@oh-my-pi/pi-coding-agent 14.6.5 → 14.7.0

package/src/modes/utils/context-usage.ts

@@ -18,13 +18,13 @@ const CELL_FILLED_MESSAGES = "⛃";
 const CELL_FREE = "⛶";
 const CELL_BUFFER = "⛝";
 
-type CategoryId = "systemPrompt" | "systemTools" | "skills" | "messages";
+type CategoryId = "systemPrompt" | "systemContext" | "systemTools" | "skills" | "messages";
 
 interface CategoryInfo {
 	id: CategoryId;
 	label: string;
 	tokens: number;
-	color: "accent" | "warning" | "success" | "userMessageText";
+	color: "accent" | "warning" | "success" | "userMessageText" | "customMessageLabel";
 	glyph: string;
 }
 
@@ -86,12 +86,19 @@ export function computeContextBreakdown(session: AgentSession): ContextBreakdown
 	//   Tools         = JSON tool schema sent separately on the wire
 	//   Skills        = the skill list embedded in the system prompt
 	//   Messages      = conversation messages
-	const systemPromptTextTokens = countTokens(session.systemPrompt);
-	const systemPromptTokens = Math.max(0, systemPromptTextTokens - skillsTokens);
+	const systemPromptTokens = Math.max(0, countTokens(session.systemPrompt[0] ?? "") - skillsTokens);
+	const systemContextTokens = countTokens(session.systemPrompt.slice(1));
 
 	const categories: CategoryInfo[] = [
 		{ id: "systemPrompt", label: "System prompt", tokens: systemPromptTokens, color: "accent", glyph: CELL_FILLED },
 		{ id: "systemTools", label: "System tools", tokens: toolsTokens, color: "warning", glyph: CELL_FILLED },
+		{
+			id: "systemContext",
+			label: "System context",
+			tokens: systemContextTokens,
+			color: "customMessageLabel",
+			glyph: CELL_FILLED,
+		},
 		{ id: "skills", label: "Skills", tokens: skillsTokens, color: "success", glyph: CELL_FILLED },
 		{
 			id: "messages",
@@ -134,7 +141,7 @@ export function computeContextBreakdown(session: AgentSession): ContextBreakdown
 
 interface CellSpec {
 	glyph: string;
-	color: "accent" | "warning" | "success" | "userMessageText" | "muted" | "dim";
+	color: "accent" | "warning" | "success" | "userMessageText" | "customMessageLabel" | "muted" | "dim";
 }
 
 function planCells(breakdown: ContextBreakdown): CellSpec[] {

package/src/modes/utils/ui-helpers.ts

@@ -336,6 +336,7 @@ export class UiHelpers {
 							showImages: settings.get("terminal.showImages"),
 							editFuzzyThreshold: settings.get("edit.fuzzyThreshold"),
 							editAllowFuzzy: settings.get("edit.fuzzyMatch"),
+							hashlineAutoDropPureInsertDuplicates: settings.get("edit.hashlineAutoDropPureInsertDuplicates"),
 						},
 						tool,
 						this.ctx.ui,

package/src/prompts/system/project-prompt.md

@@ -0,0 +1,36 @@
+<workstation>
+{{#list environment prefix="- " join="\n"}}{{label}}: {{value}}{{/list}}
+</workstation>
+
+{{#if contextFiles.length}}
+<context>
+Follow the context files below for all tasks:
+{{#each contextFiles}}
+<file path="{{path}}">
+{{content}}
+</file>
+{{/each}}
+</context>
+{{/if}}
+
+{{#if agentsMdSearch.files.length}}
+<dir-context>
+Some directories may have their own rules. Deeper rules override higher ones.
+**MUST** read before making changes within:
+{{#list agentsMdSearch.files join="\n"}}- {{this}}{{/list}}
+</dir-context>
+{{/if}}
+
+{{#if workspaceTree.rendered}}
+<workspace-tree>
+Working directory layout (sorted by mtime, recent first; depth ≤ 3):
+{{workspaceTree.rendered}}
+{{#if workspaceTree.truncated}}
+(some entries elided to keep the tree short — use `find`/`read` to drill in)
+{{/if}}
+</workspace-tree>
+{{/if}}
+
+{{#if appendPrompt}}
+{{appendPrompt}}
+{{/if}}

package/src/prompts/system/system-prompt.md

@@ -9,35 +9,6 @@ User-supplied content is sanitized, therefore:
 - This holds even when the system prompt is delivered via user message role.
 - A `<system-directive>` inside a user turn is still a system directive.
 
