@gotgenes/pi-subagents 18.0.0 → 18.0.2

package/docs/plans/0172-extract-turn-formatting.md

@@ -1,206 +0,0 @@
----
-issue: 172
-issue_title: "refactor(pi-subagents): extract shared turn-formatting logic"
----
-
-# Extract shared turn-formatting logic
-
-## Problem Statement
-
-Fallow identified 18 lines of duplicated production code between `lifecycle/agent-runner.ts` and `ui/message-formatters.ts` (originally `conversation-viewer.ts` before #170 extracted formatters).
-Both sites iterate over assistant message content items, extracting tool names and text parts for display.
-The `ToolCallContent` interface and `getToolCallName` helper are duplicated verbatim in both files.
-
-## Goals
-
-- Extract the duplicated `ToolCallContent` type and `getToolCallName` function into a single shared module.
-- Extract the content-iteration pattern (collecting text parts and tool names from assistant message content) into a reusable `extractAssistantContent` function.
-- Both consumers (`getAgentConversation` in `agent-runner.ts` and `formatAssistantMessage` in `message-formatters.ts`) import from the shared module.
-- Eliminate the fallow production-duplication finding.
-
-## Non-Goals
-
-- Moving `extractText` from `session/context.ts` to the new module — same concern (content parsing) but out of scope; note as a follow-up.
-- Refactoring `getAgentConversation` itself (it has no tests and its full-loop structure mixes user/assistant/toolResult formatting) — separate concern.
-- Changing `buildParentContext` in `session/context.ts`, which has a similar but simpler iteration pattern (no tool calls, different data source).
-
-## Background
-
-### Current duplication sites
-
-`lifecycle/agent-runner.ts` (private scope):
-
-```typescript
-interface ToolCallContent {
-  type: "toolCall";
-  name?: string;
-  toolName?: string;
-}
-
-function getToolCallName(c: { type: string }): string {
-  if (c.type !== "toolCall") return "unknown";
-  const tc = c as ToolCallContent;
-  return tc.name ?? tc.toolName ?? "unknown";
-}
-```
-
-`ui/message-formatters.ts` (exported):
-
-```typescript
-interface ToolCallContent { /* identical */ }
-export function getToolCallName(c: { type: string }): string { /* identical */ }
-```
-
-The content-iteration pattern appears in:
-
-1. `getAgentConversation` (agent-runner.ts lines 480–486) — plain-text output for LLM consumption
-2. `formatAssistantMessage` (message-formatters.ts lines 144–148) — themed display lines for TUI
-
-Both collect `textParts: string[]` and tool names via the same `for (const c of content)` loop with identical guards.
-
-### Structural analysis
-
-Per the code-design skill's "structural reasons before extracting duplication" check: the two consumers differ only in *presentation* (plain text vs. themed TUI lines).
-The *data extraction* — identifying text items and tool-call items, extracting tool names — is the same logical operation.
-This is incidental duplication suitable for extraction.
-
-### Dependencies
-
-- Issue #164 (domain directory reorganization): ✓ closed — files are already in `lifecycle/` and `ui/`.
-- Issue #170 (buildContentLines complexity reduction): ✓ closed — formatting logic is already in `message-formatters.ts`.
-- Issue #170 is related: the extraction may simplify `formatAssistantMessage` slightly (the issue body predicted this).
-
-### Placement
-
-The architecture doc (Phase 10, Step 9) says: "extracts into a shared function in the session domain."
-The `session/` directory already hosts `context.ts` which exports the related `extractText` function.
-A new `session/content-items.ts` module keeps the concern focused without overloading `context.ts`.
-
-## Design Overview
-
-### New module: `session/content-items.ts`
-
-```typescript
-/** Tool-call content item — SDK exposes this at runtime but doesn't export the type. */
-export interface ToolCallContent {
-  type: "toolCall";
-  name?: string;
-  toolName?: string;
-}
-
-/** Extracts the display name from a tool-call content item. */
-export function getToolCallName(c: { type: string }): string {
-  if (c.type !== "toolCall") return "unknown";
-  const tc = c as ToolCallContent;
-  return tc.name ?? tc.toolName ?? "unknown";
-}
-
-/** Extracted text parts and tool names from assistant message content. */
-export interface AssistantContentParts {
-  textParts: string[];
-  toolNames: string[];
-}
-
-/**
- * Extract text and tool-call names from assistant message content items.
- * Pure data extraction — consumers apply their own formatting.
