+
+- New `src/ui/session-navigator.ts` — a flat command that lists any subagent with a live record or a persisted session file (foreground included, never background-filtered), lets the operator pick one, and renders that child's transcript read-only.
+- New typed accessor on `Subagent`/`SubagentSession` returning `record.messages` as `AgentMessage[]` (the boundary currently widens it to `readonly unknown[]`).
+- `src/index.ts` — register the new command; the background widget ([#444]) is an optional secondary selection gesture, not a dependency.
+
+ADR-0004 Decision B: "Tell-Don't-Ask — hand Pi the session path; Pi owns the viewer."
+Mechanism (confirmed by the Step 1 spike and revised by [ADR-0004] Addendum 2): a **read-only** (non-interactive) transcript **dual-sourced by liveness**, rendered through Pi's own public entry components (no bespoke renderer).
+
+- **Tracked agent** (still in `manager.listAgents()`) — render live from the in-memory record: `record.messages` for history, `record.subscribeToUpdates()` to re-render on streaming updates, and `record.activeTools` / `record.responseText` for the running-agent streaming indicator.
+- **Evicted / untracked agent** — render from the file snapshot: `parseSessionEntries(readFileSync(record.outputFile, "utf8"))` → drop the `SessionHeader` → `buildSessionContext(...).messages`.
+
+Both sources yield `AgentMessage[]`, so one Pi-component renderer serves both: Pi's public entry components (`AssistantMessageComponent` / `ToolExecutionComponent` / …) or `serializeConversation` (see the [ADR-0004] addendum, Findings 0 and 1).
+Neither `switchSession` (a full takeover that invalidates the root's in-flight turn) nor `loadEntriesFromFile` (a test-only export the package's public barrel does not re-export, in both `0.79.1` and `0.79.8`) is used.
+`Subagent.outputFile` already exposes the persisted child session JSONL path via `subagentSession?.outputFile` — no new SDK dependency.
+The new surface stands up while the old `viewAgentConversation`/`ConversationViewer` path still works; the bespoke viewer is removed only by the terminal cut (Step 5).
+
+Outcome: operator views any subagent's session through Pi's native machinery — live for a running agent, a file snapshot for an evicted one; the new surface coexists with the old viewer until Step 5.
+
+Landed ([#445], sliced): #445 shipped the first releasable vertical slice — the `/subagent-sessions` command (`src/ui/session-navigator.ts`), the pure selection/sourcing/text-render core (`src/ui/session-navigation.ts`), and the typed `agentMessages` accessor (`SessionMessage` on `SubagentSession`/`Subagent`).
+It is **live-source only** behind a renderer-agnostic `TranscriptSource` seam, rendered via Pi's `serializeConversation` text.
+With the `manager.listAgents()`-only candidate set, no listed record is ever session-disposed (dispose-and-delete are atomic), so the file-snapshot branch has no reachable caller and was deferred to keep `fallow dead-code` clean.
+Step 4 (#445, the slice) is complete and released (`pi-subagents` v17.3.0); the remaining work is now tracked as two follow-up steps behind the same seam: Step 4a ([#462]) upgrades the renderer from `serializeConversation` text to Pi's per-entry TUI components (gates Step 5 for rendering parity); Step 4b ([#463]) broadens the candidate set to evicted agents and adds the file-snapshot source (`parseSessionEntries` → `buildSessionContext`, independent).
+
+`Release: independent` (spike-gated)
+
+### Step 4a — Upgrade native-navigation renderer to Pi TUI components ([#462])
+