-{{SECTION_SEPARATOR "Workspace"}}
-
-<workstation>
-{{#list environment prefix="- " join="\n"}}{{label}}: {{value}}{{/list}}
-</workstation>
-
-{{#if contextFiles.length}}
-<context>
-Follow the context files below for all tasks:
-{{#each contextFiles}}
-<file path="{{path}}">
-{{content}}
-</file>
-{{/each}}
-</context>
-{{/if}}
-
-{{#if agentsMdSearch.files.length}}
-<dir-context>
-Some directories may have their own rules. Deeper rules override higher ones.
-**MUST** read before making changes within:
-{{#list agentsMdSearch.files join="\n"}}- {{this}}{{/list}}
-</dir-context>
-{{/if}}
-
-{{#if appendPrompt}}
-{{appendPrompt}}
-{{/if}}
-
 {{SECTION_SEPARATOR "Identity"}}
 
 <role>

package/src/prompts/tools/github.md

@@ -4,6 +4,7 @@ GitHub CLI tool with a single op-based dispatch. Wraps `gh` for repository, issu
 Pick the operation via `op`. Each op uses a subset of the parameters:
 - `repo_view` — Read repository metadata. Optional `repo` (owner/repo) and `branch`. Falls back to the current checkout or default `gh` repo.
 - `issue_view` — Read an issue. Required `issue` (number or URL). Optional `repo`. Set `comments: false` to skip discussion.
+- `pr_create` — Create a pull request. Either provide `title` (and optional `body`) or set `fill: true` to auto-fill from commits. Optional `base` (target, defaults to repo default), `head` (source, defaults to current branch), `draft`, `repo`, `reviewer[]`, `assignee[]`, `label[]`. Returns the new PR URL plus a summary.
 - `pr_view` — Read one or more pull requests, including reviews and inline review comments. Optional `pr` (number, URL, branch, or array of any — pass an array to fetch multiple PRs in one call); omitting it targets the current branch's PR. Optional `repo`. Set `comments: false` for a lighter summary.
 - `pr_diff` — Read one or more pull request diffs. Optional `pr` (single identifier or array for batch). Optional `repo`. Set `nameOnly: true` for changed file names. Use `exclude` to drop generated paths from the diff.
 - `pr_checkout` — Check one or more pull requests out into dedicated git worktrees. Optional `pr` (number, URL, branch, or array of any of those — pass an array to batch-check-out multiple PRs in one call), `repo`, `force` (reset existing local branch).

package/src/prompts/tools/hashline.md

@@ -8,8 +8,8 @@ This format is purely textual. The tool has NO awareness of language, indentatio
 
 <ops>
 @PATH            header: subsequent ops apply to PATH
-< ANCHOR         insert lines BEFORE the anchored line (or BOF); payload follows as `{{hsep}}TEXT` lines
 + ANCHOR         insert lines AFTER  the anchored line (or EOF); payload follows as `{{hsep}}TEXT` lines
+< ANCHOR         insert lines BEFORE the anchored line (or BOF); payload follows as `{{hsep}}TEXT` lines
 - A..B           delete the line range (inclusive); `- A` for one line
 = A..B           replace the range with payload `{{hsep}}TEXT` lines, or with one blank line if no payload follows
 </ops>
@@ -20,6 +20,7 @@ This format is purely textual. The tool has NO awareness of language, indentatio
 - `< A` inserts before line A; `+ A` inserts after line A. `< BOF` / `+ BOF` both prepend; `< EOF` / `+ EOF` both append.
 - `= A..B` replaces the inclusive range with the following payload lines. `= A` (or `= A..B`) with no payload blanks the range to a single empty line.
 - `- A..B` deletes the inclusive range; omit `..B` for one line.
+- 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>
 