- */
-export function extractAssistantContent(
-  content: { type: string; [key: string]: unknown }[],
-): AssistantContentParts {
-  const textParts: string[] = [];
-  const toolNames: string[] = [];
-  for (const c of content) {
-    if (c.type === "text" && c.text) textParts.push(c.text as string);
-    else if (c.type === "toolCall") toolNames.push(getToolCallName(c));
-  }
-  return { textParts, toolNames };
-}
-```
-
-### Consumer call sites (pseudocode)
-
-`getAgentConversation` (agent-runner.ts):
-
-```typescript
-const { textParts, toolNames } = extractAssistantContent(msg.content);
-if (textParts.length > 0) parts.push(`[Assistant]: ${textParts.join("\n")}`);
-if (toolNames.length > 0) parts.push(`[Tool Calls]:\n${toolNames.map(n => `  Tool: ${n}`).join("\n")}`);
-```
-
-`formatAssistantMessage` (message-formatters.ts):
-
-```typescript
-const { textParts, toolNames } = extractAssistantContent(content);
-const lines: string[] = [theme.bold("[Assistant]")];
-if (textParts.length > 0) lines.push(...wrapText(textParts.join("\n").trim(), width));
-for (const name of toolNames) lines.push(truncateToWidth(theme.fg("muted", `  [Tool: ${name}]`), width));
-```
-
-Both consumers call the same extraction, then apply their own presentation.
-This follows Tell-Don't-Ask: the shared function returns structured data, not formatted strings.
-
-## Module-Level Changes
-
-### New files
-
-| File                                 | Description                                                                              |
-| ------------------------------------ | ---------------------------------------------------------------------------------------- |
-| `src/session/content-items.ts`       | `ToolCallContent`, `getToolCallName`, `AssistantContentParts`, `extractAssistantContent` |
-| `test/session/content-items.test.ts` | Unit tests for `getToolCallName` and `extractAssistantContent`                           |
-
-### Changed files
-
-| File                                | Change                                                                                                                                                                 |
-| ----------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `src/lifecycle/agent-runner.ts`     | Remove `ToolCallContent` and `getToolCallName`; import `extractAssistantContent` from `session/content-items`; refactor `getAgentConversation` assistant branch        |
-| `src/ui/message-formatters.ts`      | Remove `ToolCallContent` and `getToolCallName`; import `extractAssistantContent` and `getToolCallName` from `session/content-items`; refactor `formatAssistantMessage` |
-| `docs/architecture/architecture.md` | Update Step 9 status to "✓ Done"; update `session/` module listing to include `content-items.ts`; update production duplication section                                |
-
-### Architecture doc updates
-
-The architecture doc references this issue in three places:
-
-1. Line 487 — "Production duplication" subsection: update to note the duplication is resolved.
-2. Line 247 — `session/` module listing: add `content-items.ts`.
-3. Line 653 — Phase 10, Step 9: mark "✓ Done".
-
-## Test Impact Analysis
-
-1. **New unit tests enabled**: `getToolCallName` and `extractAssistantContent` get direct unit tests for the first time.
-   Previously, `getToolCallName` was only exercised indirectly through `formatAssistantMessage` tests.
-2. **Existing tests that stay as-is**: `message-formatters.test.ts` tests for `formatAssistantMessage` continue to exercise the full pipeline (extraction + formatting).
-   They become integration-level tests relative to the new extraction layer — they should not be simplified or removed.
-3. **No existing tests become redundant**: `getAgentConversation` has no tests today, so nothing to deduplicate.
-
-## TDD Order
-
-1. `test:` Write tests for `getToolCallName` in `test/session/content-items.test.ts` — covers `toolCall` with `name`, `toolName`, both (prefers `name`), neither (returns "unknown"), and non-toolCall type.
-   Commit: `test: add getToolCallName unit tests`
-2. `test:` Write tests for `extractAssistantContent` in the same file — covers empty array, text-only items, toolCall-only items, mixed items, and items with other types (e.g., `image`).
-   Commit: `test: add extractAssistantContent unit tests`
-3. `feat:` Create `src/session/content-items.ts` with `ToolCallContent`, `getToolCallName`, `AssistantContentParts`, and `extractAssistantContent`.
-   All tests go green.
-   Commit: `feat: extract shared content-item parsing into session/content-items`
-4. `refactor:` Update `message-formatters.ts` — remove local `ToolCallContent` and `getToolCallName`; import `getToolCallName` and `extractAssistantContent` from `session/content-items`; refactor `formatAssistantMessage` to use `extractAssistantContent`.
-   Existing `message-formatters.test.ts` stays green.
-   Commit: `refactor: use shared content-items in message-formatters`
-5. `refactor:` Update `agent-runner.ts` — remove local `ToolCallContent` and `getToolCallName`; import `extractAssistantContent` from `session/content-items`; refactor `getAgentConversation` assistant branch.