+Smell: Category C (coupling) — the #445 slice renders the transcript as `serializeConversation` plain text, while the bespoke `ConversationViewer` it replaces renders richer per-message formatting.
+Until the native renderer reaches parity, the terminal cut (Step 5) cannot delete the bespoke viewer without a fidelity regression.
+This step swaps the renderer behind the existing `TranscriptSource` seam (`src/ui/session-navigation.ts` / `session-navigator.ts`) for Pi's per-entry components (`AssistantMessageComponent` / `ToolExecutionComponent` / …); selection and sourcing are untouched.
+
+Gates Step 5: per [ADR-0004]'s no-interim-regression invariant, the native navigator must reach rendering parity with the bespoke viewer before Step 5 deletes it.
+
+Outcome: native session navigation renders at parity with the removed `ConversationViewer`; Step 5 can delete the bespoke viewer with no fidelity regression.
+
+Landed ([#462]): the renderer now mounts Pi's per-entry components (`AssistantMessageComponent` / `ToolExecutionComponent` / `BashExecutionComponent` / `UserMessageComponent` / `CompactionSummaryMessageComponent` / `BranchSummaryMessageComponent` / `SkillInvocationMessageComponent`) into a `Container`, mirroring Pi's own `renderSessionContext` mapping.
+The `TranscriptOverlay` caches that `Container` and rebuilds it on source change only (Pi's `rebuildChatFromMessages` path), keeping the lightweight `◍` streaming indicator.
+Tool calls render with their real `ToolDefinition`, resolved through a dependency-safe `getToolDefinition` read accessor on the record (mirroring `agentMessages`) surfaced on the `TranscriptSource` seam — no inbound call into the core.
+The pure `session-navigation.ts` sheds `renderTranscriptLines`/`serializeConversation`; rendering now lives in the SDK/TUI `session-navigator.ts`, which threads `cwd` from the command context.
+`custom`-role messages are skipped (the bespoke viewer never rendered them either).
+Selection and sourcing are untouched; native navigation now renders at parity, unblocking Step 5 for rendering fidelity.
+
+`Release: independent`
+
+### Step 4b — File-snapshot source for evicted agents ([#463])
+
+Smell: Category C (coupling) — the #445 slice sources transcripts live from `manager.listAgents()` only; an agent evicted by the 10-minute cleanup sweep has a persisted session JSONL but no live record, so it is unreachable.
+This step adds the file-snapshot `TranscriptSource` branch (`parseSessionEntries(readFile(outputFile))` → drop the `SessionHeader` → `buildSessionContext(...).messages`) and broadens the candidate set to evicted agents, behind the same seam; the renderer is untouched.
+
+Independent: this is a new capability the bespoke viewer never had, so it gates nothing and is not a Step 5 prerequisite.
+Best sequenced after Step 4a (shared renderer), but carries no hard dependency.
+
+Outcome: the operator can view a fully-evicted agent's transcript from its persisted session file; the dual-source design recorded in [ADR-0004] Addendum 2 is fully realized.
+
+Landed ([#463]): `fileSnapshotSource(outputFile, readFile)` lands in the pure `session-navigation.ts` (`parseSessionEntries` → drop the `SessionHeader` → `buildSessionContext(...).messages`; a static no-subscribe, no-streaming source).
+The candidate set is broadened via **manager-retained descriptors**, not a directory scan: the persisted child session carries no subagent `type`/`description` (those live only on the in-memory record), so a scan would yield degraded labels and parse every file per open.
+Instead `SubagentManager.cleanup()` stashes a lightweight `EvictedSubagent` descriptor (label fields + `outputFile`, no messages) before disposing a record with a persisted file, exposed via `listEvicted()` and cleared by `clearCompleted()`/`dispose()`.