 <case file="a.ts">
@@ -39,11 +40,9 @@ This format is purely textual. The tool has NO awareness of language, indentatio
 
 # Replace a contiguous range with multiple lines
 @a.ts
-= {{hrefr 3}}..{{hrefr 6}}
-{{hsep}}export function label(name: string): string {
+= {{hrefr 4}}..{{hrefr 5}}
 {{hsep}}	const clean = (name || DEF).trim();
 {{hsep}}	return clean.length === 0 ? DEF : clean.toUpperCase();
-{{hsep}}}
 
 # Insert BEFORE a line
 @a.ts
@@ -73,11 +72,30 @@ This format is purely textual. The tool has NO awareness of language, indentatio
 = {{hrefr 2}}
 </examples>
 
+<anti-pattern>
+# WRONG — replaces 6 lines just to add one. Use `+` at the boundary instead.
+@a.ts
+= {{hrefr 1}}..{{hrefr 6}}
+{{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
++ {{hrefr 1}}
+{{hsep}}const DEBUG = false;
+
+If your replacement payload would render with even one unchanged line in the diff, you have the wrong op or the wrong range. Stop and rewrite as `+`/`<`/`-` plus a narrower `=`.
+</anti-pattern>
+
 <critical>
 - Always copy anchors exactly from tool output, but **NEVER** include line content after the `{{hsep}}` separator in the op line.
-- Only emit changed lines. Do not restate unchanged context as payload.
 - Every inserted/replacement content line **MUST** start with `{{hsep}}`; raw content lines are invalid.
 - Do not write unified diff syntax (`@@`, `-OLD`, `+NEW`).
-- To replace a block, use one `= A..B` op followed by all replacement `{{hsep}}TEXT` payload lines.
 - `= A..B` deletes the range; payload is what's written. If a payload edge line already exists immediately outside `A..B`, widen the range to cover it — otherwise it duplicates.
+- Multiple ops in one patch are cheap. Prefer two narrow ops over one wide `=`.
 </critical>

package/src/prompts/tools/read.md

@@ -6,29 +6,30 @@ The `read` tool is multi-purpose and more capable than it looks — inspects fil
 - For URLs, `read` fetches the page and returns clean extracted text/markdown by default (reader-mode). It handles HTML pages, GitHub issues/PRs, Stack Overflow, Wikipedia, Reddit, NPM, arXiv, RSS/Atom, JSON endpoints, PDFs, etc. You **SHOULD** reach for `read` — not a browser/puppeteer tool — for fetching and inspecting web content.
 
 ## Parameters
-- `path` — file path or URL (required)
-- `sel` — optional selector for line ranges or raw mode
+- `path` — file path or URL (required). Append `:<sel>` for line ranges or raw mode (for example `src/foo.ts:50-200` or `src/foo.ts:raw`).
 - `timeout` — seconds, for URLs only
 
 ## Selectors
 
-|`sel` value|Behavior|
+|`path` suffix|Behavior|
 |---|---|
-|_(omitted)_|Read full file (up to {{DEFAULT_LIMIT}} lines)|
-|`50`|Read from line 50 onward|
-|`50-200`|Read lines 50-200|
-|`50+150`|Read 150 lines starting at line 50|
-|`20+1`|Read exactly one line|
+|_(omitted)_|For parseable code files, return a structural summary. Otherwise read from the start (up to {{DEFAULT_LIMIT}} lines).|
+|`:50`|Read from line 50 onward|
+|`:50-200`|Read lines 50-200|
+|`:50+150`|Read 150 lines starting at line 50|
+|`:20+1`|Read exactly one line|
+|`:raw`|Read verbatim text without anchors or summarization|
 