-   Existing `agent-runner.test.ts` stays green.
-   Commit: `refactor: use shared content-items in agent-runner`
-6. `docs:` Update `docs/architecture/architecture.md` — mark Step 9 done, update module listing, update duplication section.
-   Commit: `docs: mark step 9 (extract turn-formatting) done in architecture`
-
-## Risks and Mitigations
-
-| Risk                                                                                                              | Mitigation                                                                                                     |
-| ----------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- |
-| `getToolCallName` signature change breaking callers                                                               | Signature is identical — no change needed. Import path changes are the only difference.                        |
-| `extractAssistantContent` return shape mismatch with consumers                                                    | Consumer pseudocode verified above; both sites restructure trivially around `{ textParts, toolNames }`.        |
-| `message-formatters.ts` re-exports `getToolCallName` — removing the local definition could break external imports | Grep confirms no external consumers import `getToolCallName` from `message-formatters.ts`. Drop the re-export. |
-| `isBashExecution` in `message-formatters.ts` also uses a local type — could be confused with this extraction      | `isBashExecution` and `BashExecutionMessage` are unrelated to the tool-call duplication; leave them in place.  |
-
-## Open Questions
-
-- Should `extractText` (currently in `session/context.ts`) move to `session/content-items.ts` for consistency?
-  Deferred — it works fine where it is, and moving it means updating all importers for no functional benefit.

package/docs/plans/0180-reorder-append-prompt-for-kv-cache.md

@@ -1,100 +0,0 @@
----
-issue: 180
-issue_title: "perf(pi-subagents): reorder append-mode system prompt to enable KV cache reuse"
----
-
-# Reorder append-mode system prompt for KV cache reuse
-
-## Problem Statement
-
-In append mode, `buildAgentPrompt()` places varying, agent-specific content (the `<active_agent>` tag and env block) *before* the large shared inherited system prompt (~8k tokens).
-LLM KV caching works on prefixes — the cache is only reusable when the beginning of the prompt matches.
-Every subagent spawn reprocesses the entire inherited prompt from scratch because the prefix differs per agent.
-
-## Goals
-
-- Reorder the append-mode system prompt so shared/stable content comes first and varying content follows.
-- Preserve the `<active_agent>` tag at any position — pi-permission-system's `ACTIVE_AGENT_TAG_REGEX.exec()` searches the full string.
-- Keep replace-mode prompt ordering unchanged (it has no shared inherited content to cache).
-- Update tests and JSDoc to reflect the new ordering.
-
-## Non-Goals
-
-- Changing replace-mode prompt assembly (no shared prefix to cache).
-- Modifying pi-permission-system (its regex parsing is already position-independent).
-- Changing the *content* of any prompt section — only reordering.
-
-## Background
-
-`buildAgentPrompt()` in `src/session/prompts.ts` assembles the system prompt for subagents.
-In append mode, the current ordering is:
-
-```text
-1. <active_agent name="${name}"/>     ← VARIES per agent
-2. # Environment ...                  ← VARIES per runtime
-3. <inherited_system_prompt>          ← SHARED (~8k tokens)
-4. <sub_agent_context>                ← SHARED (static)
-5. <agent_instructions>               ← VARIES per agent
-6. memory / skills                    ← VARIES
-```
-
-pi-permission-system's `getActiveAgentNameFromSystemPrompt()` in `src/active-agent.ts` uses `ACTIVE_AGENT_TAG_REGEX.exec(systemPrompt)` — a regex search that finds the tag at any position, confirmed by reading the source.
-
-## Design Overview
-
-Move shared/stable sections to the front of the append-mode prompt:
-
-```text
-1. <inherited_system_prompt>          ← SHARED (~8k tokens, NOW CACHEABLE)
-2. <sub_agent_context>                ← SHARED (static)
-3. <active_agent name="${name}"/>     ← VARIES (after cached prefix)
-4. # Environment ...                  ← VARIES
-5. <agent_instructions>               ← VARIES per agent
-6. memory / skills                    ← VARIES
-```
-
-This is a pure reordering — no content changes.
-The `<active_agent>` tag remains in the system prompt for pi-permission-system to find via regex.
-The env block and agent instructions still provide context to the model; their position relative to the inherited prompt is not semantically significant.
-
-## Module-Level Changes
-
-### `src/session/prompts.ts`
-
-1. Reorder the return statement in the `config.promptMode === "append"` branch to place `identity` (wrapped in `<inherited_system_prompt>`) and `bridge` before `activeAgentTag` and `envBlock`.