+`NavigationEntry` became a `live | evicted` discriminated union; the handler selects `liveSource` vs `fileSnapshotSource` by kind inside a `try/catch` (an unreadable file notifies and skips), and `index.ts` injects `readFileSync`.
+Evicted entries carry an `· evicted (snapshot)` label marker.
+Coverage is in-session evictions (the sweep's only targets); old-session orphan files — the ones a scan would surface with degraded labels — are out of scope.
+
+`Release: independent`
+
+### Step 5 — Dissolve `/agents` and remove the conversation-viewer subtree ([#442])
+
+Smell: Category A (dead subsystem) plus Category B (oversized) — once Steps 2–4 re-home all four menu responsibilities, the `/agents` command and everything reachable only from `agent-menu.ts` is an unreferenced subtree.
+This is the first of two deletion commits (split by subtree).
+The hub `agent-menu.ts` is deleted here, not surgically narrowed, and deleting it is what orphans the leaf subtrees — so it must precede the definition-management deletion (Step 6), because `agent-menu.ts` statically imports the wizard, editor, and file-ops, and dynamically imports the viewer.
+Target files:
+
+- `src/index.ts` — remove the `registerCommand("agents", …)` block, the `AgentsMenuHandler` construction and import, and the `FsAgentFileOps` import/construction (its only use is wiring the menu).
+- Delete `src/ui/agent-menu.ts` (331 LOC) and `test/ui/agent-menu.test.ts` (185 LOC).
+- Delete `src/ui/conversation-viewer.ts` (241 LOC) and `test/conversation-viewer.test.ts` (239 LOC) — its only consumer is `agent-menu.ts`'s dynamic import, gone with the hub.
+- Delete `src/ui/message-formatters.ts` (195 LOC) and `test/message-formatters.test.ts` (388 LOC, the largest test function by LOC) — its only consumer is `ConversationViewer`.
+
+Running-agent visibility is now owned by the background widget (Step 3); session navigation replaces the bespoke overlay (Step 4); settings live in `/subagents-settings` (Step 2).
+Deleting the hub in one move avoids any surgical edit to the doomed file and leaves the definition-management leaves orphaned for Step 6.
+
+Actual approach (vs. original plan): a tidy-first preparatory commit first extracted `MenuUI` from `agent-menu.ts` into a new `src/ui/menu-ui.ts` module, breaking the bidirectional type cycle — the hub imported the wizard/editor classes while the leaves imported back the `MenuUI` type.
+Without the prep commit, deleting either subtree first would have left the other half referencing a deleted module.
+`index.ts` also shed the now-dead `join` (node:path) and `buildParentSnapshot` imports in addition to the two module imports.
+The `menu-ui.ts` module is transient and is removed with the wizard/editor in Step 6.
+
+Outcome: ✅ `/agents` dissolved; −767 LOC source (menu hub + viewer + formatters); −812 LOC test; largest test function eliminated; `index.ts` dewired.
+
+`Release: batch "dissolve-agents"`
+
+### Step 6 — Remove the orphaned agent-definition management subtree ([#441])
+
+Smell: Category A (dead subsystem) — the creation wizard and config editor are removed per ADR-0004 Decision C; after Step 5 deletes their only importer (`agent-menu.ts`), they and their file-ops helpers are pure orphans.
+This is the second deletion commit (split by subtree).
+Target files:
+
+- Delete `src/ui/agent-creation-wizard.ts` (233 LOC) and `test/ui/agent-creation-wizard.test.ts` (296 LOC).
+- Delete `src/ui/agent-config-editor.ts` (199 LOC) and `test/ui/agent-config-editor.test.ts` (392 LOC) — eliminates the 11-line internal production clone in `disableAgent`/`ejectAgent`, the package's only remaining production duplication.
