@oh-my-pi/pi-coding-agent 14.7.8 → 14.8.0

package/src/internal-urls/docs-index.generated.ts

@@ -1,6 +1,6 @@
 // Auto-generated by scripts/generate-docs-index.ts - DO NOT EDIT
 
-export const EMBEDDED_DOC_FILENAMES: readonly string[] = ["bash-tool-runtime.md","blob-artifact-architecture.md","compaction.md","config-usage.md","custom-tools.md","environment-variables.md","extension-loading.md","extensions.md","fs-scan-cache-architecture.md","gemini-manifest-extensions.md","handoff-generation-pipeline.md","hooks.md","marketplace.md","mcp-config.md","mcp-protocol-transports.md","mcp-runtime-lifecycle.md","mcp-server-tool-authoring.md","memory.md","models.md","natives-addon-loader-runtime.md","natives-architecture.md","natives-binding-contract.md","natives-build-release-debugging.md","natives-media-system-utils.md","natives-rust-task-cancellation.md","natives-shell-pty-process.md","natives-text-search-pipeline.md","non-compaction-retry-policy.md","notebook-tool-runtime.md","plugin-manager-installer-plumbing.md","porting-from-pi-mono.md","porting-to-natives.md","provider-streaming-internals.md","python-repl.md","resolve-tool-runtime.md","rpc.md","rulebook-matching-pipeline.md","sdk.md","secrets.md","session-operations-export-share-fork-resume.md","session-switching-and-recent-listing.md","session-tree-plan.md","session.md","skills.md","skills/authoring-extensions.md","skills/authoring-hooks.md","skills/authoring-marketplaces.md","skills/examples/hello-extension/README.md","skills/examples/mini-marketplace/README.md","skills/examples/safety-hook/README.md","slash-command-internals.md","task-agent-discovery.md","theme.md","tree.md","ttsr-injection-lifecycle.md","tui-runtime-internals.md","tui.md"];
+export const EMBEDDED_DOC_FILENAMES: readonly string[] = ["bash-tool-runtime.md","blob-artifact-architecture.md","compaction.md","config-usage.md","custom-tools.md","environment-variables.md","extension-loading.md","extensions.md","fs-scan-cache-architecture.md","gemini-manifest-extensions.md","handoff-generation-pipeline.md","hooks.md","marketplace.md","mcp-config.md","mcp-protocol-transports.md","mcp-runtime-lifecycle.md","mcp-server-tool-authoring.md","memory.md","models.md","natives-addon-loader-runtime.md","natives-architecture.md","natives-binding-contract.md","natives-build-release-debugging.md","natives-media-system-utils.md","natives-rust-task-cancellation.md","natives-shell-pty-process.md","natives-text-search-pipeline.md","non-compaction-retry-policy.md","notebook-tool-runtime.md","plugin-manager-installer-plumbing.md","porting-from-pi-mono.md","porting-to-natives.md","provider-streaming-internals.md","python-repl.md","render-mermaid.md","resolve-tool-runtime.md","rpc.md","rulebook-matching-pipeline.md","sdk.md","secrets.md","session-operations-export-share-fork-resume.md","session-switching-and-recent-listing.md","session-tree-plan.md","session.md","skills.md","skills/authoring-extensions.md","skills/authoring-hooks.md","skills/authoring-marketplaces.md","skills/examples/hello-extension/README.md","skills/examples/mini-marketplace/README.md","skills/examples/safety-hook/README.md","slash-command-internals.md","task-agent-discovery.md","theme.md","tree.md","ttsr-injection-lifecycle.md","tui-runtime-internals.md","tui.md"];
 