-2. Update the JSDoc comment on `buildAgentPrompt()` — replace "Both modes prepend" language with a description that notes the tag is included (not necessarily prepended) in append mode.
-
-### `test/session/prompts.test.ts`
-
-1. Update "prepends `<active_agent>` tag in append mode" — change from asserting `prompt.startsWith()` to asserting the tag appears *after* the inherited system prompt.
-2. Update "active_agent tag appears before envBlock in both modes" — the append-mode assertions change: the tag should still appear before the env block, but no longer at index 0.
-   The replace-mode assertions remain unchanged (`tagIdx === 0`).
-
-## Test Impact Analysis
-
-- Two existing tests assert `<active_agent>` is prepended (index 0) in append mode — these must change to assert the new ordering.
-- All other prompt tests use `toContain()` and are position-independent — they pass without changes.
-- No new test files or test surfaces are needed; the existing test suite covers the reordering adequately once the positional assertions are updated.
-
-## TDD Order
-
-1. **Red: update positional assertions for append mode.**
-   Change the two append-mode tests to assert the new ordering: `<inherited_system_prompt>` appears before `<active_agent>`, and the tag appears before the env block but not at index 0.
-   Commit: `test: assert cache-friendly prompt ordering in append mode (#180)`
-
-2. **Green: reorder the append-mode return statement.**
-   Move `identity` + `<inherited_system_prompt>` wrapper and `bridge` before `activeAgentTag` + `envBlock` in the return expression.
-   Update the JSDoc on `buildAgentPrompt()`.
-   Commit: `perf: reorder append-mode prompt for KV cache reuse (#180)`
-
-## Risks and Mitigations
-
-| Risk                                         | Mitigation                                                                                                                                         |
-| -------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- |
-| pi-permission-system depends on tag position | Confirmed `ACTIVE_AGENT_TAG_REGEX.exec()` searches the full string — position-independent.                                                         |
-| Model behavior changes with reordered prompt | The same content is present; only ordering changes. The inherited system prompt as the "base" followed by specialization is arguably more natural. |
-| Replace mode accidentally affected           | Replace mode has its own code path and is not touched by this change.                                                                              |
-
-## Open Questions
-
-None — the design is straightforward and confirmed safe by code inspection.

package/docs/plans/0185-remove-persistent-agent-memory.md

@@ -1,161 +0,0 @@
----
-issue: 185
-issue_title: "pi-subagents: Remove persistent agent memory feature"
----
-
-# Remove persistent agent memory feature
-
-## Problem Statement
-
-The `memory.ts` module and all supporting code for persistent agent memory (`MEMORY.md`, `agent-memory/` directories) should be removed from `pi-subagents`.
-The memory feature invents its own filesystem layout, file format, and security model — all outside the stated scope of `pi-subagents`, which is agent spawning, execution, and result retrieval.
-This follows the same scope-reduction rationale as the scheduling subsystem removal (issue #52).
-
-## Goals
-
-- Remove the `memory.ts` module and all memory-related code from the package.
-- Remove `MemoryScope` type and `memory` field from `AgentConfig`.
-- Remove memory block injection from session assembly.
-- Remove memory tool augmentation from agent-types.
-- Remove memory parsing from custom agent loading.
-- Remove memory display from UI components.
-- Extract `isSymlink`, `isUnsafeName`, and `safeReadFile` to a shared utility module — `skill-loader.ts` depends on them independently of memory.
-- Update architecture documentation to reflect the removal.
-
-## Non-Goals
-
-- Replacing memory with an alternative persistence mechanism — that is out of scope.
-- Changing the `skill-loader.ts` logic beyond updating the import source for the extracted utilities.
-
-## Background
-
-The memory system spans six source modules and two UI modules:
-
-| File                              | Memory surface                                                                                                                                                                           |
-| --------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `src/session/memory.ts`           | Core module: `buildMemoryBlock`, `buildReadOnlyMemoryBlock`, `resolveMemoryDir`, `ensureMemoryDir`, `readMemoryIndex`, plus shared utilities `isSymlink`, `isUnsafeName`, `safeReadFile` |
-| `src/types.ts`                    | `MemoryScope` type, `memory?: MemoryScope` field on `AgentConfig`                                                                                                                        |
-| `src/session/session-config.ts`   | `AssemblerIO.buildMemoryBlock` / `buildReadOnlyMemoryBlock` fields, memory logic block (~25 lines), `MemoryScope` import                                                                 |
-| `src/config/agent-types.ts`       | `getMemoryToolNames()`, `getReadOnlyMemoryToolNames()`, `MEMORY_TOOL_NAMES`, `READONLY_MEMORY_TOOL_NAMES`                                                                                |
-| `src/config/custom-agents.ts`     | `parseMemory()` function, `memory: parseMemory(fm.memory)` assignment                                                                                                                    |
-| `src/session/prompts.ts`          | `memoryBlock` field on `PromptExtras`, injection into system prompt                                                                                                                      |
-| `src/ui/agent-config-editor.ts`   | `if (cfg.memory) fmFields.push(...)`                                                                                                                                                     |
-| `src/ui/agent-creation-wizard.ts` | `memory:` line in frontmatter help text                                                                                                                                                  |
-| `src/index.ts`                    | Imports `buildMemoryBlock`, `buildReadOnlyMemoryBlock`; wires them into `assemblerIO`                                                                                                    |
-
-The three utility functions (`isSymlink`, `isUnsafeName`, `safeReadFile`) are imported by `skill-loader.ts` for filesystem safety — they are not memory-specific and must survive the removal.