+- Delete `src/ui/agent-file-ops.ts` (59 LOC) and `test/ui/agent-file-ops.test.ts` (112 LOC) — only consumers were wizard + editor.
+- Delete `src/ui/agent-file-writer.ts` (55 LOC) and `test/ui/agent-file-writer.test.ts` (148 LOC) — only consumers were wizard + editor.
+- `test/helpers/ui-stubs.ts` — delete `makeFileOps`, `createTestSubagentConfig`, and `spawnAndWait` from `makeMenuManager` if no surviving consumer remains; delete the file outright once all consumers are gone.
+
+An operator generates a new agent `.md` by asking a Pi session directly (more capable than a fixed wizard) or by writing the file in an editor; viewing and editing definitions is served by opening the `.md` files in an editor or IDE.
+These files are orphaned by Step 5, so this is a pure `git rm` with no surviving references and no edit to any doomed file.
+
+Outcome: −546 LOC source (wizard + editor + file-ops + file-writer); −948 LOC test; production duplication → 0 lines; 1 production and 1 test clone group eliminated.
+
+Landed ([#441]): deleted `agent-creation-wizard.ts`, `agent-config-editor.ts`, `agent-file-ops.ts`, `agent-file-writer.ts`, and `menu-ui.ts` (orphaned transient) from `src/ui/`, plus their four test files.
+`test/helpers/ui-stubs.ts` pruned to `makeMenuUI` only (still consumed by `subagents-settings.test.ts`); `makeFileOps`, `makeMenuManager`, and `createTestSubagentConfig` removed with their `ui-stubs.test.ts` describe blocks.
+Production duplication: 0 lines (confirmed by `fallow dupes`); `fallow dead-code`: clean. 16 test clone groups remain — Step 7 consolidation target.
+
+`Release: batch "dissolve-agents"`
+
+### Step 7 — Consolidate remaining test clone families ([#443])
+
+Smell: Category D (testability) — 16 clone groups at Phase 18 end; the terminal cut (Steps 5–6) removes ~4 groups; remaining groups are extraction targets.
+Run after the cut so no helper is extracted into a file the cut then deletes.
+Target files:
+
+- `test/lifecycle/subagent-manager.test.ts` — extract a shared assertion helper for 3 clone families (23 lines across groups at :92/:109, :282/:330, and :323 shared with `subagent.test.ts`).
+- `test/ui/agent-widget.test.ts` — merge the duplicate `makeWidget` helper defined twice across two `describe` blocks (14-line clone at :225/:284).
+- `test/session/session-config.test.ts` — extract a shared fixture for the 16-line internal clone (lines 131–146 / 151–166).
+- `test/lifecycle/concurrency-limiter.test.ts` — extract shared setup for the 10-line clone (lines 21–30 / 148–155).
+- `test/tools/spawn-config.test.ts` — extract a shared fixture for the 9-line clone (lines 22–30 / 35–43).
+
+Outcome (landed): test clone groups reduced from 16 to 9 (≤ 10 target met).
+Eight genuine arrange/fixture/helper families were extracted: `makeNavigable` (shared `test/helpers/make-navigable.ts`), `emitResumeUsageAndCompaction` (shared `test/helpers/mock-session.ts`), and local helpers for `makeWidget`, the captured-overlay render (`renderCapturedOverlay`), the `resultConsumed` observer (`seedResultConsumedObserver`), the ready subagent (`makeReadySubagent`), and the prepared bracket (`preparedBracket`).
+The nine residual families are the repeated system-under-test call (`resolveSpawnConfig`, `assembleSessionConfig`, `schedule`, `SessionNavigatorHandler.handle`, `spawnBg`+`await`, `agent.run()`, `execute`), left intact per the testing guardrail — the repeated act is the test subject, not duplication to remove.