 export const EMBEDDED_DOCS: Readonly<Record<string, string>> = {
 	"bash-tool-runtime.md": "# Bash tool runtime\n\nThis document describes the **`bash` tool** runtime path used by agent tool calls, from command normalization to execution, truncation/artifacts, and rendering.\n\nIt also calls out where behavior diverges in interactive TUI, print mode, RPC mode, and user-initiated bang (`!`) shell execution.\n\n## Scope and runtime surfaces\n\nThere are two different bash execution surfaces in coding-agent:\n\n1. **Tool-call surface** (`toolName: \"bash\"`): used when the model calls the bash tool.\n   - Entry point: `BashTool.execute()`.\n   - Parameters include `command`, optional `env`, `timeout`, `cwd`, `head`, `tail`, `pty`, and, when `async.enabled` is true, `async`.\n2. **User bang-command surface** (`!cmd` from interactive input or RPC `bash` command): session-level helper path.\n   - Entry point: `AgentSession.executeBash()`.\n\nBoth eventually use `executeBash()` in `src/exec/bash-executor.ts` for non-PTY execution, but only the tool-call path runs normalization/interception, optional managed background-job handling, and tool renderer logic.\n\n## End-to-end tool-call pipeline\n\n## 1) Input handling and parameter merge\n\n`BashTool.execute()` currently handles input before execution as follows:\n\n- validates optional `env` names against shell-variable syntax,\n- extracts a leading `cd <path> && ...` into `cwd` when `cwd` was not supplied,\n- rejects `async: true` when `async.enabled` is false,\n- uses only explicit `head`/`tail` tool args for post-run filtering.\n\n`normalizeBashCommand()` still exists in `src/tools/bash-normalize.ts`, but `BashTool.execute()` does not call it in the current source. Trailing shell pipes such as `| head -n 50` remain part of the shell command unless the caller uses the structured `head`/`tail` args.\n\n## 2) Optional interception (blocked-command path)\n\nIf `bashInterceptor.enabled` is true, `BashTool` loads rules from settings and runs `checkBashInterception()` against the normalized command.\n\nInterception behavior:\n\n- command is blocked **only** when:\n  - regex rule matches, and\n  - the suggested tool is present in `ctx.toolNames`.\n- invalid regex rules are silently skipped.\n- on block, `BashTool` throws `ToolError` with message:\n  - `Blocked: ...`\n  - original command included.\n\nDefault rule patterns (defined in code) target common misuses:\n\n- file readers (`cat`, `head`, `tail`, ...)\n- search tools (`grep`, `rg`, ...)\n- file finders (`find`, `fd`, ...)\n- in-place editors (`sed -i`, `perl -i`, `awk -i inplace`)\n- shell redirection writes (`echo ... > file`, heredoc redirection)\n\n### Caveat\n\n`InterceptionResult` includes `suggestedTool`, but `BashTool` currently surfaces only the message text (no structured suggested-tool field in `details`).\n\n## 3) CWD validation and timeout clamping\n\n`cwd` is resolved relative to session cwd (`resolveToCwd`), then validated via `stat`:\n\n- missing path -> `ToolError(\"Working directory does not exist: ...\")`\n- non-directory -> `ToolError(\"Working directory is not a directory: ...\")`\n\nTimeout is clamped to `[1, 3600]` seconds and converted to milliseconds.\n\n## 4) Artifact allocation\n\nBefore execution, the tool allocates an artifact path/id (best-effort) for truncated output storage.\n\n- artifact allocation failure is non-fatal (execution continues without artifact spill file),\n- artifact id/path are passed into execution path for full-output persistence on truncation.\n\n## 5) PTY vs non-PTY execution selection\n\n`BashTool` chooses PTY execution only when all are true:\n\n- tool input `pty === true`\n- `PI_NO_PTY !== \"1\"`\n- tool context has UI (`ctx.hasUI === true` and `ctx.ui` set)\n\nOtherwise it uses non-interactive `executeBash()`.\n\nThat means print mode and non-UI RPC/tool contexts always use non-PTY.\n\n## Non-interactive execution engine (`executeBash`)\n\n## Shell session reuse model\n\n`executeBash()` caches native `Shell` instances in a process-global map keyed by:\n\n- shell path,\n- configured command prefix,\n- snapshot path,\n- serialized shell env,\n- optional agent session key.\n\nSession-level bang-command executions pass `sessionKey: this.sessionId`.\n\nTool-call executions pass `sessionKey: this.session.getSessionId?.()`, when available. In both surfaces, a session key isolates shell reuse per session; without one, reuse falls back to shell config/snapshot/env.\n\n## Shell config and snapshot behavior\n\nAt each call, executor loads settings shell config (`shell`, `env`, optional `prefix`).\n\nIf selected shell includes `bash`, it attempts `getOrCreateSnapshot()`:\n\n- snapshot captures aliases/functions/options from user rc,\n- snapshot creation is best-effort,\n- failure falls back to no snapshot.\n\nIf `prefix` is configured, command becomes:\n\n```text\n<prefix> <command>\n```\n\n## Streaming and cancellation\n\n`Shell.run()` streams chunks to `OutputSink` and optional `onChunk` callback.\n\nCancellation:\n\n- aborted signal triggers `shellSession.abort(...)`,\n- timeout from native result is mapped to `cancelled: true` + annotation text,\n- explicit cancellation similarly returns `cancelled: true` + annotation.\n\nNo exception is thrown inside executor for timeout/cancel; it returns structured `BashResult` and lets caller map error semantics.\n\n## Interactive PTY path (`runInteractiveBashPty`)\n\nWhen PTY is enabled, tool runs `runInteractiveBashPty()` which opens an overlay console component and drives a native `PtySession`.\n\nBehavior highlights:\n\n- xterm-headless virtual terminal renders viewport in overlay,\n- keyboard input is normalized (including Kitty sequences and application cursor mode handling),\n- `esc` while running kills the PTY session,\n- terminal resize propagates to PTY (`session.resize(cols, rows)`).\n\nEnvironment hardening defaults are injected for unattended runs:\n\n- pagers disabled (`PAGER=cat`, `GIT_PAGER=cat`, etc.),\n- editor prompts disabled (`GIT_EDITOR=true`, `EDITOR=true`, ...),\n- terminal/auth prompts reduced (`GIT_TERMINAL_PROMPT=0`, `SSH_ASKPASS=/usr/bin/false`, `CI=1`),\n- package-manager/tool automation flags for non-interactive behavior.\n\nPTY output is normalized (`CRLF`/`CR` to `LF`, `sanitizeText`) and written into `OutputSink`, including artifact spill support.\n\nOn PTY startup/runtime error, sink receives `PTY error: ...` line and command finalizes with undefined exit code.\n\n## Output handling: streaming, truncation, artifact spill\n\nBoth PTY and non-PTY paths use `OutputSink`.\n\n## OutputSink semantics\n\n- keeps an in-memory UTF-8-safe tail buffer (`DEFAULT_MAX_BYTES`, currently 50KB),\n- tracks total bytes/lines seen,\n- if artifact path exists and output overflows (or file already active), writes full stream to artifact file,\n- when memory threshold overflows, trims in-memory buffer to tail (UTF-8 boundary safe),\n- marks `truncated` when overflow/file spill occurs.\n\n`dump()` returns:\n\n- `output` (possibly annotated prefix),\n- `truncated`,\n- `totalLines/totalBytes`,\n- `outputLines/outputBytes`,\n- `artifactId` if artifact file was active.\n\n### Long-output caveat\n\nRuntime truncation is byte-threshold based in `OutputSink` (50KB default). It does not enforce a hard 2000-line cap in this code path.\n\n## Live tool updates and async jobs\n\nFor non-PTY foreground execution, `BashTool` uses a separate `TailBuffer` for partial updates and emits `onUpdate` snapshots while command is running.\n\nFor PTY execution, live rendering is handled by custom UI overlay, not by `onUpdate` text chunks.\n\nWhen `async.enabled` is true and the call passes `async: true`, `BashTool` starts a managed bash job, returns a running job result with a job id, and stores completion through the session managed-job path. Auto-backgrounding can also start this path after `bash.autoBackground.thresholdMs`.\n\n## Result shaping, metadata, and error mapping\n\nAfter execution:\n\n1. `cancelled` handling:\n   - if abort signal is aborted -> throw `ToolAbortError` (abort semantics),\n   - else -> throw `ToolError` (treated as tool failure).\n2. PTY `timedOut` -> throw `ToolError`.\n3. apply head/tail filters to final output text (`applyHeadTail`, head then tail).\n4. empty output becomes `(no output)`.\n5. attach truncation metadata via `toolResult(...).truncationFromSummary(result, { direction: \"tail\" })`.\n6. exit-code mapping:\n   - missing exit code -> `ToolError(\"... missing exit status\")`\n   - non-zero exit -> `ToolError(\"... Command exited with code N\")`\n   - zero exit -> success result.\n\nSuccess payload structure:\n\n- `content`: text output,\n- `details.meta.truncation` when truncated, including:\n  - `direction`, `truncatedBy`, total/output line+byte counts,\n  - `shownRange`,\n  - `artifactId` when available.\n\nBecause built-in tools are wrapped with `wrapToolWithMetaNotice()`, truncation notice text is appended to final text content automatically (for example: `Full: artifact://<id>`).\n\n## Rendering paths\n\n## Tool-call renderer (`bashToolRenderer`)\n\n`bashToolRenderer` is used for tool-call messages (`toolCall` / `toolResult`):\n\n- collapsed mode shows visual-line-truncated preview,\n- expanded mode shows all currently available output text,\n- warning line includes truncation reason and `artifact://<id>` when truncated,\n- timeout value (from args) is shown in footer metadata line.\n\n### Caveat: full artifact expansion\n\n`BashRenderContext` has `isFullOutput`, but current renderer context builder does not set it for bash tool results. Expanded view still uses the text already in result content (tail/truncated output) unless another caller provides full artifact content.\n\n## User bang-command component (`BashExecutionComponent`)\n\n`BashExecutionComponent` is for user `!` commands in interactive mode (not model tool calls):\n\n- streams chunks live,\n- collapsed preview keeps last 20 logical lines,\n- line clamp at 4000 chars per line,\n- shows truncation + artifact warnings when metadata is present,\n- marks cancelled/error/exit state separately.\n\nThis component is wired by `CommandController.handleBashCommand()` and fed from `AgentSession.executeBash()`.\n\n## Mode-specific behavior differences\n\n| Surface                        | Entry path                                            | PTY eligible                                                         | Live output UX                                                           | Error surfacing                                  |\n| ------------------------------ | ----------------------------------------------------- | -------------------------------------------------------------------- | ------------------------------------------------------------------------ | ------------------------------------------------ |\n| Interactive tool call          | `BashTool.execute`                                    | Yes, when `pty=true` and UI exists and `PI_NO_PTY!=1`                | PTY overlay (interactive) or streamed tail updates                       | Tool errors become `toolResult.isError`          |\n| Print mode tool call           | `BashTool.execute`                                    | No (no UI context)                                                   | No TUI overlay; output appears in event stream/final assistant text flow | Same tool error mapping                          |\n| RPC tool call (agent tooling)  | `BashTool.execute`                                    | Usually no UI -> non-PTY                                             | Structured tool events/results                                           | Same tool error mapping                          |\n| Interactive bang command (`!`) | `AgentSession.executeBash` + `BashExecutionComponent` | No (uses executor directly)                                          | Dedicated bash execution component                                       | Controller catches exceptions and shows UI error |\n| RPC `bash` command             | `rpc-mode` -> `session.executeBash`                   | No                                                                   | Returns `BashResult` directly                                            | Consumer handles returned fields                 |\n\n## Operational caveats\n\n- Interceptor only blocks commands when suggested tool is currently available in context.\n- If artifact allocation fails, truncation still occurs but no `artifact://` back-reference is available.\n- Shell session cache has no explicit eviction in this module; lifetime is process-scoped.\n- PTY and non-PTY timeout surfaces differ:\n  - PTY exposes explicit `timedOut` result field,\n  - non-PTY maps timeout into `cancelled + annotation` summary.\n\n## Implementation files\n\n- [`src/tools/bash.ts`](../packages/coding-agent/src/tools/bash.ts) — tool entrypoint, input handling/interception, async and PTY/non-PTY selection, result/error mapping, bash tool renderer.\n- [`src/tools/bash-normalize.ts`](../packages/coding-agent/src/tools/bash-normalize.ts) — post-run head/tail filtering; also contains an unused command-normalization helper.\n- [`src/tools/bash-interceptor.ts`](../packages/coding-agent/src/tools/bash-interceptor.ts) — interceptor rule matching and blocked-command messages.\n- [`src/exec/bash-executor.ts`](../packages/coding-agent/src/exec/bash-executor.ts) — non-PTY executor, shell session reuse, cancellation wiring, output sink integration.\n- [`src/tools/bash-interactive.ts`](../packages/coding-agent/src/tools/bash-interactive.ts) — PTY runtime, overlay UI, input normalization, non-interactive env defaults.\n- [`src/session/streaming-output.ts`](../packages/coding-agent/src/session/streaming-output.ts) — `OutputSink`, `TailBuffer`, truncation/artifact spill, and summary metadata.\n- [`src/tools/output-meta.ts`](../packages/coding-agent/src/tools/output-meta.ts) — truncation metadata shape + notice injection wrapper.\n- [`src/session/agent-session.ts`](../packages/coding-agent/src/session/agent-session.ts) — session-level `executeBash`, message recording, abort lifecycle.\n- [`src/modes/components/bash-execution.ts`](../packages/coding-agent/src/modes/components/bash-execution.ts) — interactive `!` command execution component.\n- [`src/modes/controllers/command-controller.ts`](../packages/coding-agent/src/modes/controllers/command-controller.ts) — wiring for interactive `!` command UI stream/update completion.\n- [`src/modes/rpc/rpc-mode.ts`](../packages/coding-agent/src/modes/rpc/rpc-mode.ts) — RPC `bash` and `abort_bash` command surface.\n- [`src/internal-urls/artifact-protocol.ts`](../packages/coding-agent/src/internal-urls/artifact-protocol.ts) — `artifact://<id>` resolution.\n",
@@ -37,6 +37,7 @@ export const EMBEDDED_DOCS: Readonly<Record<string, string>> = {