-
-AGENTS.md constraint: the architecture doc in `docs/architecture/architecture.md` references `memory.ts` in the module listing and domain model diagram — both need updating.
-
-## Design Overview
-
-This is a pure removal with one extraction.
-The only design decision is where to place the extracted utilities.
-
-The three functions (`isSymlink`, `isUnsafeName`, `safeReadFile`) are filesystem safety primitives used by `skill-loader.ts`.
-They belong in a new `src/session/safe-fs.ts` module in the session domain, co-located with their sole remaining consumer.
-
-The removal works consumers-first, declaration-last:
-
-1. Extract shared utilities to `safe-fs.ts` (additive, no behavior change).
-2. Remove all memory consumers — session-config, prompts, agent-types, custom-agents, UI, index wiring.
-3. Remove the declarations — `MemoryScope`, `AgentConfig.memory`, `memory.ts`.
-4. Update architecture docs.
-
-## Module-Level Changes
-
-### New files
-
-- `src/session/safe-fs.ts` — extracted `isSymlink`, `isUnsafeName`, `safeReadFile`.
-- `test/session/safe-fs.test.ts` — tests moved from `memory.test.ts` for these three functions.
-
-### Modified files
-
-- `src/session/skill-loader.ts` — change import from `#src/session/memory` to `#src/session/safe-fs`.
-- `src/session/session-config.ts` — remove `buildMemoryBlock` / `buildReadOnlyMemoryBlock` from `AssemblerIO`; remove `MemoryScope` import; remove entire memory logic block (~lines 215–242).
-- `src/session/prompts.ts` — remove `memoryBlock` from `PromptExtras` interface; remove `extras?.memoryBlock` injection.
-- `src/config/agent-types.ts` — remove `MEMORY_TOOL_NAMES`, `READONLY_MEMORY_TOOL_NAMES`, `getMemoryToolNames()`, `getReadOnlyMemoryToolNames()`.
-- `src/config/custom-agents.ts` — remove `MemoryScope` import; remove `memory: parseMemory(fm.memory)` assignment; remove `parseMemory()` function.
-- `src/types.ts` — remove `MemoryScope` type; remove `memory?: MemoryScope` from `AgentConfig`; remove associated doc comment.
-- `src/ui/agent-config-editor.ts` — remove `if (cfg.memory)` line.
-- `src/ui/agent-creation-wizard.ts` — remove `memory:` line from frontmatter help text.
-- `src/index.ts` — remove `buildMemoryBlock` / `buildReadOnlyMemoryBlock` import; remove those fields from `assemblerIO` object.
-- `docs/architecture/architecture.md` — remove `memory.ts` from module listing; remove Memory node from domain model Mermaid diagram; update session domain description.
-
-### Deleted files
-
-- `src/session/memory.ts` — entire module.
-- `test/session/memory.test.ts` — entire test file (tests for `isSymlink`, `isUnsafeName`, `safeReadFile` are moved to `safe-fs.test.ts` first; remaining memory-specific tests are deleted).
-
-### Test files modified
-
-- `test/session/session-config.test.ts` — remove `mockBuildMemoryBlock` / `mockBuildReadOnlyMemoryBlock` mocks from `AssemblerIO` construction; remove "assembleSessionConfig — memory block selection" describe block (~lines 354–427).
-- `test/session/prompts.test.ts` — remove "injects memory block in replace mode", "injects memory block in append mode", and "injects both memory and skills" test cases.
-- `test/config/agent-types.test.ts` — remove `getMemoryToolNames` / `getReadOnlyMemoryToolNames` imports and test suite (~lines 26–51).
-- `test/config/custom-agents.test.ts` — remove memory scope parsing tests (~lines 361–403).