+One family the plan pre-classified as captured-overlay boilerplate (`dup:ea0a1bce`) proved to be an act-clone once the boilerplate was extracted, so it joins the residual set rather than being wrapped.
+
+`Release: independent`
+
+## Step dependency diagram
+
+```mermaid
+flowchart LR
+    S1["✅ Step 1 - Spike (#446)"]
+    S2["✅ Step 2 - Settings command (#447)"]
+    S3["✅ Step 3 - Background widget (#444)"]
+    S4["✅ Step 4 - Native session nav slice (#445)"]
+    S4a["✅ Step 4a - Renderer to TUI components (#462)"]
+    S4b["✅ Step 4b - File-snapshot source (#463)"]
+    S5["✅ Step 5 - Dissolve /agents + viewer (#442)"]
+    S6["✅ Step 6 - Remove definition mgmt (#441)"]
+    S7["✅ Step 7 - Test clones (#443)"]
+
+    S1 --> S4
+    S4 --> S4a
+    S4 --> S4b
+    S2 --> S5
+    S3 --> S5
+    S4a --> S5
+    S5 --> S6
+    S6 --> S7
+```
+
+The terminal cut (Step 5) depends on all three replacements — settings (Step 2), widget (Step 3), and session navigation **at rendering parity** (Step 4a, which completes the #445 slice) — because each of the four `/agents` options must have its responsibility re-homed, and the viewer replacement at parity, before its branch can die.
+Step 4b (file-snapshot source) is a new capability and gates nothing.
+The old `S1 → S6 → S7` chain hid the widget dependency; this diagram makes it explicit.
+
+## Parallel tracks
+
+- **Track A — Replacements (Steps 1–4):** the spike gates session navigation (Step 1 → Step 4); settings (Step 2) and the background widget (Step 3) are independent of the spike and of each other.
+  None of these steps edits `agent-menu.ts`, so they carry no shared-file collision on the menu — genuinely parallelizable, unlike the prior plan's Steps 2/3/5, which all collided on `agent-menu.ts` and `index.ts`.
+  Steps 2 and 4 each append a command registration to `index.ts` (additive, low-conflict).
+  Steps 4a (renderer parity) and 4b (file-snapshot source) complete the #445 slice behind its `TranscriptSource` seam; Step 4a gates Step 5, Step 4b is independent.
+- **Track B — Dissolution (Steps 5 → 6):** the terminal cut, gated on all of Track A landing.
+  Hub-first ordering is forced: Step 5 deletes `agent-menu.ts` (orphaning the leaves), then Step 6 `git rm`s the now-orphaned definition-management subtree.
+- **Track C — Test health (Step 7):** clone consolidation, run after the cut so no surviving helper is extracted into a doomed file.
+
+## Release batches
+
+- **Batch "dissolve-agents":** Steps 5, 6 (ship together; tail = Step 6).
+  Depends on Steps 2, 3, 4 already merged.
+- Independently releasable: Steps 1, 2, 3, 4, 4a, 4b, 7.
+
+## Follow-on issues
+
+- [#470] — the terminal cut (Steps 5–6) removed `/agents`, the conversation viewer, and the creation wizard/config editor, but `README.md` was never updated and continued documenting the removed surface through the `pi-subagents-v18.0.0` release.
+  Filed and closed independently after the phase's steps landed; the README now documents `/subagents:settings`, `/subagents:sessions`, and the background widget.
+
+[ADR-0004]: ../../decisions/0004-reconsider-ui-direction.md
+[#441]: https://github.com/gotgenes/pi-packages/issues/441
+[#442]: https://github.com/gotgenes/pi-packages/issues/442
+[#443]: https://github.com/gotgenes/pi-packages/issues/443
+[#444]: https://github.com/gotgenes/pi-packages/issues/444
+[#445]: https://github.com/gotgenes/pi-packages/issues/445
+[#446]: https://github.com/gotgenes/pi-packages/issues/446
+[#447]: https://github.com/gotgenes/pi-packages/issues/447
+[#462]: https://github.com/gotgenes/pi-packages/issues/462
+[#463]: https://github.com/gotgenes/pi-packages/issues/463
+[#470]: https://github.com/gotgenes/pi-packages/issues/470