 	"porting-to-natives.md": "# Porting to pi-natives (N-API) — Field Notes\n\nThis is a practical guide for moving hot paths into `crates/pi-natives` and wiring them through the generated native package entrypoint. It exists to avoid the same failures happening twice.\n\n## When to port\n\nPort when any of these are true:\n\n- The hot path runs in render loops, tight UI updates, or large batches.\n- JS allocations dominate (string churn, regex backtracking, large arrays).\n- You already have a JS baseline and can benchmark both versions side by side.\n- The work is CPU-bound or blocking I/O that can run on the libuv thread pool.\n- The work is async I/O that can run on Tokio's runtime (for example shell execution).\n\nAvoid ports that depend on JS-only state or dynamic imports. N-API exports should be data-in/data-out. Long-running work should go through `task::blocking` (CPU-bound/blocking I/O) or `task::future` (async I/O) with cancellation where the caller needs `timeoutMs` or `AbortSignal`.\n\n## Current package shape\n\n`@oh-my-pi/pi-natives` no longer has a `packages/natives/src/<module>` TypeScript wrapper layer. The package root points at generated native artifacts:\n\n- runtime entry: `packages/natives/native/index.js`\n- types entry: `packages/natives/native/index.d.ts`\n- loader helpers: `packages/natives/native/loader-state.js`\n- embedded manifest: `packages/natives/native/embedded-addon.js`\n\nConsumers import directly from `@oh-my-pi/pi-natives`. The generated declarations are produced during `bun --cwd=packages/natives run build`.\n\n## Anatomy of a native export\n\n**Rust side:**\n\n- Implementation lives in `crates/pi-natives/src/<module>.rs`.\n- If you add a new module, register it in `crates/pi-natives/src/lib.rs`.\n- Export with `#[napi]`; snake_case exports are converted to camelCase automatically. Use explicit JS names only for true aliases/non-default names. Use `#[napi(object)]` for object-shaped structs.\n- For CPU-bound or blocking work, use `task::blocking(tag, cancel_token, work)`.\n- For async work that needs Tokio, use `task::future(env, tag, work)`.\n- Pass a `CancelToken` when the API exposes `timeoutMs` or `AbortSignal`, and call `heartbeat()` inside long loops.\n\n**Package/build side:**\n\n- `packages/natives/scripts/build-native.ts` runs napi-rs, installs the `.node` artifact, copies generated `index.js`/`index.d.ts`, and appends enum runtime exports.\n- `packages/natives/native/index.js` is the loader that chooses a candidate `.node` file and returns the loaded addon.\n- `packages/natives/package.json` exposes only the package root (`@oh-my-pi/pi-natives`).\n\n**Consumer side:**\n\n- Update direct imports/callsites in `packages/coding-agent` or `packages/tui` when the new export replaces a JS implementation.\n- Keep higher-level policy in consumers unless it belongs in the native primitive itself.\n\n## Porting checklist\n\n1. **Add the Rust implementation**\n\n- Put the core logic in a plain Rust function.\n- If it is a new module, add it to `crates/pi-natives/src/lib.rs`.\n- Expose it with `#[napi]` so the default snake_case -> camelCase mapping stays consistent.\n- Keep signatures owned and simple: `String`, `Vec<String>`, `Uint8Array`, `Either<JsString, Uint8Array>`, or `#[napi(object)]` structs.\n- For CPU-bound or blocking work, use `task::blocking`; for async work, use `task::future`.\n- If exposing cancellation, include `timeout_ms: Option<u32>` and `signal: Option<Unknown<'env>>` in options, create `CancelToken::new(...)`, and heartbeat in long loops.\n\n2. **Build generated bindings**\n\n- Run `bun --cwd=packages/natives run build`.\n- Confirm the generated `packages/natives/native/index.d.ts` includes the new export with the intended JS name/signature.\n- Confirm `packages/natives/native/index.js` still has generated enum exports appended when enum changes are involved.\n\n3. **Update consumers**\n\n- Import the new export directly from `@oh-my-pi/pi-natives`.\n- Replace only callsites where the native implementation is faster/equivalent and preserves behavior.\n- Remove obsolete JS implementation code in the same change when the native path becomes canonical.\n\n4. **Add benchmarks**\n\n- Put benchmarks next to the owning package (`packages/tui/bench`, `packages/natives/bench`, or `packages/coding-agent/bench`).\n- Include a JS baseline and native version in the same run.\n- Use `Bun.nanoseconds()` and a fixed iteration count.\n- Keep benchmark inputs realistic for the hot path.\n\n5. **Run focused verification**\n\n- Build the native package.\n- Run the benchmark.\n- Run the narrow tests or scenario covering the changed export/callsites.\n\n## Pain points and how to avoid them\n\n### 1) Stale platform/variant artifacts\n\nThe loader probes platform-tagged artifacts in deterministic order. For x64, selected variant candidates are tried before the unsuffixed default fallback:\n\n- `modern`: `pi_natives.<tag>-modern.node`, then `...-baseline.node`, then `pi_natives.<tag>.node`.\n- `baseline`: `pi_natives.<tag>-baseline.node`, then `pi_natives.<tag>.node`.\n\nNon-x64 uses `pi_natives.<tag>.node`.\n\nCompiled binaries also probe `<getNativesDir()>/<version>/...` and a legacy user-data directory before package/executable locations. If any earlier candidate is stale, a new export may appear missing.\n\n**Fix:** remove stale candidate/cache files and rebuild.\n\n```bash\nrm packages/natives/native/pi_natives.<platform>-<arch>.node\nrm packages/natives/native/pi_natives.<platform>-<arch>-modern.node\nrm packages/natives/native/pi_natives.<platform>-<arch>-baseline.node\nbun --cwd=packages/natives run build\n```\n\nFor compiled binaries, delete the versioned addon cache shown in the loader error (normally under `~/.omp/natives/<version>` unless `$XDG_DATA_HOME/omp` is used).\n\n### 2) Generated types do not match loaded binary\n\nThis can happen when `native/index.d.ts` was regenerated but the `.node` file being loaded is stale or from a different platform/variant.\n\nVerify the loaded export set from the actual candidate path:\n\n```bash\nbun -e 'const tag = `${process.platform}-${process.arch}`; const mod = require(`./packages/natives/native/pi_natives.${tag}.node`); console.log(Object.keys(mod).sort())'\n```\n\nFix the build/candidate mismatch. Do not paper over it with optional consumer checks if the export is required.\n\n### 3) Rust signature mismatch\n\nKeep N-API signatures simple and owned. Avoid borrowed references like `&str` in public exports. If you need structured data, use `#[napi(object)]` structs. If you need callbacks, use napi-rs `ThreadsafeFunction` and keep callback error/value behavior explicit.\n\n### 4) Enum runtime exports\n\nnapi-rs declarations alone are not enough for JS callers that use enum objects at runtime. `scripts/gen-enums.ts` appends enum objects to `native/index.js`. If you add or change a native enum, verify both `native/index.d.ts` and the generated enum export block in `native/index.js`.\n\n### 5) Benchmarking mistakes\n\n- Do not compare different inputs or allocations.\n- Keep JS and native using identical input arrays.\n- Run both in the same benchmark file to avoid skew.\n- Include enough iterations to smooth startup noise, but keep inputs realistic.\n\n## Benchmark template\n\n```ts\nconst ITERATIONS = 2000;\n\nfunction bench(name: string, fn: () => void): number {\n  const start = Bun.nanoseconds();\n  for (let i = 0; i < ITERATIONS; i++) fn();\n  const elapsed = (Bun.nanoseconds() - start) / 1e6;\n  console.log(\n    `${name}: ${elapsed.toFixed(2)}ms total (${(elapsed / ITERATIONS).toFixed(6)}ms/op)`,\n  );\n  return elapsed;\n}\n\nbench(\"feature/js\", () => {\n  jsImpl(sample);\n});\n\nbench(\"feature/native\", () => {\n  nativeImpl(sample);\n});\n```\n\n## Verification checklist\n\n- Generated `native/index.d.ts` includes the new export and intended TS signature.\n- The loaded `.node` file's `Object.keys(require(candidate))` includes the new export.\n- Runtime enum objects are present when the change adds/changes enums.\n- Bench numbers are recorded in the PR/notes.\n- Call sites are updated only if native is faster/equal and behavior-compatible.\n- Obsolete JS code is removed when the native implementation becomes canonical.\n\n## Rule of thumb\n\n- If native is slower, do not switch callsites. Keep or remove the export based on whether it has a near-term owner.\n- If native is faster and behavior-compatible, switch callsites and keep a benchmark to catch regressions.\n",
 	"provider-streaming-internals.md": "# Provider streaming internals\n\nThis document explains how token/tool streaming is normalized in `@oh-my-pi/pi-ai`, then propagated through `@oh-my-pi/pi-agent-core` and `coding-agent` session events.\n\n## End-to-end flow\n\n1. `streamSimple()` (`packages/ai/src/stream.ts`) maps generic options and dispatches to a provider stream function.\n2. Provider stream functions translate provider-native stream events into the unified `AssistantMessageEvent` sequence. Current built-ins include Anthropic, OpenAI Responses/Completions/Codex/Azure Responses, Google Gemini/Gemini CLI/Vertex, Bedrock Converse, Ollama, Cursor, plus GitLab Duo/Kimi wrappers and extension-registered custom APIs.\n3. Each provider pushes events into `AssistantMessageEventStream` (`packages/ai/src/utils/event-stream.ts`), which throttles delta events and exposes:\n   - async iteration for incremental updates\n   - `result()` for final `AssistantMessage`\n4. `agentLoop` (`packages/agent/src/agent-loop.ts`) consumes those events, mutates in-flight assistant state, and emits `message_update` events carrying the raw `assistantMessageEvent`.\n5. `AgentSession` (`packages/coding-agent/src/session/agent-session.ts`) subscribes to agent events, persists messages, drives extension hooks, and applies session behaviors (retry, compaction, TTSR, streaming-edit abort checks).\n\n## Unified stream contract in `@oh-my-pi/pi-ai`\n\nAll providers emit the same shape (`AssistantMessageEvent` in `packages/ai/src/types.ts`):\n\n- `start`\n- content block lifecycle triplets:\n  - text: `text_start` → `text_delta`\\* → `text_end`\n  - thinking: `thinking_start` → `thinking_delta`\\* → `thinking_end`\n  - tool call: `toolcall_start` → `toolcall_delta`\\* → `toolcall_end`\n- terminal event:\n  - `done` with `reason: \"stop\" | \"length\" | \"toolUse\"`\n  - or `error` with `reason: \"aborted\" | \"error\"`\n\n`AssistantMessageEventStream` guarantees:\n\n- final result is resolved by terminal event (`done` or `error`)\n- deltas are batched/throttled (~50ms)\n- buffered deltas are flushed before non-delta events and before completion\n\n## Delta throttling and harmonization behavior\n\n`AssistantMessageEventStream` treats `text_delta`, `thinking_delta`, and `toolcall_delta` as mergeable events:\n\n- buffered deltas are merged only when **type + contentIndex** match\n- merge keeps the latest `partial` snapshot\n- non-delta events force immediate flush\n\nThis smooths high-frequency provider streams for TUI/event consumers, but is not provider backpressure: providers still produce at full speed, while the local stream buffers.\n\n## Provider normalization details\n\n## Anthropic (`anthropic-messages`)\n\nSource: `packages/ai/src/providers/anthropic.ts`\n\nNormalization points:\n\n- `message_start` initializes usage (input/output/cache tokens)\n- `content_block_start` maps to text/thinking/toolcall starts\n- `content_block_delta` maps:\n  - `text_delta` → `text_delta`\n  - `thinking_delta` → `thinking_delta`\n  - `input_json_delta` → `toolcall_delta`\n  - `signature_delta` updates `thinkingSignature` only (no event)\n- `content_block_stop` emits corresponding `*_end`\n- `message_delta.stop_reason` maps via `mapStopReason()`\n\nTool-call argument streaming:\n\n- each tool block carries internal `partialJson`\n- every JSON delta appends to `partialJson`\n- `arguments` are reparsed on each delta via `parseStreamingJson()`\n- `toolcall_end` reparses once more, then strips `partialJson`\n\n## OpenAI Responses family (`openai-responses`, `openai-codex-responses`, `azure-openai-responses`)\n\nSources: `packages/ai/src/providers/openai-responses.ts`, `openai-codex-responses.ts`, and `azure-openai-responses.ts`\n\nNormalization points:\n\n- `response.output_item.added` starts reasoning/text/function-call blocks\n- reasoning summary events (`response.reasoning_summary_text.delta`) become `thinking_delta`\n- output/refusal deltas become `text_delta`\n- `response.function_call_arguments.delta` becomes `toolcall_delta`\n- `response.output_item.done` emits `thinking_end` / `text_end` / `toolcall_end`\n- `response.completed` maps status to stop reason and usage\n\nTool-call argument streaming:\n\n- same `partialJson` accumulation pattern as Anthropic\n- providers that send only `response.function_call_arguments.done` still populate final args\n- tool call IDs are normalized as `\"<call_id>|<item_id>\"`\n\n## Google Generative AI (`google-generative-ai`)\n\nSource: `packages/ai/src/providers/google.ts`\n\nNormalization points:\n\n- iterates `candidate.content.parts`\n- text parts are split into thinking vs text by `isThinkingPart(part)`\n- block transitions close previous block before starting a new one\n- `part.functionCall` is treated as a complete tool call (start/delta/end emitted immediately)\n- finish reason mapped by `mapStopReason()` from `google-shared.ts`\n\nTool-call argument streaming:\n\n- function call args arrive as structured object, not incremental JSON text\n- implementation emits one synthetic `toolcall_delta` containing `JSON.stringify(arguments)`\n- no partial JSON parser needed for Google in this path\n\n## Partial tool-call JSON accumulation and recovery\n\nShared behavior for Anthropic/OpenAI Responses uses `parseStreamingJson()` (`packages/ai/src/utils/json-parse.ts`):\n\n1. try `JSON.parse`\n2. fallback to `partial-json` parser for incomplete fragments\n3. if both fail, return `{}`\n\nImplications:\n\n- malformed or truncated argument deltas do not crash stream processing immediately\n- in-progress `arguments` may temporarily be `{}`\n- later valid deltas can recover structured arguments because parsing is retried on every append\n- final `toolcall_end` performs one more parse attempt before emission\n\n## Stop reasons vs transport/runtime errors\n\nProvider stop reasons are mapped to normalized `stopReason`:\n\n- Anthropic: `end_turn`→`stop`, `max_tokens`→`length`, `tool_use`→`toolUse`, safety/refusal cases→`error`\n- OpenAI Responses: `completed`→`stop`, `incomplete`→`length`, `failed/cancelled`→`error`\n- Google: `STOP`→`stop`, `MAX_TOKENS`→`length`, safety/prohibited/malformed-function-call classes→`error`\n\nError semantics are split in two stages:\n\n1. **Model completion semantics** (provider reported finish reason/status)\n2. **Transport/runtime failure** (network/client/parser/abort exceptions)\n\nIf provider stream throws or signals failure, each provider wrapper catches and emits terminal `error` event with:\n\n- `stopReason = \"aborted\"` when abort signal is set\n- otherwise `stopReason = \"error\"`\n- `errorMessage = formatErrorMessageWithRetryAfter(error)`\n\n## Malformed chunk / SSE parse failure behavior\n\nFor these provider paths, chunk/SSE framing is handled by vendor SDK streams (Anthropic SDK, OpenAI SDK, Google SDK). This code does not implement a custom SSE decoder here.\n\nObserved behavior in current implementation:\n\n- malformed chunk/SSE parsing at SDK level surfaces as an exception or stream `error` event\n- provider wrapper converts that into unified terminal `error` event\n- no provider-specific resume/retry inside the stream function itself\n- higher-level retries are handled in `AgentSession` auto-retry logic (message-level retry, not stream-chunk replay)\n\n## Cancellation boundaries\n\nCancellation is layered:\n\n- AI provider request: `options.signal` is passed into provider client stream call.\n- Provider wrapper: after stream loop, aborted signal forces error path (`\"Request was aborted\"`).\n- Agent loop: checks `signal.aborted` before handling each provider event and can synthesize an aborted assistant message from the latest partial.\n- Session/agent controls: `AgentSession.abort()` -> `agent.abort()` -> shared abort controller cancellation.\n\nTool execution cancellation is separate from model stream cancellation:\n\n- tool runners use `AbortSignal.any([agentSignal, steeringAbortSignal])`\n- steering interrupts can abort remaining tool execution while preserving already-produced tool results\n\n## Backpressure boundaries\n\nThere is no hard backpressure mechanism between provider SDK stream and downstream consumers:\n\n- `EventStream` uses in-memory queues with no max size\n- throttling reduces UI update rate but does not slow provider intake\n- if consumers lag significantly, queued events can grow until completion\n\nCurrent design favors responsiveness and simple ordering over bounded-buffer flow control.