-
-## Test Impact Analysis
-
-1. The extraction of `isSymlink`, `isUnsafeName`, `safeReadFile` to `safe-fs.ts` enables their tests to exist independently of the memory module — currently they are co-located with memory-specific tests in `memory.test.ts`.
-2. The memory-specific tests in `memory.test.ts` (`resolveMemoryDir`, `ensureMemoryDir`, `readMemoryIndex`, `buildMemoryBlock`, `buildReadOnlyMemoryBlock`) become redundant and are deleted — the code they test is being removed.
-3. The memory block selection tests in `session-config.test.ts` (~70 lines) test the memory branching logic in `assembleSessionConfig` — they are deleted because that logic is removed.
-4. The memory injection tests in `prompts.test.ts` test `memoryBlock` injection — deleted because the field and injection code are removed.
-5. The memory parsing tests in `custom-agents.test.ts` test `parseMemory` — deleted because the function is removed.
-6. The memory tool name helper tests in `agent-types.test.ts` test `getMemoryToolNames` / `getReadOnlyMemoryToolNames` — deleted because the functions are removed.
-7. All other tests remain as-is — they do not depend on memory functionality.
-
-## TDD Order
-
-1. **Extract utilities to `safe-fs.ts`.**
-   Create `src/session/safe-fs.ts` with `isSymlink`, `isUnsafeName`, `safeReadFile`.
-   Create `test/session/safe-fs.test.ts` with tests moved from `memory.test.ts` for these three functions.
-   Update `src/session/skill-loader.ts` import to point to `#src/session/safe-fs`.
-   Update `src/session/memory.ts` import to point to `#src/session/safe-fs` (temporary — keeps memory working until removal).
-   Verify: `pnpm vitest run` and `pnpm run check`.
-   Commit: `refactor: extract safe-fs utilities from memory module`
-
-2. **Remove memory from session assembly and config layers.**
-   Remove `buildMemoryBlock` / `buildReadOnlyMemoryBlock` from `AssemblerIO` in `session-config.ts`.
-   Remove `MemoryScope` import and all memory logic from `session-config.ts`.
-   Remove `memoryBlock` from `PromptExtras` in `prompts.ts` and its injection logic.
-   Remove `getMemoryToolNames`, `getReadOnlyMemoryToolNames`, `MEMORY_TOOL_NAMES`, `READONLY_MEMORY_TOOL_NAMES` from `agent-types.ts`.
-   Remove `getMemoryToolNames` / `getReadOnlyMemoryToolNames` import from `session-config.ts`.
-   Remove `buildMemoryBlock` / `buildReadOnlyMemoryBlock` import and `assemblerIO` fields from `index.ts`.
-   Update `test/session/session-config.test.ts`: remove memory mocks from IO construction and memory block selection test suite.
-   Update `test/session/prompts.test.ts`: remove memory injection tests.
-   Update `test/config/agent-types.test.ts`: remove memory tool name helper tests.
-   Verify: `pnpm vitest run` and `pnpm run check`.
-   Commit: `feat!: remove memory from session assembly and config layers`
-
-3. **Remove memory from types, custom-agents, and UI.**
-   Remove `MemoryScope` type and `memory` field from `AgentConfig` in `types.ts`.
-   Remove `parseMemory()` function, `MemoryScope` import, and `memory:` assignment from `custom-agents.ts`.
-   Remove `if (cfg.memory)` line from `agent-config-editor.ts`.
-   Remove `memory:` help text line from `agent-creation-wizard.ts`.
-   Update `test/config/custom-agents.test.ts`: remove memory parsing tests.
-   Verify: `pnpm vitest run` and `pnpm run check`.
-   Commit: `feat!: remove MemoryScope type and memory config field`
-
-4. **Delete `memory.ts` and its test file.**
-   Delete `src/session/memory.ts`.
-   Delete `test/session/memory.test.ts`.
-   Verify: `pnpm vitest run` and `pnpm run check`.
-   Commit: `feat!: delete memory module`
-
-5. **Update architecture documentation.**
-   Remove Memory node from the domain model Mermaid diagram.
-   Remove `memory.ts` from the module listing.
-   Update session domain description.
-   Commit: `docs: update architecture after memory removal`
-
-## Risks and Mitigations
-
-| Risk                                                                                   | Mitigation                                                                                                                                                              |
-| -------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| Users with `memory:` in custom agent frontmatter — the field silently becomes a no-op. | `parseMemory` removal means unknown frontmatter keys are ignored by the markdown parser. No error, no crash — just no memory. This is acceptable for a feature removal. |
-| `skill-loader.ts` breaks if the utility extraction has a typo or missing re-export.    | Step 1 runs full test suite before proceeding. The skill-loader tests exercise `isUnsafeName` and `safeReadFile` indirectly.                                            |
-| Architecture doc Mermaid diagram breaks after node removal.                            | Verify the diagram renders correctly after editing — remove both the node and all edges referencing it.                                                                 |
-
-## Open Questions
-
-None — the issue scope is unambiguous and all affected files have been traced.