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gotgenes/pi-subagents",
-  "version": "18.0.1",
+  "version": "18.0.3",
   "type": "module",
   "exports": {
     ".": {
@@ -15,11 +15,10 @@
   "files": [
     "src",
     "dist",
-    "docs",
-    "vitest.config.ts",
-    "AGENTS.md",
-    "CHANGELOG.md",
-    ".prettierignore"
+    "docs/*.md",
+    "docs/architecture",
+    "docs/decisions",
+    "CHANGELOG.md"
   ],
   "imports": {
     "#src/*": "./src/*",
@@ -67,7 +66,7 @@
     "@earendil-works/pi-coding-agent": "0.79.1",
     "@earendil-works/pi-tui": "0.79.1",
     "@types/node": "^22.15.3",
-    "rollup": "^4.61.1",
+    "rollup": "^4.62.2",
     "rollup-plugin-dts": "^6.4.1",
     "rumdl": "^0.2.10",
     "typescript": "^6.0.3",

package/src/config/default-agents.ts

@@ -15,6 +15,7 @@ export const DEFAULT_AGENTS: Map<string, AgentConfig> = new Map([
       name: "general-purpose",
       displayName: "Agent",
       description: "General-purpose agent for complex, multi-step tasks",
+      toolGuideline: "- Use general-purpose for complex tasks that need file editing.",
       // builtinToolNames omitted — means "all available tools" (resolved at lookup time)
       // inheritContext / runInBackground omitted — strategy fields, callers decide per-call.
       // Setting them to false would lock callsite intent (see resolveAgentInvocationConfig in invocation-config.ts).
@@ -29,6 +30,7 @@ export const DEFAULT_AGENTS: Map<string, AgentConfig> = new Map([
       name: "Explore",
       displayName: "Explore",
       description: "Fast codebase exploration agent (read-only)",
+      toolGuideline: "- Use Explore for codebase searches and code understanding.",
       builtinToolNames: READ_ONLY_TOOLS,
       model: "anthropic/claude-haiku-4-5-20251001",
       systemPrompt: `# CRITICAL: READ-ONLY MODE - NO FILE MODIFICATIONS
@@ -69,6 +71,7 @@ Use Bash ONLY for read-only operations: ls, git status, git log, git diff, find,
       name: "Plan",
       displayName: "Plan",
       description: "Software architect for implementation planning (read-only)",
+      toolGuideline: "- Use Plan for architecture and implementation planning.",
       builtinToolNames: READ_ONLY_TOOLS,
       systemPrompt: `# CRITICAL: READ-ONLY MODE - NO FILE MODIFICATIONS
 You are a software architect and planning specialist.

package/src/lifecycle/subagent.ts

@@ -17,7 +17,6 @@ import { SubagentState, type SubagentStatus } from "#src/lifecycle/subagent-stat
 import type { LifetimeUsage } from "#src/lifecycle/usage";
 import type { WorkspaceProvider } from "#src/lifecycle/workspace";
 import { WorkspaceBracket } from "#src/lifecycle/workspace-bracket";
-import { NotificationState } from "#src/observation/notification-state";
 import { subscribeSubagentObserver } from "#src/observation/record-observer";
 import type { RunConfig } from "#src/runtime";
 import type { AgentInvocation, CompactionInfo, ParentSessionInfo, SessionMessage, SubagentType, ThinkingLevel } from "#src/types";
@@ -36,6 +35,16 @@ export interface SubagentLifecycleObserver {
 
 export type { SubagentStatus } from "#src/lifecycle/subagent-state";
 
+/**
+ * The result of a steer attempt. `Subagent.steer` owns the non-running
+ * rejection rule and reports it here, so coordinators switch on the outcome
+ * instead of pre-checking status (tell by id, with outcomes).
+ */
+export type SteerOutcome =
+	| { kind: "delivered" }
+	| { kind: "buffered" }
+	| { kind: "rejected"; status: SubagentStatus };
+
 /**
  * The execution machinery a Subagent needs to run. A single mandatory
  * collaborator: production (SubagentManager.spawn) always supplies it, so run()
@@ -108,8 +117,6 @@ export class Subagent {
 	private readonly workspaceBracket: WorkspaceBracket;
 
 	subagentSession?: SubagentSession;
-	private _notification?: NotificationState;
-	get notification(): NotificationState | undefined { return this._notification; }
 
 	// Steer buffer — messages queued before the session is ready
 	private _pendingSteers: string[] = [];
@@ -121,22 +128,32 @@ export class Subagent {
 		return this.subagentSession?.outputFile;
 	}
 
+	/** The tool call ID that spawned this background agent, if any. */
+	get toolCallId(): string | undefined {
+		return this.execution.parentSession?.toolCallId;
+	}
+
 	/** Returns true when a SubagentSession is available (session is ready). */
 	isSessionReady(): boolean {
 		return this.subagentSession != null;
 	}
 