\n\n## How stream events surface as agent/session events\n\n`agentLoop.streamAssistantResponse()` bridges `AssistantMessageEvent` to `AgentEvent`:\n\n- on `start`: pushes placeholder assistant message and emits `message_start`\n- on block events (`text_*`, `thinking_*`, `toolcall_*`): updates last assistant message, emits `message_update` with raw `assistantMessageEvent`\n- on terminal (`done`/`error`): resolves final message from `response.result()`, emits `message_end`\n\n`AgentSession` then consumes those events for session-level behaviors:\n\n- TTSR watches `message_update.assistantMessageEvent` for `text_delta`, `thinking_delta`, and `toolcall_delta`\n- streaming edit guard inspects `toolcall_delta`/`toolcall_end` on `edit` calls and can abort early\n- persistence writes finalized messages at `message_end`\n- auto-retry examines assistant `stopReason === \"error\"` plus `errorMessage` heuristics\n\n## Unified vs provider-specific responsibilities\n\nUnified (common contract):\n\n- event shape (`AssistantMessageEvent`)\n- final result extraction (`done`/`error`)\n- delta throttling + merge rules\n- agent/session event propagation model\n\nProvider-specific (not fully abstracted):\n\n- upstream event taxonomies and mapping logic\n- stop-reason translation tables\n- tool-call ID conventions\n- reasoning/thinking block semantics and signatures\n- usage token semantics and availability timing\n- message conversion constraints per API\n\n## Implementation files\n\n- [`../../ai/src/stream.ts`](../packages/ai/src/stream.ts) — provider dispatch, option mapping, API key/session plumbing, custom API dispatch, and provider-specific credential handling.\n- [`../../ai/src/utils/event-stream.ts`](../packages/ai/src/utils/event-stream.ts) — generic stream queue + assistant delta throttling.\n- [`../../ai/src/utils/json-parse.ts`](../packages/ai/src/utils/json-parse.ts) — partial JSON parsing for streamed tool arguments.\n- [`../../ai/src/providers/anthropic.ts`](../packages/ai/src/providers/anthropic.ts) — Anthropic event translation and tool JSON delta accumulation.\n- [`../../ai/src/providers/openai-responses.ts`](../packages/ai/src/providers/openai-responses.ts), [`openai-codex-responses.ts`](../packages/ai/src/providers/openai-codex-responses.ts), [`azure-openai-responses.ts`](../packages/ai/src/providers/azure-openai-responses.ts) — Responses-family event translation and status mapping.\n- [`../../ai/src/providers/google.ts`](../packages/ai/src/providers/google.ts), [`google-gemini-cli.ts`](../packages/ai/src/providers/google-gemini-cli.ts), [`google-vertex.ts`](../packages/ai/src/providers/google-vertex.ts) — Gemini stream chunk-to-block translation variants.\n- [`../../ai/src/providers/google-shared.ts`](../packages/ai/src/providers/google-shared.ts) — Gemini finish-reason mapping and shared conversion rules.\n- [`../../ai/src/providers/amazon-bedrock.ts`](../packages/ai/src/providers/amazon-bedrock.ts), [`openai-completions.ts`](../packages/ai/src/providers/openai-completions.ts), [`ollama.ts`](../packages/ai/src/providers/ollama.ts), [`cursor.ts`](../packages/ai/src/providers/cursor.ts) — additional built-in stream adapters using the same event contract.\n- [`../../agent/src/agent-loop.ts`](../packages/agent/src/agent-loop.ts) — provider stream consumption and `message_update` bridging.\n- [`../src/session/agent-session.ts`](../packages/coding-agent/src/session/agent-session.ts) — session-level handling of streaming updates, abort, retry, and persistence.\n",
 	"python-repl.md": "# Eval Tool Python Backend and IPython Runtime\n\nThis document describes the current Python execution stack in `packages/coding-agent`.\nIt covers tool behavior, kernel/gateway lifecycle, environment handling, execution semantics, output rendering, and operational failure modes.\n\n## Scope and Key Files\n\n- Tool surface: `src/tools/eval.ts`\n- Session/per-call kernel orchestration: `src/eval/py/executor.ts`\n- Kernel protocol + gateway integration: `src/eval/py/kernel.ts`\n- Shared local gateway coordinator: `src/eval/py/gateway-coordinator.ts`\n- Interactive-mode renderer for user-triggered Python runs: `src/modes/components/eval-execution.ts`\n- Runtime/env filtering and Python resolution: `src/eval/py/runtime.ts`\n\n## What eval's Python backend is\n\nThe `eval` tool executes one or more Python cells through a Jupyter Kernel Gateway-backed kernel when `language: \"python\"` is selected or inferred (not by spawning `python -c` directly per cell).\n\nTool params:\n\n```ts\n{\n  cells: Array<{ code: string; title?: string }>;\n  timeout?: number; // seconds, clamped to 1..600, default 30\n  reset?: boolean; // reset selected runtime before the first cell only\n}\n```\n\nThe tool is `concurrency = \"exclusive\"` for a session, so calls do not overlap.\n\n## Gateway lifecycle\n\n### Modes\n\nThere are two gateway paths:\n\n1. **External gateway** (`PI_PYTHON_GATEWAY_URL` set)\n   - Uses the configured URL directly.\n   - Optional auth with `PI_PYTHON_GATEWAY_TOKEN`.\n   - No local gateway process is spawned or managed.\n\n2. **Local shared gateway** (default path)\n   - Uses a single shared process coordinated under `~/.omp/agent/python-gateway`.\n   - Metadata file: `gateway.json`\n   - Lock file: `gateway.lock`\n   - Spawn command:\n     - `python -m kernel_gateway`\n     - bound to `127.0.0.1:<allocated-port>`\n     - startup health check: `GET /api/kernelspecs`\n\n### Local shared gateway coordination\n\n`acquireSharedGateway()`:\n\n- Takes a file lock (`gateway.lock`) with heartbeat.\n- Reuses `gateway.json` if PID is alive and health check passes.\n- Cleans stale info/PIDs when needed.\n- Starts a new gateway when no healthy one exists.\n\n`releaseSharedGateway()` is currently a no-op (kernel shutdown does not tear down shared gateway).\n\n`shutdownSharedGateway()` explicitly terminates the shared process and clears gateway metadata.\n\n### Important constraint\n\n`python.sharedGateway=false` is rejected at kernel start:\n\n- Error: `Shared Python gateway required; local gateways are disabled`\n- There is no per-process non-shared local gateway mode.\n\n## Kernel lifecycle\n\nKernels are created via `POST /api/kernels` on the selected gateway when a retained session needs a kernel or when `per-call` mode starts a request.\n\nKernel startup sequence:\n\n1. Availability check (`checkPythonKernelAvailability`)\n2. Create kernel (`/api/kernels`)\n3. Open websocket (`/api/kernels/:id/channels`)\n4. Initialize kernel env (`cwd`, env vars, `sys.path`)\n5. Execute `PYTHON_PRELUDE`\n6. Load extension modules from:\n   - user: `~/.omp/agent/modules/*.py`\n   - project: `<cwd>/.omp/modules/*.py` (overrides same-name user module)\n\nKernel shutdown:\n\n- Deletes remote kernel via `DELETE /api/kernels/:id`\n- Closes websocket\n- Calls shared gateway release hook (no-op today)\n\n## Session persistence semantics\n\n`python.kernelMode` controls retained kernel reuse:\n\n- `session` (default)\n  - Reuses kernel sessions keyed by session file plus cwd when a session file exists; otherwise by cwd.\n  - Execution is serialized per session via a queue.\n  - Idle sessions are evicted after 5 minutes.\n  - At most 4 sessions; oldest is evicted on overflow.\n  - Heartbeat checks detect dead kernels.\n  - Auto-restart allowed once; repeated crash => hard failure.\n\n- `per-call`\n  - Creates a fresh kernel for each execute request.\n  - Shuts kernel down after the request.\n  - No cross-call state persistence.\n\n### Multi-cell behavior in a single tool call\n\nCells run sequentially in the same kernel instance for that tool call.\n\nIf an intermediate cell fails:\n\n- Earlier cell state remains in memory.\n- Tool returns a targeted error indicating which cell failed.\n- Later cells are not executed.\n\n`reset=true` only applies to the first cell execution in that call.\n\n## Environment filtering and runtime resolution\n\nEnvironment is filtered before launching gateway/kernel runtime:\n\n- Allowlist includes core vars like `PATH`, `HOME`, locale vars, `VIRTUAL_ENV`, `PYTHONPATH`, etc.\n- Allow-prefixes: `LC_`, `XDG_`, `PI_`\n- Denylist strips common API keys (OpenAI/Anthropic/Gemini/etc.)\n\nRuntime selection order:\n\n1. Active/located venv (`VIRTUAL_ENV`, then `<cwd>/.venv`, `<cwd>/venv`)\n2. Managed venv at `~/.omp/python-env`\n3. `python` or `python3` on PATH\n\nWhen a venv is selected, its bin/Scripts path is prepended to `PATH`.\n\nKernel startup receives the optional session file path from the executor:\n\n- `PI_SESSION_FILE` (session state file path)\n\n`PythonKernel.#initializeKernelEnvironment(...)` then runs init script inside kernel to:\n\n- `os.chdir(cwd)`\n- injects provided env entries into `os.environ`\n- ensures cwd is in `sys.path`\n\n## Tool availability and mode selection\n\n`eval.py` / `eval.js` (both default `true`) plus optional `PI_PY` override controls eval backend exposure:\n\n- Python backend only (`eval.py=true`, `eval.js=false`)\n- JavaScript backend only (`eval.py=false`, `eval.js=true`)\n- both backends\n\n`PI_PY` accepted values:\n\n- `0` / `bash` -> JavaScript backend only\n- `1` / `py` -> Python backend only\n- `mix` / `both` -> both backends\n\nIf Python preflight fails and `eval.js` is enabled, `eval` remains available and dispatches to JavaScript unless `language: \"python\"` is explicitly requested.\n\n## Execution flow and cancellation/timeout\n\n### Tool-level timeout\n\n`eval` timeout is in seconds, default 30, clamped to `1..600`.\n\nThe tool combines:\n\n- caller abort signal\n- timeout abort signal\n\nwith `AbortSignal.any(...)`.\n\n### Kernel execution cancellation\n\nOn abort/timeout:\n\n- Execution is marked cancelled.\n- Kernel interrupt is attempted via REST (`POST /interrupt`) and control-channel `interrupt_request`.\n- Result includes `cancelled=true`.\n- Timeout path annotates output as `Command timed out after <n> seconds`.\n\n### stdin behavior\n\nInteractive stdin is not supported.\n\nIf kernel emits `input_request`:\n\n- Tool records `stdinRequested=true`\n- Emits explanatory text\n- Sends empty `input_reply`\n- Execution is treated as failure at executor layer\n\n## Output capture and rendering\n\n### Captured output classes\n\nFrom kernel messages:\n\n- `stream` -> plain text chunks\n- `display_data`/`execute_result` -> rich display handling\n- `error` -> traceback text\n- custom MIME `application/x-omp-status` -> structured status events\n\nDisplay MIME precedence:\n\n1. `text/markdown`\n2. `text/plain`\n3. `text/html` (converted to basic markdown)\n\nAdditionally captured as structured outputs:\n\n- `application/json` -> JSON tree data\n- `image/png` -> image payloads\n- `application/x-omp-status` -> status events\n\n### Storage and truncation\n\nOutput is streamed through `OutputSink` and may be persisted to artifact storage.\n\nTool results can include truncation metadata and `artifact://<id>` for full output recovery.\n\n### Renderer behavior\n\n- Tool renderer (`eval.ts`):\n  - shows code-cell blocks with per-cell status\n  - collapsed preview defaults to 10 lines\n  - supports expanded mode for full output and richer status detail\n- Interactive renderer (`eval-execution.ts`):\n  - used for user-triggered Python execution in TUI\n  - collapsed preview defaults to 20 lines\n  - clamps very long individual lines to 4000 chars for display safety\n  - shows cancellation/error/truncation notices\n\n## External gateway support\n\nSet:\n\n```bash\nexport PI_PYTHON_GATEWAY_URL=\"http://127.0.0.1:8888\"\n# Optional:\nexport PI_PYTHON_GATEWAY_TOKEN=\"...\"\n```\n\nBehavior differences from local shared gateway:\n\n- No local gateway lock/info files\n- No local process spawn/termination\n- Health checks and kernel CRUD run against external endpoint\n- Auth failures are surfaced with explicit token guidance\n\n## Operational troubleshooting (current failure modes)\n\n- **Python backend not available**\n  - Check `eval.py`, Python kernel dependencies, and `PI_PY`.\n  - If preflight fails and `eval.js` is enabled, omit `language` or pass `language: \"js\"` to use JavaScript.\n\n- **Kernel availability errors**\n  - Local mode requires both `kernel_gateway` and `ipykernel` importable in resolved Python runtime.\n  - Install with:\n    ```bash\n    python -m pip install jupyter_kernel_gateway ipykernel\n    ```\n\n- **`python.sharedGateway=false` causes startup failure**\n  - This is expected with current implementation.\n\n- **External gateway auth/reachability failures**\n  - 401/403 -> set `PI_PYTHON_GATEWAY_TOKEN`.\n  - timeout/unreachable -> verify URL/network and gateway health.\n\n- **Execution hangs then times out**\n  - Increase tool `timeout` (max 600s) if workload is legitimate.\n  - For stuck code, cancellation triggers kernel interrupt but user code may still need refactor.\n\n- **stdin/input prompts in Python code**\n  - `input()` is not supported interactively in this runtime path; pass data programmatically.\n\n- **Resource exhaustion (`EMFILE` / too many open files)**\n  - Session manager triggers shared-gateway recovery (session teardown + shared gateway restart).\n\n- **Working directory errors**\n  - Tool validates `cwd` exists and is a directory before execution.\n\n## Relevant environment variables\n\n- `PI_PY` — tool exposure override (`bash-only`/`ipy-only`/`both` mapping above)\n- `PI_PYTHON_GATEWAY_URL` — use external gateway\n- `PI_PYTHON_GATEWAY_TOKEN` — optional external gateway auth token\n- `PI_PYTHON_SKIP_CHECK=1` — bypass Python preflight/warm checks\n- `PI_PYTHON_IPC_TRACE=1` — log kernel IPC send/receive traces\n- `PI_DEBUG_STARTUP=1` — emit startup-stage debug markers\n",
+	"render-mermaid.md": "# RenderMermaid\n\n`RenderMermaid` is an optional built-in tool that renders Mermaid source to terminal-friendly text.\n\n## Enable it\n\nDisabled by default. Turn it on in `/settings` under **Tools → Render Mermaid**, or in `~/.omp/agent/config.yml`:\n\n```yaml\nrenderMermaid:\n  enabled: true\n```\n\n## What it does\n\n- Tool name: `render_mermaid`\n- Input: Mermaid source in the required `mermaid` field\n- Output: rendered ASCII/Unicode text, not SVG or PNG\n- Storage: when artifact storage is available, the full render is also saved as an `artifact://...`\n\nThere are no model-specific or environment-variable prerequisites. Once enabled, any model that can call built-in tools can use it.\n\n## Parameters\n\n```json\n{\n  \"mermaid\": \"graph TD\\n  A[Start] --> B[Stop]\",\n  \"config\": {\n    \"useAscii\": false,\n    \"paddingX\": 2,\n    \"paddingY\": 2,\n    \"boxBorderPadding\": 0\n  }\n}\n```\n\nAvailable `config` fields:\n\n- `useAscii` — `true` for plain ASCII, `false` for Unicode box-drawing characters (default and usually more readable)\n- `paddingX` — horizontal spacing between nodes\n- `paddingY` — vertical spacing between nodes\n- `boxBorderPadding` — inner padding inside node boxes\n\n## Current limitations\n\n`RenderMermaid` uses the `beautiful-mermaid` ASCII renderer. It works best for flowcharts and small diagrams.\n\nComplex sequence diagrams, especially with `alt` / `else` blocks, can become very wide in a terminal. That is current renderer behavior, not a provider or model configuration problem.\n\nIf a sequence diagram is hard to read:\n\n1. Keep Unicode output (`useAscii: false`)\n2. Reduce spacing with a tighter config such as `paddingX: 2`, `paddingY: 2`, `boxBorderPadding: 0`\n3. Prefer smaller sub-diagrams over one large sequence diagram\n4. Open the saved artifact if the inline preview is truncated in the TUI\n\n## Example\n\nInput:\n\n```mermaid\ngraph TD\n  A[Start] --> B{Decision}\n  B -->|Yes| C[Action]\n  B -->|No| D[End]\n```\n\nTypical result:\n\n```text\n┌─────┐\n│Start│\n└─────┘\n   │\n   ▼\n┌────────┐\n│Decision│\n└────────┘\n```\n",