package/docs/plans/0188-replace-any-casts-with-sdk-types.md

@@ -1,162 +0,0 @@
----
-issue: 188
-issue_title: "refactor(pi-subagents): replace any casts with SDK types in extractText and SubscribableSession"
----
-
-# Replace `any` casts with SDK types
-
-## Problem Statement
-
-Two places in pi-subagents use `any` where proper SDK types are available and already imported in adjacent files.
-`extractText` in `session/context.ts` uses `(c: any)` in a filter/map chain, requiring a top-level `eslint-disable` for `no-unsafe-member-access` and `no-unsafe-return`.
-`record-observer.ts` and `ui-observer.ts` each define an identical local `SubscribableSession` interface with `(event: any) => void`, creating both a type hole and duplicated boilerplate.
-
-## Goals
-
-- Replace `any` casts in `extractText` with a `TextContent` type predicate.
-- Remove the `eslint-disable` comment from `session/context.ts`.
-- Replace `any` in the `SubscribableSession` interface with `AgentSessionEvent`.
-- Deduplicate the `SubscribableSession` interface into a single shared definition.
-
-## Non-Goals
-
-- Changing the `extractText` parameter type from `unknown[]` — callers in `message-formatters.ts` pass `unknown[]`, and widening the refactoring surface is out of scope.
-- Replacing the `SubscribableSession` interface with the full `AgentSession` class — ISP requires a narrow interface (the observers only need `subscribe`).
-- Addressing the `eslint-disable` in `record-observer.ts` and `ui-observer.ts` for `no-unsafe-member-access` / `no-unsafe-assignment` — those are caused by the `event` property access pattern inside the callback body, not by the parameter type.
-  Once the callback parameter is typed as `AgentSessionEvent`, the unsafe-access rules should be satisfied and those `eslint-disable` comments can be removed too.
-
-## Background
-
-### Existing conventions
-
-`content-items.ts` already imports `TextContent` from `@earendil-works/pi-ai` and uses `(c as TextContent).text` after a `c.type === "text"` guard.
-`agent-runner.ts` already imports `AgentSessionEvent` from `@earendil-works/pi-coding-agent` and uses it as the parameter type in `session.subscribe((event: AgentSessionEvent) => { ... })`.
-Both SDK types are proven to work in this package.
-
-### `extractText` callers
-
-`extractText(content: unknown[])` is called from:
-
-- `session/context.ts` — `buildParentContext` passes `msg.content` from session entries.
-- `lifecycle/agent-runner.ts` — `getLastAssistantText` and `getAgentConversation` pass `msg.content`.
-- `ui/message-formatters.ts` — `formatUserMessage` and `formatToolResult` pass `unknown[]` content.
-
-The parameter type stays `unknown[]` to avoid rippling through callers.
-The type predicate narrows inside the function body.
-
-### `SubscribableSession` consumers
-
-Both `subscribeRecordObserver` and `subscribeUIObserver` accept a `SubscribableSession` parameter.
-Tests use `createMockSession()` from `test/helpers/mock-session.ts`, which returns a `MockSession` with `subscribe: Mock<(fn: (event: unknown) => void) => () => void>`.
-
-Changing `SubscribableSession.subscribe` to accept `(event: AgentSessionEvent) => void` is structurally sound: the mock's `subscribe` accepting `(fn: (event: unknown) => void)` is a supertype — a function that accepts any event can accept an `AgentSessionEvent`.
-The TypeScript compiler allows this because of function parameter contravariance.
-Tests construct inline event objects that match `AgentSessionEvent` member shapes, so no test changes are needed.
-
-### Shared location for `SubscribableSession`
-
-The interface is used by two domains (observation, UI).
-A new shared types location is needed.
-The existing `types.ts` at the package root contains cross-cutting types (`SubagentType`, `ThinkingLevel`, `ShellExec`).
-`SubscribableSession` fits there — it's a narrow cross-domain interface for session event subscription.
-
-## Design Overview
-
-### `extractText` type predicate
-
-Replace the `any` casts with a user-defined type guard:
-
-```typescript
-import type { TextContent } from "@earendil-works/pi-ai";
-
-function isTextContent(c: unknown): c is TextContent {
-  return typeof c === "object" && c !== null && (c as { type: string }).type === "text";
-}
-
-export function extractText(content: unknown[]): string {
-  return content
-    .filter(isTextContent)
-    .map((c) => c.text ?? "")
-    .join("\n");
-}
-```
-
-The type predicate eliminates both `any` casts and the `eslint-disable` at the top of the file.