 	/**
-	 * Deliver or buffer a steer message.
-	 * Returns true when delivered immediately; false when buffered for later delivery.
+	 * Steer a running agent, owning the non-running rejection rule.
+	 * Returns a `rejected` outcome (with the observed status) when the agent is
+	 * not running, a `buffered` outcome when the session is not yet ready, or a
+	 * `delivered` outcome once the message reaches the session.
 	 */
-	async steer(message: string): Promise<boolean> {
+	async steer(message: string): Promise<SteerOutcome> {
+		if (this.status !== "running") {
+			return { kind: "rejected", status: this.status };
+		}
 		if (!this.subagentSession) {
 			this.queueSteer(message);
-			return false;
+			return { kind: "buffered" };
 		}
 		await this.subagentSession.steer(message);
-		return true;
+		return { kind: "delivered" };
 	}
 
 	/** Return the session conversation as formatted text, or undefined if no session. */
@@ -192,12 +209,6 @@ export class Subagent {
 		this.workspaceBracket = new WorkspaceBracket(
 			this.execution.getWorkspaceProvider ?? (() => undefined),
 		);
-
-		// Notification state — created from parentSession.toolCallId if present
-		const toolCallId = init.execution.parentSession?.toolCallId;
-		if (toolCallId) {
-			this._notification = new NotificationState(toolCallId);
-		}
 	}
 
 	/**

package/src/observation/notification.ts

@@ -55,7 +55,7 @@ export function formatTaskNotification(record: Subagent, resultMaxLen: number):
       : record.result
     : "No output.";
 
-  const toolCallId = record.notification?.toolCallId;
+  const toolCallId = record.toolCallId;
   const outputFile = record.outputFile;
   return [
     "<task-notification>",
@@ -123,15 +123,21 @@ export function buildEventData(record: Subagent) {
 // ---- Notification system factory ----
 
 export interface NotificationSystem {
-  cancelNudge: (key: string) => void;
   sendCompletion: (record: Subagent) => void;
   dispose: () => void;
 }
 
+/** Delivery-consumption operation: get-result-tool's dependency on NotificationManager. */
+export interface ResultDelivery {
+  /** Record the parent consumed this agent's result: suppress its completion nudge. */
+  consume: (id: string) => void;
+}
+
 const NUDGE_HOLD_MS = 200;
 
-export class NotificationManager implements NotificationSystem {
+export class NotificationManager implements NotificationSystem, ResultDelivery {
   private pendingNudges = new Map<string, ReturnType<typeof setTimeout>>();
+  private consumed = new Set<string>();
 
   constructor(
     private sendMessage: (
@@ -140,21 +146,28 @@ export class NotificationManager implements NotificationSystem {
     ) => void,
   ) {}
 
-  cancelNudge(key: string): void {
-    const timer = this.pendingNudges.get(key);
-    if (timer != null) {
-      clearTimeout(timer);
-      this.pendingNudges.delete(key);
-    }
+  consume(id: string): void {
+    this.consumed.add(id);
+    this.cancelNudge(id);
   }
 
   sendCompletion(record: Subagent): void {
+    if (this.consumed.has(record.id)) return;
     this.scheduleNudge(record.id, () => this.emitIndividualNudge(record));
   }
 
   dispose(): void {
     for (const timer of this.pendingNudges.values()) clearTimeout(timer);
     this.pendingNudges.clear();
+    this.consumed.clear();
+  }
+
+  private cancelNudge(key: string): void {
+    const timer = this.pendingNudges.get(key);
+    if (timer != null) {
+      clearTimeout(timer);
+      this.pendingNudges.delete(key);
+    }
   }
 
   private scheduleNudge(key: string, send: () => void, delay = NUDGE_HOLD_MS): void {
@@ -173,7 +186,7 @@ export class NotificationManager implements NotificationSystem {
   }
 
   private emitIndividualNudge(record: Subagent): void {
-    if (record.notification?.resultConsumed) return;
+    if (this.consumed.has(record.id)) return;
 
     const notification = formatTaskNotification(record, 500);
     const outputFile = record.outputFile;

package/src/observation/subagent-events-observer.ts

@@ -64,9 +64,7 @@ export class SubagentEventsObserver implements SubagentManagerObserver {
 			completedAt: record.completedAt,
 		});
 