 	"resolve-tool-runtime.md": "# Resolve tool runtime internals\n\nThis document explains how preview/apply workflows are modeled in coding-agent and how built-in or custom tools can participate via the tool-choice queue and `pushPendingAction`.\n\n## Scope and key files\n\n- [`src/tools/resolve.ts`](../packages/coding-agent/src/tools/resolve.ts)\n- [`src/tools/ast-edit.ts`](../packages/coding-agent/src/tools/ast-edit.ts)\n- [`src/extensibility/custom-tools/types.ts`](../packages/coding-agent/src/extensibility/custom-tools/types.ts)\n- [`src/extensibility/custom-tools/loader.ts`](../packages/coding-agent/src/extensibility/custom-tools/loader.ts)\n- [`src/sdk.ts`](../packages/coding-agent/src/sdk.ts)\n\n## What `resolve` does\n\n`resolve` is a hidden tool that finalizes a pending preview action.\n\n- `action: \"apply\"` executes the queued action's `apply(reason)` callback and returns that result with resolve metadata.\n- `action: \"discard\"` invokes `reject(reason)` if provided; otherwise returns `Discarded: <label>. Reason: <reason>`.\n\nIf no pending action exists, `resolve` fails with:\n\n- `No pending action to resolve. Nothing to apply or discard.`\n\n## Pending actions use the tool-choice queue\n\nPreview producers call `queueResolveHandler(...)`, which pushes a one-shot forced `resolve` directive onto the session tool-choice queue and adds a `resolve-reminder` steering message.\n\nRuntime behavior:\n\n- the queued handler owns the pending `apply`/`reject` callbacks,\n- `resolve` looks up the current queue invoker with `session.peekQueueInvoker()`,\n- if the model rejects the forced tool choice, the queue directive is requeued,\n- `resolve` does not maintain a separate pending-action stack.\n\nMultiple pending previews therefore follow the active tool-choice queue ordering, not an independent pending-action store.\n\n## Built-in producer example (`ast_edit`)\n\n`ast_edit` previews structural replacements first. When the preview has replacements and is not applied yet, it queues a resolve handler that contains:\n\n- label (human-readable summary)\n- `sourceToolName` (`ast_edit`)\n- `apply(reason: string)` callback that reruns AST edit with `dryRun: false`\n\n`resolve(action=\"apply\", reason=\"...\")` passes `reason` into this callback.\n\n## Custom tools: `pushPendingAction`\n\nCustom tools can register resolve-compatible pending actions through `CustomToolAPI.pushPendingAction(...)`. The custom tool loader forwards these actions to `queueResolveHandler(...)` when that hook is available.\n\n`CustomToolPendingAction`:\n\n- `label: string` (required)\n- `apply(reason: string): Promise<AgentToolResult<unknown>>` (required) — invoked on apply; `reason` is the string passed to `resolve`\n- `reject?(reason: string): Promise<AgentToolResult<unknown> | undefined>` (optional) — invoked on discard; return value replaces the default \"Discarded\" message if provided\n- `details?: unknown` exists on the public custom-tool type but is not currently forwarded by the loader into resolve metadata\n- `sourceToolName?: string` (optional, defaults to `\"custom_tool\"`)\n\n### Minimal usage example\n\n```ts\nimport type { CustomToolFactory } from \"@oh-my-pi/pi-coding-agent\";\n\nconst factory: CustomToolFactory = (pi) => ({\n  name: \"batch_rename_preview\",\n  label: \"Batch Rename Preview\",\n  description: \"Previews renames and defers commit to resolve\",\n  parameters: pi.typebox.Type.Object({\n    files: pi.typebox.Type.Array(pi.typebox.Type.String()),\n  }),\n\n  async execute(_toolCallId, params) {\n    const previewSummary = `Prepared rename plan for ${params.files.length} files`;\n\n    pi.pushPendingAction({\n      label: `Batch rename: ${params.files.length} files`,\n      sourceToolName: \"batch_rename_preview\",\n      apply: async (reason) => {\n        // apply writes here\n        return {\n          content: [\n            { type: \"text\", text: `Applied batch rename. Reason: ${reason}` },\n          ],\n        };\n      },\n      reject: async (reason) => {\n        // optional: cleanup or notify on discard\n        return {\n          content: [\n            { type: \"text\", text: `Discarded batch rename. Reason: ${reason}` },\n          ],\n        };\n      },\n    });\n\n    return {\n      content: [\n        {\n          type: \"text\",\n          text: `${previewSummary}. Call resolve to apply or discard.`,\n        },\n      ],\n    };\n  },\n});\n\nexport default factory;\n```\n\n## Runtime availability and failures\n\n`pushPendingAction` is wired by the custom tool loader through the active session's resolve queue hook.\n\nIf the runtime did not provide the resolve queue hook, `pushPendingAction` throws:\n\n- `Pending action store unavailable for custom tools in this runtime.`\n\n## Tool-choice behavior\n\nWhen `queueResolveHandler(...)` registers a preview, the agent runtime forces a one-shot `resolve` tool choice so pending previews are explicitly finalized before normal tool flow continues.\n\n## Developer guidance\n\n- Use pending actions only for destructive or high-impact operations that should support explicit apply/discard.\n- Keep `label` concise and specific; it is shown in resolve renderer output.\n- Ensure `apply(reason)` is deterministic and idempotent enough for one-shot execution; `reason` is informational and should not change behavior.\n- Implement `reject(reason)` when the discard needs cleanup (temp state, locks, notifications); omit it for stateless previews where the default message suffices.\n- If your tool can stage multiple previews, remember they are mediated by the tool-choice queue rather than a separate pending-action stack.\n",
 	"rpc.md": "# RPC Protocol Reference\n\nRPC mode runs the coding agent as a newline-delimited JSON protocol over stdio.\n\n- **stdin**: commands (`RpcCommand`), extension UI responses, and host-tool updates/results\n- **stdout**: a ready frame, command responses (`RpcResponse`), session/agent events, extension UI requests, host-tool requests/cancellations\n\nPrimary implementation:\n\n- `src/modes/rpc/rpc-mode.ts`\n- `src/modes/rpc/rpc-types.ts`\n- `src/session/agent-session.ts`\n- `packages/agent/src/agent.ts`\n- `packages/agent/src/agent-loop.ts`\n\n## Startup\n\n```bash\nomp --mode rpc [regular CLI options]\n```\n\nBehavior notes:\n\n- `@file` CLI arguments are rejected in RPC mode.\n- RPC mode disables automatic session title generation by default to avoid an extra model call.\n- RPC mode resets workflow-altering `todo.*`, `task.*`, `async.*`, and `bash.autoBackground.*` settings to their built-in defaults instead of inheriting user overrides.\n- The process reads stdin as JSONL (`readJsonl(Bun.stdin.stream())`).\n- At startup it writes `{ \"type\": \"ready\" }` before processing commands.\n- When stdin closes, pending host-tool calls are rejected and the process exits with code `0`.\n- Responses/events are written as one JSON object per line.\n\n## Transport and Framing\n\nEach frame is a single JSON object followed by `\\n`.\n\nThere is no envelope beyond the object shape itself.\n\n### Outbound frame categories (stdout)\n\n1. Ready frame (`{ type: \"ready\" }`)\n2. `RpcResponse` (`{ type: \"response\", ... }`)\n3. `AgentSessionEvent` objects (`agent_start`, `message_update`, etc.)\n4. `RpcExtensionUIRequest` (`{ type: \"extension_ui_request\", ... }`)\n5. Host tool requests/cancellations (`host_tool_call`, `host_tool_cancel`)\n6. Extension errors (`{ type: \"extension_error\", extensionPath, event, error }`)\n\n### Inbound frame categories (stdin)\n\n1. `RpcCommand`\n2. `RpcExtensionUIResponse` (`{ type: \"extension_ui_response\", ... }`)\n3. Host tool updates/results (`host_tool_update`, `host_tool_result`)\n\n## Request/Response Correlation\n\nAll commands accept optional `id?: string`.\n\n- If provided, normal command responses echo the same `id`.\n- `RpcClient` relies on this for pending-request resolution.\n\nImportant edge behavior from runtime:\n\n- Unknown command responses are emitted with `id: undefined` (even if the request had an `id`).\n- Parse/handler exceptions in the input loop emit `command: \"parse\"` with `id: undefined`.\n- `prompt` and `abort_and_prompt` return immediate success, then may emit a later error response with the **same** id if async prompt scheduling fails.\n\n## Command Schema (canonical)\n\n`RpcCommand` is defined in `src/modes/rpc/rpc-types.ts`:\n\n### Prompting\n\n- `{ id?, type: \"prompt\", message: string, images?: ImageContent[], streamingBehavior?: \"steer\" | \"followUp\" }`\n- `{ id?, type: \"steer\", message: string, images?: ImageContent[] }`\n- `{ id?, type: \"follow_up\", message: string, images?: ImageContent[] }`\n- `{ id?, type: \"abort\" }`\n- `{ id?, type: \"abort_and_prompt\", message: string, images?: ImageContent[] }`\n- `{ id?, type: \"new_session\", parentSession?: string }`\n\n### State\n\n- `{ id?, type: \"get_state\" }`\n- `{ id?, type: \"set_todos\", phases: TodoPhase[] }`\n- `{ id?, type: \"set_host_tools\", tools: RpcHostToolDefinition[] }`\n\n### Model\n\n- `{ id?, type: \"set_model\", provider: string, modelId: string }`\n- `{ id?, type: \"cycle_model\" }`\n- `{ id?, type: \"get_available_models\" }`\n\n### Thinking\n\n- `{ id?, type: \"set_thinking_level\", level: ThinkingLevel }`\n- `{ id?, type: \"cycle_thinking_level\" }`\n\n### Queue modes\n\n- `{ id?, type: \"set_steering_mode\", mode: \"all\" | \"one-at-a-time\" }`\n- `{ id?, type: \"set_follow_up_mode\", mode: \"all\" | \"one-at-a-time\" }`\n- `{ id?, type: \"set_interrupt_mode\", mode: \"immediate\" | \"wait\" }`\n\n### Compaction\n\n- `{ id?, type: \"compact\", customInstructions?: string }`\n- `{ id?, type: \"set_auto_compaction\", enabled: boolean }`\n\n### Retry\n\n- `{ id?, type: \"set_auto_retry\", enabled: boolean }`\n- `{ id?, type: \"abort_retry\" }`\n\n### Bash\n\n- `{ id?, type: \"bash\", command: string }`\n- `{ id?, type: \"abort_bash\" }`\n\n### Session\n\n- `{ id?, type: \"get_session_stats\" }`\n- `{ id?, type: \"export_html\", outputPath?: string }`\n- `{ id?, type: \"switch_session\", sessionPath: string }`\n- `{ id?, type: \"branch\", entryId: string }`\n- `{ id?, type: \"get_branch_messages\" }`\n- `{ id?, type: \"get_last_assistant_text\" }`\n- `{ id?, type: \"set_session_name\", name: string }`\n\n### Messages\n\n- `{ id?, type: \"get_messages\" }`\n\n## Response Schema\n\nAll command results use `RpcResponse`:\n\n- Success: `{ id?, type: \"response\", command: <command>, success: true, data?: ... }`\n- Failure: `{ id?, type: \"response\", command: string, success: false, error: string }`\n\nData payloads are command-specific and defined in `rpc-types.ts`.\n\n### `get_state` payload\n\n```json\n{\n  \"model\": { \"provider\": \"...\", \"id\": \"...\" },\n  \"thinkingLevel\": \"off|minimal|low|medium|high|xhigh\",\n  \"isStreaming\": false,\n  \"isCompacting\": false,\n  \"steeringMode\": \"all|one-at-a-time\",\n  \"followUpMode\": \"all|one-at-a-time\",\n  \"interruptMode\": \"immediate|wait\",\n  \"sessionFile\": \"...\",\n  \"sessionId\": \"...\",\n  \"sessionName\": \"...\",\n  \"autoCompactionEnabled\": true,\n  \"messageCount\": 0,\n  \"queuedMessageCount\": 0,\n  \"todoPhases\": [\n    {\n      \"id\": \"phase-1\",\n      \"name\": \"Todos\",\n      \"tasks\": [\n        {\n          \"id\": \"task-1\",\n          \"content\": \"Map the tool surface\",\n          \"status\": \"in_progress\"\n        }\n      ]\n    }\n  ],\n  \"systemPrompt\": \"...\",\n  \"dumpTools\": [\n    {\n      \"name\": \"read\",\n      \"description\": \"Read files and URLs\",\n      \"parameters\": {}\n    }\n  ]\n}\n```\n\n### `set_todos` payload\n\nReplaces the in-memory todo state for the current session and returns the normalized phase list:\n\n```json\n{\n  \"id\": \"req_2\",\n  \"type\": \"set_todos\",\n  \"phases\": [\n    {\n      \"id\": \"phase-1\",\n      \"name\": \"Evaluation\",\n      \"tasks\": [\n        {\n          \"id\": \"task-1\",\n          \"content\": \"Map the read tool surface\",\n          \"status\": \"in_progress\"\n        },\n        {\n          \"id\": \"task-2\",\n          \"content\": \"Exercise edit operations\",\n          \"status\": \"pending\"\n        }\n      ]\n    }\n  ]\n}\n```\n\nThis is useful for hosts that want to pre-seed a plan before the first prompt.\n\n### `set_host_tools` payload\n\nReplaces the current set of host-owned tools that the RPC server may call back\ninto over stdio:\n\n```json\n{\n  \"id\": \"req_3\",\n  \"type\": \"set_host_tools\",\n  \"tools\": [\n    {\n      \"name\": \"echo_host\",\n      \"label\": \"Echo Host\",\n      \"description\": \"Echo a value from the embedding host\",\n      \"parameters\": {\n        \"type\": \"object\",\n        \"properties\": {\n          \"message\": { \"type\": \"string\" }\n        },\n        \"required\": [\"message\"],\n        \"additionalProperties\": false\n      }\n    }\n  ]\n}\n```\n\nThe response payload is:\n\n```json\n{\n  \"toolNames\": [\"echo_host\"]\n}\n```\n\nThese tools are added to the active session tool registry before the next model\ncall. Re-sending `set_host_tools` replaces the previous host-owned set.\n\n## Event Stream Schema\n\nRPC mode forwards `AgentSessionEvent` objects from `AgentSession.subscribe(...)`.\n\nCommon event types:\n\n- `agent_start`, `agent_end`\n- `turn_start`, `turn_end`\n- `message_start`, `message_update`, `message_end`\n- `tool_execution_start`, `tool_execution_update`, `tool_execution_end`\n- `auto_compaction_start`, `auto_compaction_end`\n- `auto_retry_start`, `auto_retry_end`\n- `ttsr_triggered`\n- `todo_reminder`\n- `todo_auto_clear`\n\nExtension runner errors are emitted separately as:\n\n```json\n{\n  \"type\": \"extension_error\",\n  \"extensionPath\": \"...\",\n  \"event\": \"...\",\n  \"error\": \"...\"\n}\n```\n\n`message_update` includes streaming deltas in `assistantMessageEvent` (text/thinking/toolcall deltas).\n\n## Prompt/Queue Concurrency and Ordering\n\nThis is the most important operational behavior.\n\n### Immediate ack vs completion\n\n`prompt` and `abort_and_prompt` are **acknowledged immediately**:\n\n```json\n{ \"id\": \"req_1\", \"type\": \"response\", \"command\": \"prompt\", \"success\": true }\n```\n\nThat means:\n\n- command acceptance != run completion\n- final completion is observed via `agent_end`\n\n### While streaming\n\n`AgentSession.prompt()` requires `streamingBehavior` during active streaming:\n\n- `\"steer\"` => queued steering message (interrupt path)\n- `\"followUp\"` => queued follow-up message (post-turn path)\n\nIf omitted during streaming, prompt fails.\n\n### Queue defaults\n\nFrom `packages/agent/src/agent.ts` defaults:\n\n- `steeringMode`: `\"one-at-a-time\"`\n- `followUpMode`: `\"one-at-a-time\"`\n- `interruptMode`: `\"immediate\"`\n\n### Mode semantics\n\n- `set_steering_mode` / `set_follow_up_mode`\n  - `\"one-at-a-time\"`: dequeue one queued message per turn\n  - `\"all\"`: dequeue entire queue at once\n- `set_interrupt_mode`\n  - `\"immediate\"`: tool execution checks steering between tool calls; pending steering can abort remaining tool calls in the turn\n  - `\"wait\"`: defer steering until turn completion\n\n## Extension UI Sub-Protocol\n\nExtensions in RPC mode use request/response UI frames.