 # Filesystem
 - Reading a directory path returns a list of dirents.
   {{#if IS_HL_MODE}}
-- Reading a file returns lines prefixed with anchors (line+hash): `41th|def alpha():`
+- Reading a file with an explicit selector returns lines prefixed with anchors (line+hash): `41th|def alpha():`
   {{else}}
   {{#if IS_LINE_NUMBER_MODE}}
-- Reading a file returns lines prefixed with line numbers: `41|def alpha():`
+- Reading a file with an explicit selector returns lines prefixed with line numbers: `41|def alpha():`
   {{/if}}
   {{/if}}
+- Reading a parseable code file without a selector returns a structural summary with signatures/declarations kept and large bodies collapsed to `…`. Use `:raw` or an explicit range such as `:1-9999` for verbatim content.
 
 # Inspection
 
@@ -36,7 +37,7 @@ Extracts text from PDF, Word, PowerPoint, Excel, RTF, EPUB, and Jupyter notebook
 
 # Directories & Archives
 
-Directories and archive roots return a list of entries. Supports `.tar`, `.tar.gz`, `.tgz`, `.zip`. Use `archive.ext:path/inside/archive` to read contents.
+Directories and archive roots return a list of entries. Supports `.tar`, `.tar.gz`, `.tgz`, `.zip`. Use `archive.ext:path/inside/archive` to read contents, and append a selector to the archive entry such as `archive.zip:dir/file.ts:50-60`.
 
 # SQLite Databases
 
@@ -50,13 +51,13 @@ For `.sqlite`, `.sqlite3`, `.db`, `.db3`:
 
 # URLs
 
-Extracts content from web pages, GitHub issues/PRs, Stack Overflow, Wikipedia, Reddit, NPM, arXiv, RSS/Atom feeds, JSON endpoints, PDFs at URLs, and similar text-based resources. Returns clean reader-mode text/markdown — no browser required. Use `sel="raw"` for untouched HTML; `timeout` to override the default request timeout.
+Extracts content from web pages, GitHub issues/PRs, Stack Overflow, Wikipedia, Reddit, NPM, arXiv, RSS/Atom feeds, JSON endpoints, PDFs at URLs, and similar text-based resources. Returns clean reader-mode text/markdown — no browser required. Use a `:raw` suffix for untouched HTML; `timeout` to override the default request timeout. URL line selectors require the `L` form, for example `https://example.com/page:L50-L60`.
 </instruction>
 
 <critical>
 - You **MUST** use `read` for every file, directory, archive, and URL read. `cat`, `head`, `tail`, `less`, `more`, `ls`, `tar`, `unzip`, `curl`, and `wget` are **FORBIDDEN** for inspection — any such Bash call is a bug, regardless of how short or convenient it looks.
 - You **MUST** prefer `read` over a browser/puppeteer tool for fetching URL content; only use a browser if `read` fails to deliver reasonable content.
 - You **MUST** always include the `path` parameter — never call `read` with an empty argument object `{}`.
-- For specific line ranges, use `sel` (e.g. `sel="50-200"`, `sel="50+150"`) — do **NOT** reach for `sed -n`, `awk NR`, or `head`/`tail` pipelines.
-- You **MAY** use `sel` with URL reads; the tool paginates cached fetched output.
+- For specific line ranges, append the selector to `path` (e.g. `path="src/foo.ts:50-200"`, `path="src/foo.ts:50+150"`) — do **NOT** reach for `sed -n`, `awk NR`, or `head`/`tail` pipelines.
+- You **MAY** use path suffix selectors with URL reads; the tool paginates cached fetched output.
 </critical>

package/src/sdk.ts

@@ -102,6 +102,7 @@ import { closeAllConnections } from "./ssh/connection-manager";
 import { unmountAll } from "./ssh/sshfs-mount";
 import {
 	type AgentsMdSearch,
+	type BuildSystemPromptResult,
 	buildAgentsMdSearch,
 	buildSystemPrompt as buildSystemPromptInternal,
 	buildSystemPromptToolMetadata,
@@ -140,6 +141,7 @@ import { wrapToolWithMetaNotice } from "./tools/output-meta";
 import { queueResolveHandler } from "./tools/resolve";
 import { EventBus } from "./utils/event-bus";
 import { buildNamedToolChoice } from "./utils/tool-choice";
+import { buildWorkspaceTree, type WorkspaceTree } from "./workspace-tree";
 