-		// Skip notification if result was already consumed via get_subagent_result.
-		if (record.notification?.resultConsumed) return;
-
+		// The manager decides whether to nudge (it owns the consumed-result state).
 		this.notifications.sendCompletion(record);
 	}
 

package/src/runtime.ts

@@ -6,6 +6,7 @@
  * Follows the same pattern as pi-permission-system's ExtensionRuntime.
  */
 
+import type { Model } from "@earendil-works/pi-ai";
 import { buildParentSnapshot, type ParentSnapshot } from "#src/lifecycle/parent-snapshot";
 import type { ModelInfo } from "#src/tools/spawn-config";
 import type { SessionContext } from "#src/types";
@@ -54,7 +55,7 @@ export class SubagentRuntime {
   /** Extract model info from the current session context. */
   getModelInfo(): ModelInfo {
     return {
-      parentModel: this.currentCtx?.model as ModelInfo["parentModel"],
+      parentModel: this.currentCtx?.model as Model<any> | undefined,
       modelRegistry: this.currentCtx?.modelRegistry,
     };
   }

package/src/service/service-adapter.ts

@@ -5,6 +5,7 @@
  * (stripping non-serializable fields), and session gating.
  */
 
+import type { Model } from "@earendil-works/pi-ai";
 import type { ParentSnapshot } from "#src/lifecycle/parent-snapshot";
 import type { WorkspaceProvider } from "#src/lifecycle/workspace";
 import type { SpawnOptions, SubagentRecord, SubagentsService } from "#src/service/service";
@@ -35,7 +36,7 @@ export interface ServiceRuntimeLike {
 export class SubagentsServiceAdapter implements SubagentsService {
   constructor(
     private readonly manager: SubagentManagerLike,
-    private readonly resolveModel: (input: string, registry: ModelRegistry) => unknown,
+    private readonly resolveModel: (input: string, registry: ModelRegistry) => Model<any> | string,
     private readonly runtime: ServiceRuntimeLike,
   ) {}
 
@@ -44,19 +45,7 @@ export class SubagentsServiceAdapter implements SubagentsService {
       throw new Error("No active session — cannot spawn agents outside a session.");
     }
 
-    let model: unknown;
-    if (options?.model) {
-      const registry = this.runtime.currentCtx.modelRegistry;
-      if (!registry) {
-        throw new Error("No model registry available.");
-      }
-      const resolved = this.resolveModel(options.model, registry);
-      if (typeof resolved === "string") {
-        throw new Error(resolved);
-      }
-      model = resolved;
-    }
-
+    const model = this.resolveModelOption(options?.model);
     const description = options?.description ?? prompt.slice(0, 80);
     const isBackground = !(options?.foreground ?? false);
 
@@ -87,11 +76,11 @@ export class SubagentsServiceAdapter implements SubagentsService {
 
   async steer(id: string, message: string): Promise<boolean> {
     const record = this.manager.getRecord(id);
-    if (record?.status !== "running") {
+    if (!record) {
       return false;
     }
-    await record.steer(message);
-    return true;
+    const outcome = await record.steer(message);
+    return outcome.kind !== "rejected";
   }
 
   async waitForAll(): Promise<void> {
@@ -105,6 +94,20 @@ export class SubagentsServiceAdapter implements SubagentsService {
   registerWorkspaceProvider(provider: WorkspaceProvider): () => void {
     return this.manager.registerWorkspaceProvider(provider);
   }
+
+  /** Resolve an optional model-string override against the current session's registry. */
+  private resolveModelOption(modelInput: string | undefined): Model<any> | undefined {
+    if (!modelInput) return undefined;
+    const registry = this.runtime.currentCtx?.modelRegistry;
+    if (!registry) {
+      throw new Error("No model registry available.");
+    }
+    const resolved = this.resolveModel(modelInput, registry);
+    if (typeof resolved === "string") {
+      throw new Error(resolved);
+    }
+    return resolved;
+  }
 }
 
 /**