\n\n### Outbound request\n\n`RpcExtensionUIRequest` (`type: \"extension_ui_request\"`) methods:\n\n- `select`, `confirm`, `input`, `editor`, `cancel`\n- `notify`, `setStatus`, `setWidget`, `setTitle`, `set_editor_text`\n\nRuntime note:\n\n- Automatic session title generation is disabled in RPC mode, and `setTitle` UI\n  requests are also suppressed by default because most hosts do not have a\n  meaningful terminal-title surface. Set `PI_RPC_EMIT_TITLE=1` to opt back in to\n  the UI event only.\n\nExample:\n\n```json\n{\n  \"type\": \"extension_ui_request\",\n  \"id\": \"123\",\n  \"method\": \"confirm\",\n  \"title\": \"Confirm\",\n  \"message\": \"Continue?\",\n  \"timeout\": 30000\n}\n```\n\n### Inbound response\n\n`RpcExtensionUIResponse` (`type: \"extension_ui_response\"`):\n\n- `{ type: \"extension_ui_response\", id: string, value: string }`\n- `{ type: \"extension_ui_response\", id: string, confirmed: boolean }`\n- `{ type: \"extension_ui_response\", id: string, cancelled: true, timedOut?: boolean }`\n\nIf a dialog has a timeout, RPC mode resolves to a default value when timeout/abort fires.\n\n## Host Tool Sub-Protocol\n\nRPC hosts can expose custom tools to the agent by sending `set_host_tools`, then\nserving execution requests over the same transport.\n\n### Outbound request\n\nWhen the agent wants the host to execute one of those tools, RPC mode emits:\n\n```json\n{\n  \"type\": \"host_tool_call\",\n  \"id\": \"host_1\",\n  \"toolCallId\": \"toolu_123\",\n  \"toolName\": \"echo_host\",\n  \"arguments\": { \"message\": \"hello\" }\n}\n```\n\nIf the tool execution is later aborted, RPC mode emits:\n\n```json\n{\n  \"type\": \"host_tool_cancel\",\n  \"id\": \"host_cancel_1\",\n  \"targetId\": \"host_1\"\n}\n```\n\n### Inbound updates and completion\n\nHosts can optionally stream progress:\n\n```json\n{\n  \"type\": \"host_tool_update\",\n  \"id\": \"host_1\",\n  \"partialResult\": {\n    \"content\": [{ \"type\": \"text\", \"text\": \"working\" }]\n  }\n}\n```\n\nCompletion uses:\n\n```json\n{\n  \"type\": \"host_tool_result\",\n  \"id\": \"host_1\",\n  \"result\": {\n    \"content\": [{ \"type\": \"text\", \"text\": \"done\" }]\n  }\n}\n```\n\nSet top-level `isError: true` on `host_tool_result` to reject the pending host tool call and surface the returned text content as a tool error.\n\n## Error Model and Recoverability\n\n### Command-level failures\n\nFailures are `success: false` with string `error`.\n\n```json\n{\n  \"id\": \"req_2\",\n  \"type\": \"response\",\n  \"command\": \"set_model\",\n  \"success\": false,\n  \"error\": \"Model not found: provider/model\"\n}\n```\n\n### Recoverability expectations\n\n- Most command failures are recoverable; process remains alive.\n- Malformed JSONL / parse-loop exceptions emit a `parse` error response and continue reading subsequent lines.\n- Empty `set_session_name` is rejected (`Session name cannot be empty`).\n- Extension UI responses with unknown `id` are ignored.\n- Process termination conditions are stdin close or explicit extension-triggered shutdown after the current command.\n\n## Compact Command Flows\n\n### 1) Prompt and stream\n\nstdin:\n\n```json\n{ \"id\": \"req_1\", \"type\": \"prompt\", \"message\": \"Summarize this repo\" }\n```\n\nstdout sequence (typical):\n\n```json\n{ \"id\": \"req_1\", \"type\": \"response\", \"command\": \"prompt\", \"success\": true }\n{ \"type\": \"agent_start\" }\n{ \"type\": \"message_update\", \"assistantMessageEvent\": { \"type\": \"text_delta\", \"delta\": \"...\" }, \"message\": { \"role\": \"assistant\", \"content\": [] } }\n{ \"type\": \"agent_end\", \"messages\": [] }\n```\n\n### 2) Prompt during streaming with explicit queue policy\n\nstdin:\n\n```json\n{\n  \"id\": \"req_2\",\n  \"type\": \"prompt\",\n  \"message\": \"Also include risks\",\n  \"streamingBehavior\": \"followUp\"\n}\n```\n\n### 3) Inspect and tune queue behavior\n\nstdin:\n\n```json\n{ \"id\": \"q1\", \"type\": \"get_state\" }\n{ \"id\": \"q2\", \"type\": \"set_steering_mode\", \"mode\": \"all\" }\n{ \"id\": \"q3\", \"type\": \"set_interrupt_mode\", \"mode\": \"wait\" }\n```\n\n### 4) Extension UI round trip\n\nstdout:\n\n```json\n{\n  \"type\": \"extension_ui_request\",\n  \"id\": \"ui_7\",\n  \"method\": \"input\",\n  \"title\": \"Branch name\",\n  \"placeholder\": \"feature/...\"\n}\n```\n\nstdin:\n\n```json\n{ \"type\": \"extension_ui_response\", \"id\": \"ui_7\", \"value\": \"feature/rpc-host\" }\n```\n\n## Notes on `RpcClient` helper\n\n`src/modes/rpc/rpc-client.ts` is a convenience wrapper, not the protocol definition.\n\nCurrent helper characteristics:\n\n- Spawns `bun <cliPath> --mode rpc`\n- Correlates responses by generated `req_<n>` ids\n- Dispatches only recognized `AgentEvent` types to listeners\n- Supports host-owned custom tools via `setCustomTools()` and automatic handling of `host_tool_call` / `host_tool_cancel`\n- Does **not** expose helper methods for every protocol command (for example, `set_interrupt_mode` and `set_session_name` are in protocol types but not wrapped as dedicated methods)\n\nUse raw protocol frames if you need complete surface coverage.\n",
 	"rulebook-matching-pipeline.md": "# Rulebook Matching Pipeline\n\nThis document describes how coding-agent discovers rules from supported config formats, normalizes them into a single `Rule` shape, resolves precedence conflicts, and splits the result into:\n\n- **Rulebook rules** (available to the model via system prompt + `rule://` URLs)\n- **TTSR rules** (time-travel stream interruption rules)\n\nIt reflects the current implementation, including partial semantics and metadata that is parsed but not enforced.\n\n## Implementation files\n\n- [`../src/capability/rule.ts`](../packages/coding-agent/src/capability/rule.ts)\n- [`../src/capability/index.ts`](../packages/coding-agent/src/capability/index.ts)\n- [`../src/discovery/index.ts`](../packages/coding-agent/src/discovery/index.ts)\n- [`../src/discovery/helpers.ts`](../packages/coding-agent/src/discovery/helpers.ts)\n- [`../src/discovery/builtin.ts`](../packages/coding-agent/src/discovery/builtin.ts)\n- [`../src/discovery/cursor.ts`](../packages/coding-agent/src/discovery/cursor.ts)\n- [`../src/discovery/windsurf.ts`](../packages/coding-agent/src/discovery/windsurf.ts)\n- [`../src/discovery/cline.ts`](../packages/coding-agent/src/discovery/cline.ts)\n- [`../src/sdk.ts`](../packages/coding-agent/src/sdk.ts)\n- [`../src/system-prompt.ts`](../packages/coding-agent/src/system-prompt.ts)\n- [`../src/internal-urls/rule-protocol.ts`](../packages/coding-agent/src/internal-urls/rule-protocol.ts)\n- [`../src/utils/frontmatter.ts`](../packages/coding-agent/src/utils/frontmatter.ts)\n\n## 1. Canonical rule shape\n\nAll providers normalize source files into `Rule`:\n\n```ts\ninterface Rule {\n  name: string;\n  path: string;\n  content: string;\n  globs?: string[];\n  alwaysApply?: boolean;\n  description?: string;\n  condition?: string[];\n  scope?: string[];\n  interruptMode?: \"never\" | \"prose-only\" | \"tool-only\" | \"always\";\n  _source: SourceMeta;\n}\n```\n\nCapability identity is `rule.name` (`ruleCapability.key = rule => rule.name`).\n\nConsequence: precedence and deduplication are **name-based only**. Two different files with the same `name` are considered the same logical rule.\n\n## 2. Discovery sources and normalization\n\n`src/discovery/index.ts` auto-registers providers. For `rules`, current providers are:\n\n- `native` (priority `100`)\n- `cursor` (priority `50`)\n- `windsurf` (priority `50`)\n- `cline` (priority `40`)\n\n### Native provider (`builtin.ts`)\n\nLoads `.omp` rules from:\n\n- project: `<cwd>/.omp/rules/*.{md,mdc}`\n- user: `~/.omp/agent/rules/*.{md,mdc}`\n\nNormalization:\n\n- `name` = filename without `.md`/`.mdc`\n- frontmatter parsed via `parseFrontmatter`\n- `content` = body (frontmatter stripped)\n- `globs`, `alwaysApply`, `description`, `condition`/legacy `ttsr_trigger`, `scope`, and `interruptMode` are parsed by `buildRuleFromMarkdown`\n\nImportant caveat: `condition` values that look like file globs are converted into `tool:edit(...)` / `tool:write(...)` scope shorthands with catch-all condition `.*`.\n\n### Cursor provider (`cursor.ts`)\n\nLoads from:\n\n- user: `~/.cursor/rules/*.{mdc,md}`\n- project: `<cwd>/.cursor/rules/*.{mdc,md}`\n\nNormalization (`transformMDCRule`):\n\n- `description`: kept only if string\n- `alwaysApply`: only `true` is preserved (`false` becomes `undefined`)\n- `globs`: accepts array (string elements only) or single string\n- `condition`/legacy `ttsr_trigger`, `scope`, and `interruptMode` are parsed by shared rule helpers\n- `name` from filename without extension\n\n### Windsurf provider (`windsurf.ts`)\n\nLoads from:\n\n- user: `~/.codeium/windsurf/memories/global_rules.md` (fixed rule name `global_rules`)\n- project: `<cwd>/.windsurf/rules/*.md`\n\nNormalization:\n\n- `globs`: array-of-string or single string\n- `alwaysApply`, `description`, `condition`/legacy `ttsr_trigger`, `scope`, and `interruptMode` parsed by shared rule helpers\n- `name` is fixed to `global_rules` for the user global file and derived from filename for project rules\n\n### Cline provider (`cline.ts`)\n\nSearches upward from `cwd` for nearest `.clinerules`:\n\n- if directory: loads `*.md` inside it\n- if file: loads single file as rule named `clinerules`\n\nNormalization:\n\n- `globs`: array-of-string or single string\n- `alwaysApply`, `description`, `condition`/legacy `ttsr_trigger`, `scope`, and `interruptMode` parsed by shared rule helpers\n- `name` is fixed to `clinerules` for a `.clinerules` file and derived from filename for `.clinerules/*.md`\n\n## 3. Frontmatter parsing behavior and ambiguity\n\nAll providers use `parseFrontmatter` (`utils/frontmatter.ts`) with these semantics:\n\n1. Frontmatter is parsed only when content starts with `---` and has a closing `\\n---`.\n2. Body is trimmed after frontmatter extraction.\n3. If YAML parse fails:\n   - warning is logged,\n   - parser falls back to simple `key: value` line parsing (`^(\\w+):\\s*(.*)$`).\n\nAmbiguity consequences:\n\n- Fallback parser does not support arrays, nested objects, quoting rules, or hyphenated keys.\n- Fallback values become strings (for example `alwaysApply: true` becomes string `\"true\"`), so providers requiring boolean/string types may drop metadata.\n- `ttsr_trigger` works in fallback (underscore key); keys like `thinking-level` would not.\n- Files without valid frontmatter still load as rules with empty metadata and full content body.\n\n## 4. Provider precedence and deduplication\n\n`loadCapability(\"rules\")` (`capability/index.ts`) merges provider outputs and then deduplicates by `rule.name`.\n\n### Precedence model\n\n- Providers are ordered by priority descending.\n- Equal priority keeps registration order (`cursor` before `windsurf` from `discovery/index.ts`).\n- Dedup is first-wins: first encountered rule name is kept; later same-name items are marked `_shadowed` in `all` and excluded from `items`.\n\nEffective rule provider order is currently:\n\n1. `native` (100)\n2. `cursor` (50)\n3. `windsurf` (50)\n4. `cline` (40)\n\n### Intra-provider ordering caveat\n\nWithin a provider, item order comes from `loadFilesFromDir` glob result ordering plus explicit push order. This is deterministic enough for normal use but not explicitly sorted in code.\n\nNotable source-order differences:\n\n- `native` appends project then user config dirs.\n- `cursor` appends user then project results.\n- `windsurf` appends user `global_rules` first, then project rules.\n- `cline` loads only nearest `.clinerules` source.\n\n## 5. Split into Rulebook, Always-Apply, and TTSR buckets\n\nAfter rule discovery in `createAgentSession` (`sdk.ts`):\n\n1. All discovered rules are scanned.\n2. Rules with `condition` entries are registered into `TtsrManager`; legacy `ttsr_trigger` / `ttsrTrigger` are accepted during rule parsing as condition fallbacks.\n3. A separate `rulebookRules` list is built with this predicate:\n\n```ts\n!isTtsrRule && rule.alwaysApply !== true && !!rule.description;\n```\n\n4. An `alwaysApplyRules` list is built:\n\n```ts\n!isTtsrRule && rule.alwaysApply === true;\n```\n\n### Bucket behavior\n\n- **TTSR bucket**: any rule with a non-empty parsed `condition` that `TtsrManager.addRule(...)` accepts. Takes priority over other buckets.\n- **Always-apply bucket**: `alwaysApply === true`, not TTSR. Full content injected into system prompt. Resolvable via `rule://`.\n- **Rulebook bucket**: must have description, must not be TTSR, must not be `alwaysApply`. Listed in system prompt by name+description; content read on demand via `rule://`.\n- A rule with both `condition` and `alwaysApply` goes to TTSR only (TTSR takes priority).\n- A rule with both `alwaysApply` and `description` goes to always-apply only (not rulebook).\n\n## 6. How metadata affects runtime surfaces\n\n### `description`\n\n- Required for inclusion in rulebook.\n- Rendered in system prompt `<rules>` block.\n- Missing description means rule is not available via `rule://` and not listed in system prompt rules.\n\n### `globs`\n\n- Carried through on `Rule`.\n- Rendered as `<glob>...</glob>` entries in the system prompt rules block.\n- Exposed in rules UI state (`extensions` mode list).\n- **Not enforced for automatic matching in this pipeline.** There is no runtime glob matcher selecting rules by current file/tool target.\n\n### `alwaysApply`\n\n- Parsed and preserved by providers.\n- Used in UI display (`\"always\"` trigger label in extensions state manager).\n- Used as an exclusion condition from `rulebookRules`.\n- **Full rule content is auto-injected into the system prompt** (before the rulebook rules section).\n- Rule is also addressable via `rule://<name>` for re-reading.\n\n### `condition`, `scope`, and `interruptMode`\n\n- `condition` is the current TTSR trigger field; legacy `ttsr_trigger` / `ttsrTrigger` are accepted as fallback inputs during parsing.\n- `scope` narrows TTSR matching scope. A condition token that looks like a file glob becomes `tool:edit(<glob>)` and `tool:write(<glob>)` scope entries plus catch-all condition `.*`.\n- `interruptMode` can override the global TTSR interrupt mode for the rule.\n\n## 7. System prompt inclusion path\n\n`buildSystemPromptInternal` receives both `rules` (rulebook) and `alwaysApplyRules`.\n\nAlways-apply rules are rendered first, injecting their raw content directly into the prompt.\n\nRulebook rules are rendered in a `# Rules` section with:\n\n- `Read rule://<name> when working in matching domain`\n- Each rule's `name`, `description`, and optional `<glob>` list\n\nThis is advisory/contextual: prompt text asks the model to read applicable rules, but code does not enforce glob applicability.\n\n## 8. `rule://` internal URL behavior\n\n`RuleProtocolHandler` is registered with:\n\n```ts\nnew RuleProtocolHandler({\n  getRules: () => [...rulebookRules, ...alwaysApplyRules],\n});\n```\n\nImplications:\n\n- `rule://<name>` resolves against both **rulebookRules** and **alwaysApplyRules**.\n- TTSR-only rules and rules with no description and no `alwaysApply` are not addressable via `rule://`.\n- Resolution is exact name match.\n- Unknown names return error listing available rule names.\n- Returned content is raw `rule.content` (frontmatter stripped), content type `text/markdown`.\n\n## 9. Known partial / non-enforced semantics\n\n1. Provider descriptions mention legacy files (`.cursorrules`, `.windsurfrules`), but current loader code paths do not actually read those files.\n2. `globs` metadata is surfaced to prompt/UI but not enforced by rule selection logic.\n3. Rule selection for `rule://` includes rulebook and always-apply rules, but not TTSR-only rules.\n4. Discovery warnings (`loadCapability(\"rules\").warnings`) are produced but `createAgentSession` does not currently surface/log them in this path.\n",