-
-### `SubscribableSession` with `AgentSessionEvent`
-
-Move the interface to `types.ts` and type the callback:
-
-```typescript
-import type { AgentSessionEvent } from "@earendil-works/pi-coding-agent";
-
-export interface SubscribableSession {
-  subscribe(fn: (event: AgentSessionEvent) => void): () => void;
-}
-```
-
-Both observer files import from `types.ts` instead of defining their own.
-
-### Event property access in observer callbacks
-
-Once the callback parameter is typed as `AgentSessionEvent`, TypeScript knows the event's discriminated union members.
-The `event.type` checks narrow the union, so `event.toolName`, `event.message`, etc. become type-safe.
-The `eslint-disable` comments for `no-unsafe-member-access` and `no-unsafe-assignment` can be removed from both observer files.
-
-## Module-Level Changes
-
-| File                                 | Change                                                                                                                                                        |
-| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `src/session/context.ts`             | Import `TextContent`; add `isTextContent` type predicate; replace `any` filter/map; remove top-level `eslint-disable`                                         |
-| `src/types.ts`                       | Add `SubscribableSession` interface with `AgentSessionEvent` callback type; add `AgentSessionEvent` import                                                    |
-| `src/observation/record-observer.ts` | Import `SubscribableSession` from `types.ts`; remove local interface; remove top-level `eslint-disable`; remove inline `any` annotation on callback parameter |
-| `src/ui/ui-observer.ts`              | Import `SubscribableSession` from `types.ts`; remove local interface; remove top-level `eslint-disable`; remove inline `any` annotation on callback parameter |
-
-No test file changes expected — the mock session's structural typing remains compatible.
-
-## Test Impact Analysis
-
-1. No new unit tests are needed — the refactoring is type-only (no behavioral change).
-2. No existing tests become redundant.
-3. All existing tests for `subscribeRecordObserver` and `subscribeUIObserver` must pass as-is — they verify the same event-handling behavior.
-
-## TDD Order
-
-This is a pure refactoring with no behavioral change.
-Each step should pass `pnpm run check` (type-check) and `pnpm vitest run` (tests) before committing.
-
-1. **Add `isTextContent` type predicate and remove `any` from `extractText`.**
-   Import `TextContent` from `@earendil-works/pi-ai`.
-   Add `isTextContent` predicate function.
-   Replace the `any`-cast filter/map chain with the predicate.
-   Remove the top-level `eslint-disable` comment.
-   Verify: `pnpm run check`, `pnpm vitest run`.
-   Commit: `refactor: replace any casts in extractText with TextContent type predicate (#188)`
-
-2. **Move `SubscribableSession` to `types.ts` with `AgentSessionEvent`.**
-   Add `AgentSessionEvent` import and `SubscribableSession` interface to `src/types.ts`.
-   Update `record-observer.ts`: import from `types.ts`, remove local interface, remove `eslint-disable`, remove `any` from callback parameter.
-   Update `ui-observer.ts`: import from `types.ts`, remove local interface, remove `eslint-disable`, remove `any` from callback parameter.
-   Verify: `pnpm run check`, `pnpm vitest run`.
-   Commit: `refactor: replace any in SubscribableSession with AgentSessionEvent (#188)`
-
-## Risks and Mitigations
-
-1. **`AgentSessionEvent` union may not cover all event shapes accessed in observers.**
-   Mitigation: `agent-runner.ts` already uses the same type for identical event patterns (`event.type`, `event.toolName`, `event.message`).
-   The type checker will flag any property access that the union doesn't support.
-   Run `pnpm run check` after each step.
-
-2. **Mock session type incompatibility.**
-   The mock's `subscribe` accepts `(fn: (event: unknown) => void)`.
-   A `SubscribableSession` with `(fn: (event: AgentSessionEvent) => void)` is structurally compatible via contravariance.
-   If the compiler disagrees, the mitigation is to update `MockSession.subscribe` to accept `(fn: (event: AgentSessionEvent) => void)` — a one-line change.
-
-3. **`TextContent.text` is non-optional in the SDK type.**
-   The current code uses `c.text ?? ""` which implies `text` could be undefined.
-   `TextContent` defines `text: string` (required), so the nullish coalescing is harmless but unnecessary.
-   Keep it for safety — removing it is a separate cleanup.
-
-## Open Questions
-
-None — the issue's proposed approach is unambiguous and the SDK types are already validated in adjacent files.