 // Types
 export interface CreateAgentSessionOptions {
@@ -165,8 +167,8 @@ export interface CreateAgentSessionOptions {
 	/** Models available for cycling (Ctrl+P in interactive mode) */
 	scopedModels?: Array<{ model: Model; thinkingLevel?: ThinkingLevel }>;
 
-	/** System prompt. String replaces default, function receives default and returns final. */
-	systemPrompt?: string | ((defaultPrompt: string) => string);
+	/** System prompt blocks. Array replaces default, function receives default blocks and returns final blocks. */
+	systemPrompt?: string[] | ((defaultPrompt: string[]) => string[]);
 	/** Optional provider-facing session identifier for prompt caches and sticky auth selection.
 	 * Keeps persisted session files isolated while reusing provider-side caches. */
 	providerSessionId?: string;
@@ -270,6 +272,7 @@ export type { Skill } from "./extensibility/skills";
 export type { FileSlashCommand } from "./extensibility/slash-commands";
 export type { MCPManager, MCPServerConfig, MCPServerConnection, MCPToolsLoadResult } from "./mcp";
 export type { Tool } from "./tools";
+export { buildDirectoryTree, buildWorkspaceTree, type DirectoryTree, type WorkspaceTree } from "./workspace-tree";
 
 export {
 	// Individual tool classes (for custom usage)
@@ -399,9 +402,12 @@ export interface BuildSystemPromptOptions {
 }
 
 /**
- * Build the default system prompt.
+ * Build the default provider-facing system prompt blocks.
+ *
+ * The returned `systemPrompt` preserves the stable harness prompt and dynamic project context
+ * as separate entries so providers can cache prompt prefixes without concatenating blocks.
  */
-export async function buildSystemPrompt(options: BuildSystemPromptOptions = {}): Promise<string> {
+export async function buildSystemPrompt(options: BuildSystemPromptOptions = {}): Promise<BuildSystemPromptResult> {
 	return await buildSystemPromptInternal({
 		cwd: options.cwd,
 		skills: options.skills,
@@ -652,7 +658,7 @@ function buildMCPPromptCommands(manager: MCPManager): LoadedCustomCommand[] {
  * const { session } = await createAgentSession({
  *   model: myModel,
  *   getApiKey: async () => Bun.env.MY_KEY,
- *   systemPrompt: 'You are helpful.',
+ *   systemPrompt: ['You are helpful.'],
  *   tools: codingTools({ cwd: getProjectDir() }),
  *   skills: [],
  *   sessionManager: SessionManager.inMemory(),
@@ -680,6 +686,8 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 	// (~200ms on large repos) and only needs `cwd`, so it can overlap with everything that follows.
 	const agentsMdSearchPromise: Promise<AgentsMdSearch> = logger.time("buildAgentsMdSearch", buildAgentsMdSearch, cwd);
 	agentsMdSearchPromise.catch(() => {});
+	const workspaceTreePromise: Promise<WorkspaceTree> = logger.time("buildWorkspaceTree", buildWorkspaceTree, cwd);
+	workspaceTreePromise.catch(() => {});
 
 	// Independent discoveries that depend only on cwd/agentDir — kicked off in parallel and awaited
 	// at their respective consumer sites. Their work can overlap with model resolution, secret loading,
@@ -1330,7 +1338,10 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		const repeatToolDescriptions = settings.get("repeatToolDescriptions");
 		const eagerTasks = settings.get("task.eager");
 		const intentField = settings.get("tools.intentTracing") || $flag("PI_INTENT_TRACING") ? INTENT_FIELD : undefined;
-		const rebuildSystemPrompt = async (toolNames: string[], tools: Map<string, AgentTool>): Promise<string> => {
+		const rebuildSystemPrompt = async (
+			toolNames: string[],
+			tools: Map<string, AgentTool>,
+		): Promise<BuildSystemPromptResult> => {
 			toolContextStore.setToolNames(toolNames);
 			const discoverableMCPTools = mcpDiscoveryEnabled ? collectDiscoverableMCPTools(tools.values()) : [];
 			const discoverableMCPSummary = summarizeDiscoverableMCPTools(discoverableMCPTools);
@@ -1380,33 +1391,18 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 				eagerTasks,
 				secretsEnabled,
 				agentsMdSearch: agentsMdSearchPromise,
+				workspaceTree: workspaceTreePromise,
 			});
 