package/src/modes/components/model-selector.ts

@@ -561,6 +561,21 @@ export class ModelSelectorComponent extends Container {
 		return `${ageMinutes}m ago`;
 	}
 
+	#formatDiscoveryErrorHint(error: string | undefined): string | undefined {
+		if (!error) {
+			return undefined;
+		}
+		const httpMatch = error.match(/^HTTP (\d+) from (.+)$/);
+		if (!httpMatch) {
+			return undefined;
+		}
+		const [, statusCode, url] = httpMatch;
+		if (statusCode === "404") {
+			return `  Discovery endpoint ${url} returned 404. Point baseUrl at the host that serves /models (usually .../v1).`;
+		}
+		return `  Discovery failed: ${error}`;
+	}
+
 	#getProviderEmptyStateMessage(): string | undefined {
 		const activeProvider = this.#getActiveProvider();
 		if (activeProvider === ALL_TAB || activeProvider === CANONICAL_TAB || this.#searchInput.getValue().trim()) {
@@ -577,13 +592,18 @@ export class ModelSelectorComponent extends Container {
 					? `  Using cached model list from ${age}. Live refresh is still pending.`
 					: "  Using cached model list. Live refresh is still pending.";
 			case "unavailable":
-				return age ? `  Provider unavailable. Using cached model list from ${age}.` : "  Provider unavailable.";
+				return (
+					this.#formatDiscoveryErrorHint(state.error) ??
+					(age ? `  Provider unavailable. Using cached model list from ${age}.` : "  Provider unavailable.")
+				);
 			case "unauthenticated":
 				return "  Provider requires authentication before models can be discovered.";
 			case "idle":
 				return "  Provider has not been refreshed yet.";
+			case "empty":
+				return "  Discovery succeeded but returned 0 models. Check that /models returns { data: [{ id }] }.";
 			case "ok":
-				return "  Provider reported no models.";
+				return undefined;
 		}
 	}
 

package/src/modes/components/status-line/segments.ts

@@ -344,7 +344,7 @@ const cacheReadSegment: StatusLineSegment = {
 		const { cacheRead } = ctx.usageStats;
 		if (!cacheRead) return { content: "", visible: false };
 
-		const parts = [theme.icon.cache, theme.icon.input, formatNumber(cacheRead)].filter(Boolean);
+		const parts = [theme.icon.cache, theme.icon.output, formatNumber(cacheRead)].filter(Boolean);
 		const content = parts.join(" ");
 		return { content: theme.fg("statusLineSpend", content), visible: true };
 	},
@@ -356,7 +356,7 @@ const cacheWriteSegment: StatusLineSegment = {
 		const { cacheWrite } = ctx.usageStats;
 		if (!cacheWrite) return { content: "", visible: false };
 
-		const parts = [theme.icon.cache, theme.icon.output, formatNumber(cacheWrite)].filter(Boolean);
+		const parts = [theme.icon.cache, theme.icon.input, formatNumber(cacheWrite)].filter(Boolean);
 		const content = parts.join(" ");
 		return { content: theme.fg("statusLineOutput", content), visible: true };
 	},

package/src/modes/components/status-line-segment-editor.ts

@@ -34,8 +34,8 @@ const SEGMENT_INFO: Record<StatusLineSegmentId, { label: string; short: string }
 	time: { label: "Clock", short: "current time" },
 	session: { label: "Session", short: "session ID" },
 	hostname: { label: "Host", short: "hostname" },
-	cache_read: { label: "Cache ↓", short: "cache read" },
-	cache_write: { label: "Cache ↑", short: "cache write" },
+	cache_read: { label: "Cache ↑", short: "cache read" },
+	cache_write: { label: "Cache ↓", short: "cache write" },
 	session_name: { label: "Session Name", short: "named session" },
 };
 

package/src/modes/controllers/mcp-command-controller.ts

@@ -1114,24 +1114,16 @@ export class MCPCommandController {
 
 		let connection: MCPServerConnection | undefined;
 		try {
-			const cwd = getProjectDir();
-			const userPath = getMCPConfigPath("user", cwd);
-			const projectPath = getMCPConfigPath("project", cwd);
-
-			// Find the server config
-			const [userConfig, projectConfig] = await Promise.all([
-				readMCPConfigFile(userPath),
-				readMCPConfigFile(projectPath),
-			]);
-
-			const config = userConfig.mcpServers?.[name] ?? projectConfig.mcpServers?.[name];
+			const found = await this.#findConfiguredServer(name);
 
-			if (!config) {
+			if (!found) {
 				this.ctx.showError(
 					`Server "${name}" not found.\n\nTip: Run ${theme.fg("accent", "/mcp list")} to see available servers.`,
 				);
 				return;
 			}
+
+			const { config } = found;
 			if (config.enabled === false) {
 				this.ctx.showError(`Server "${name}" is disabled. Run /mcp enable ${name} first.`);
 				return;

package/src/prompts/tools/eval.md

@@ -20,7 +20,7 @@ At least 5 equal signs on each side. Content between one header and the next (or
 - Pass multiple small cells in one call.
 - Define small reusable functions for individual debugging.
 - Put workflow explanations in the assistant message or cell title — never inside cell code.
-
+{{#if py}}- Python cells run inside an IPython kernel with a live event loop. Use top-level `await` directly (e.g. `await main()`); `asyncio.run(…)` raises "cannot be called from a running event loop".{{/if}}
 **On failure:** errors identify the failing cell (e.g., "Cell 3 failed"). Resubmit only the fixed cell (or fixed cell + remaining cells).
 </instruction>
 

package/src/prompts/tools/hashline.md

@@ -23,6 +23,21 @@ Purely textual format. The tool has NO awareness of language, indentation, brack
 - Pick the smallest op for the change: pure addition → `+`/`<`; pure deletion → `-`; `= A..B` ONLY when content inside `A..B` is actually being modified or removed.
 </rules>
 
+<brace-shapes>
+When your edit involves brace boundaries (`{` / `}`), prefer these shapes:
+- **Whole block replace/delete**: pick the range so it spans both halves of the brace pair — start on the line that ends with `{`, end on the matching `}`. For pure removal use `-` with empty payload; for replacement, the payload's first line ends with `{` and last line is the matching `}`.
+- **Signature-only edit**: if you are only changing the line that ends with `{` (function signature, control statement, etc.), use a one-line `=` on that opener; the body and matching `}` are untouched and stay outside the range.
+- **Insert inside a block**: anchor on the opener (`+ ANCHOR` after the `{` line) or just above the closer (`+ ANCHOR` after the last interior line); emit only the new interior lines. Do not include the surrounding `{` or `}` in the payload — they're already there.
+- **Range ending on `}`**: only end on `}` when that `}` is itself part of what you're changing. The line at B+1 should be blank, an opener (next block), or a signature — not another `}`. Otherwise extend B past the closer or stop one line earlier.
+</brace-shapes>
+
+<common-failures>
+- **Do not replay the line past your range.** For `= A..B`, never end the payload with content that already exists at B+1. Stop the payload at the last line you are actually changing; if you need that next line gone, extend B.
+- **Do not duplicate chunks inside one payload.** When emitting a long `=` payload, never paste the same multi-line block twice. If you catch yourself re-emitting an earlier run of lines, stop and rewrite the op.
+- **Anchor only inside the visible region.** If the read output around your `=`/`-` end anchor is truncated (you cannot see the line at B+1), issue a fresh `read` before editing — anchoring blind drops or duplicates the boundary line.
+- **Prefer narrow ops over wide `=`.** A `+`/`<` insert plus a small `-` delete is almost always clearer and safer than a single wide `= A..B` that re-emits unchanged context.
+</common-failures>
+
 <case file="a.ts">
 {{hline 1 "const DEF = \"guest\";"}}
 {{hline 2 ""}}
@@ -73,16 +88,15 @@ Purely textual format. The tool has NO awareness of language, indentation, brack
 </examples>
 
 <anti-pattern>
-# WRONG — replaces 6 lines just to add one. Use `+` at the boundary instead.
+# WRONG — replaces 5 lines just to add one. Use `+` at the boundary instead.
 @a.ts
-= {{hrefr 1}}..{{hrefr 6}}
+= {{hrefr 1}}..{{hrefr 5}}
 {{hsep}}const DEF = "guest";
 {{hsep}}const DEBUG = false;
 {{hsep}}
 {{hsep}}export function label(name) {
 {{hsep}}	const clean = name || DEF;
 {{hsep}}	return clean.trim();
-{{hsep}}}
 
 # RIGHT — same effect, one-line insert
 @a.ts

package/src/tools/eval.ts

@@ -218,6 +218,16 @@ export class EvalTool implements AgentTool<typeof evalSchema> {
 	readonly parameters = evalSchema;
 	readonly concurrency = "exclusive";
 	readonly strict = true;
+	readonly intent = (args: Partial<Static<typeof evalSchema>>): string | undefined => {
+		const input = args.input;
+		if (input) {
+			try {
+				const cells = parseEvalInput(input).cells;
+				return cells.map(cell => cell.title || `running ${cell.language}`).join("\n");
+			} catch {}
+		}
+		return "evaluating";
+	};
 
 	get customFormat(): { syntax: "lark"; definition: string } {
 		return { syntax: "lark", definition: evalGrammar };

package/src/tools/fs-cache-invalidation.ts

@@ -17,8 +17,8 @@ export function invalidateFsScanAfterDelete(path: string): void {
 /**
  * Invalidate shared filesystem scan caches after a rename/move.
  *
- * Both source and destination paths must be invalidated because cached roots can
- * include either side of the move.
+ * Some watchers care about the disappearance at the old path; others about the
+ * appearance at the new one. Bust both to keep callers honest.
  */
 export function invalidateFsScanAfterRename(oldPath: string, newPath: string): void {
 	invalidateFsScanCache(oldPath);

package/src/tools/index.ts

@@ -230,6 +230,12 @@ export interface ToolSession {
 	/** Set or clear active checkpoint state. */
 	setCheckpointState?: (state: CheckpointState | null) => void;
 
+	/** Per-session cache of file contents as last shown to the model by
+	 *  `read`/`search`. Used by hashline anchor-stale recovery to reconstruct
+	 *  the version the model authored anchors against when the file changed
+	 *  out-of-band. Lazily initialized by `getFileReadCache`. */
+	fileReadCache?: import("../edit/file-read-cache").FileReadCache;
+
 	/** Queue a hidden message to be injected at the next agent turn. */
 	queueDeferredMessage?(message: CustomMessage): void;
 }

package/src/tools/read.ts

@@ -8,6 +8,7 @@ import type { Component } from "@oh-my-pi/pi-tui";
 import { Text } from "@oh-my-pi/pi-tui";
 import { getRemoteDir, prompt, readImageMetadata, untilAborted } from "@oh-my-pi/pi-utils";
 import { type Static, Type } from "@sinclair/typebox";
+import { getFileReadCache } from "../edit/file-read-cache";
 import { formatHashLine, formatHashLines, formatLineHash, HL_BODY_SEP } from "../edit/line-hash";
 import { isNotebookPath, readEditableNotebookText } from "../edit/notebook";
 import type { RenderResultOptions } from "../extensibility/custom-tools/types";
@@ -163,6 +164,33 @@ function countTextLines(text: string): number {
 }
 const READ_CHUNK_SIZE = 8 * 1024;
 
+/**
+ * Number of unanchored context lines to include before/after a user-requested
+ * range. Anchor-stale failures are heavily concentrated on edits whose anchors
+ * land just outside the most recent read window — a few lines of pre-anchored
+ * context covers off-by-one anchor selection without much cost.
+ */
+const RANGE_CONTEXT_LINES = 3;
+
+/**
+ * Expand a [start, end) range with ±RANGE_CONTEXT_LINES context lines on the
+ * sides where the user actually constrained the range. A start of 0 (no
+ * explicit offset) does not get leading context — that's already an open-ended
+ * read from the top.
+ */
+function expandRangeWithContext(
+	requestedStart: number,
+	requestedEnd: number,
+	totalLines: number,
+	expandStart: boolean,
+	expandEnd: boolean,
+): { startLine: number; endLine: number } {
+	return {
+		startLine: expandStart ? Math.max(0, requestedStart - RANGE_CONTEXT_LINES) : requestedStart,
+		endLine: expandEnd ? Math.min(totalLines, requestedEnd + RANGE_CONTEXT_LINES) : requestedEnd,
+	};
+}
+
 async function streamLinesFromFile(
 	filePath: string,
 	startLine: number,
@@ -645,9 +673,26 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 		const details = options.details ?? {};
 		const allLines = text.split("\n");
 		const totalLines = allLines.length;
-		const startLine = offset ? Math.max(0, offset - 1) : 0;
-		const startLineDisplay = startLine + 1;
+		// User-requested 0-indexed range start. Lines BEFORE this are leading
+		// context (added below if offset is explicit).
+		const requestedStart = offset ? Math.max(0, offset - 1) : 0;
 		const ignoreResultLimits = options.ignoreResultLimits ?? false;
+		const requestedEnd =
+			limit !== undefined && !ignoreResultLimits
+				? Math.min(requestedStart + limit, allLines.length)
+				: allLines.length;
+		// Expand only on sides the user actually constrained: leading context
+		// when offset>1, trailing context when a finite limit was set.
+		const expanded = expandRangeWithContext(
+			requestedStart,
+			requestedEnd,
+			allLines.length,
+			offset !== undefined && offset > 1,
+			limit !== undefined && !ignoreResultLimits,
+		);
+		const startLine = expanded.startLine;
+		const endLineExpanded = expanded.endLine;
+		const startLineDisplay = startLine + 1;
 
 		const resultBuilder = toolResult(details);
 		if (options.sourcePath) {
@@ -660,20 +705,19 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 			resultBuilder.sourceInternal(options.sourceInternal);
 		}
 
-		if (startLine >= allLines.length) {
+		if (requestedStart >= allLines.length) {
 			const suggestion =
 				allLines.length === 0
 					? `The ${options.entityLabel} is empty.`
 					: `Use :1 to read from the start, or :${allLines.length} to read the last line.`;
 			return resultBuilder
 				.text(
-					`Line ${startLineDisplay} is beyond end of ${options.entityLabel} (${allLines.length} lines total). ${suggestion}`,
+					`Line ${requestedStart + 1} is beyond end of ${options.entityLabel} (${allLines.length} lines total). ${suggestion}`,
 				)
 				.done();
 		}
 
-		const endLine =
-			limit !== undefined && !ignoreResultLimits ? Math.min(startLine + limit, allLines.length) : allLines.length;
+		const endLine = endLineExpanded;
 		const selectedContent = allLines.slice(startLine, endLine).join("\n");
 		const userLimitedLines = limit !== undefined && !ignoreResultLimits ? endLine - startLine : undefined;
 		const truncation = ignoreResultLimits ? noTruncResult(selectedContent) : truncateHead(selectedContent);
@@ -1318,13 +1362,20 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 			if (!content) {
 				// Raw text or line-range mode
 				const { offset, limit } = selToOffsetLimit(parsed);
-				const startLine = offset ? Math.max(0, offset - 1) : 0;
+				// User-requested 0-indexed range start. Lines BEFORE this become
+				// leading context (added below if offset is explicit).
+				const requestedStart = offset ? Math.max(0, offset - 1) : 0;
+				const expandStart = offset !== undefined && offset > 1;
+				const expandEnd = limit !== undefined;
+				const leadingContext = expandStart ? Math.min(requestedStart, RANGE_CONTEXT_LINES) : 0;
+				const trailingContext = expandEnd ? RANGE_CONTEXT_LINES : 0;
+				const startLine = requestedStart - leadingContext;
 				const startLineDisplay = startLine + 1;
 
 				const DEFAULT_LIMIT = this.#defaultLimit;
 				const effectiveLimit = limit ?? DEFAULT_LIMIT;
-				const maxLinesToCollect = Math.min(effectiveLimit, DEFAULT_MAX_LINES);
-				const selectedLineLimit = effectiveLimit;
+				const maxLinesToCollect = Math.min(effectiveLimit + leadingContext + trailingContext, DEFAULT_MAX_LINES);
+				const selectedLineLimit = effectiveLimit + leadingContext + trailingContext;
 				// Scale byte budget with line limit so the configured line count actually fits.
 				// Assume ~512 bytes/line average; never go below the shared default.
 				const maxBytesForRead = Math.max(DEFAULT_MAX_BYTES, maxLinesToCollect * 512);
@@ -1348,13 +1399,15 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 				} = streamResult;
 
 				// Check if offset is out of bounds - return graceful message instead of throwing
-				if (startLine >= totalFileLines) {
+				if (requestedStart >= totalFileLines) {
 					const suggestion =
 						totalFileLines === 0
 							? "The file is empty."
 							: `Use :1 to read from the start, or :${totalFileLines} to read the last line.`;
 					return toolResult<ReadToolDetails>({ resolvedPath: absolutePath, suffixResolution })
-						.text(`Line ${startLineDisplay} is beyond end of file (${totalFileLines} lines total). ${suggestion}`)
+						.text(
+							`Line ${requestedStart + 1} is beyond end of file (${totalFileLines} lines total). ${suggestion}`,
+						)
 						.done();
 				}
 
@@ -1378,6 +1431,10 @@ export class ReadTool implements AgentTool<typeof readSchema, ReadToolDetails> {
 					firstLineExceedsLimit,
 				};
 
+				if (collectedLines.length > 0 && !firstLineExceedsLimit) {
+					getFileReadCache(this.session).recordContiguous(absolutePath, startLineDisplay, collectedLines);
+				}
+
 				const isRawMode = parsed.kind === "raw";
 				const shouldAddHashLines = !isRawMode && displayMode.hashLines;
 				const shouldAddLineNumbers = isRawMode ? false : shouldAddHashLines ? false : displayMode.lineNumbers;

package/src/tools/search.ts

@@ -6,6 +6,7 @@ import type { Component } from "@oh-my-pi/pi-tui";
 import { Text } from "@oh-my-pi/pi-tui";
 import { prompt, untilAborted } from "@oh-my-pi/pi-utils";
 import { type Static, Type } from "@sinclair/typebox";
+import { getFileReadCache } from "../edit/file-read-cache";
 import type { RenderResultOptions } from "../extensibility/custom-tools/types";
 import type { Theme } from "../modes/theme/theme";
 import searchDescription from "../prompts/tools/search.md" with { type: "text" };
@@ -345,27 +346,32 @@ export class SearchTool implements AgentTool<typeof searchSchema, SearchToolDeta
 					}
 					return nextWidth;
 				}, 0);
+				const cacheEntries: Array<readonly [number, string]> = [];
 				for (const match of fileMatches) {
-					const pushLine = (lineNumber: number, line: string, isMatch: boolean) => {
+					const pushLine = (lineNumber: number, line: string, isMatch: boolean, recordable: boolean) => {
 						modelOut.push(formatMatchLine(lineNumber, line, isMatch, { useHashLines }));
 						displayOut.push(formatCodeFrameLine(isMatch ? "*" : " ", lineNumber, line, lineNumberWidth));
+						if (recordable) cacheEntries.push([lineNumber, line] as const);
 					};
 					if (match.contextBefore) {
 						for (const ctx of match.contextBefore) {
-							pushLine(ctx.lineNumber, ctx.line, false);
+							pushLine(ctx.lineNumber, ctx.line, false, true);
 						}
 					}
-					pushLine(match.lineNumber, match.line, true);
+					pushLine(match.lineNumber, match.line, true, !match.truncated);
 					if (match.truncated) {
 						linesTruncated = true;
 					}
 					if (match.contextAfter) {
 						for (const ctx of match.contextAfter) {
-							pushLine(ctx.lineNumber, ctx.line, false);
+							pushLine(ctx.lineNumber, ctx.line, false, true);
 						}
 					}
 					fileMatchCounts.set(relativePath, (fileMatchCounts.get(relativePath) ?? 0) + 1);
 				}
+				if (cacheEntries.length > 0) {
+					getFileReadCache(this.session).recordSparse(path.resolve(searchPath, relativePath), cacheEntries);
+				}
 				return { model: modelOut, display: displayOut };
 			};
 			if (isDirectory) {