 			if (options.systemPrompt === undefined) {
 				return defaultPrompt;
 			}
-			if (typeof options.systemPrompt === "string") {
-				return await buildSystemPromptInternal({
-					cwd,
-					skills,
-					contextFiles,
-					tools: promptTools,
-					toolNames,
-					rules: rulebookRules,
-					alwaysApplyRules,
-					skillsSettings: settings.getGroup("skills"),
-					customPrompt: options.systemPrompt,
-					appendSystemPrompt: appendPrompt,
-					repeatToolDescriptions,
-					intentField,
-					mcpDiscoveryMode: hasDiscoverableMCPTools,
-					mcpDiscoveryServerSummaries: discoverableMCPSummary.servers.map(formatDiscoverableMCPToolServerSummary),
-					eagerTasks,
-					secretsEnabled,
-					agentsMdSearch: agentsMdSearchPromise,
-				});
+			if (Array.isArray(options.systemPrompt)) {
+				return { systemPrompt: options.systemPrompt };
 			}
-			return options.systemPrompt(defaultPrompt);
+			return {
+				systemPrompt: options.systemPrompt(defaultPrompt.systemPrompt),
+			};
 		};
 
 		const toolNamesFromRegistry = Array.from(toolRegistry.keys());
@@ -1472,7 +1468,12 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			}
 		}
 
-		const systemPrompt = await logger.time("buildSystemPrompt", rebuildSystemPrompt, initialToolNames, toolRegistry);
+		const { systemPrompt } = await logger.time(
+			"buildSystemPrompt",
+			rebuildSystemPrompt,
+			initialToolNames,
+			toolRegistry,
+		);
 
 		const promptTemplates = await promptTemplatesPromise;
 		toolSession.promptTemplates = promptTemplates;

package/src/session/agent-session.ts

@@ -245,8 +245,8 @@ export interface AgentSessionConfig {
 	onResponse?: SimpleStreamOptions["onResponse"];
 	/** Current session message-to-LLM conversion pipeline */
 	convertToLlm?: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
-	/** System prompt builder that can consider tool availability */
-	rebuildSystemPrompt?: (toolNames: string[], tools: Map<string, AgentTool>) => Promise<string>;
+	/** System prompt builder that can consider tool availability. Returns ordered provider-facing blocks. */
+	rebuildSystemPrompt?: (toolNames: string[], tools: Map<string, AgentTool>) => Promise<{ systemPrompt: string[] }>;
 	/**
 	 * Optional accessor for live MCP server instructions. Read by the session's
 	 * `rebuildSystemPrompt`-skip optimization to detect server-side instruction
@@ -520,9 +520,11 @@ export class AgentSession {
 	#onPayload: SimpleStreamOptions["onPayload"] | undefined;
 	#onResponse: SimpleStreamOptions["onResponse"] | undefined;
 	#convertToLlm: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
-	#rebuildSystemPrompt: ((toolNames: string[], tools: Map<string, AgentTool>) => Promise<string>) | undefined;
+	#rebuildSystemPrompt:
+		| ((toolNames: string[], tools: Map<string, AgentTool>) => Promise<{ systemPrompt: string[] }>)
+		| undefined;
 	#getMcpServerInstructions: (() => Map<string, string> | undefined) | undefined;
-	#baseSystemPrompt: string;
+	#baseSystemPrompt: string[];
 	/**
 	 * Signature of the (toolNames, tool descriptions) tuple passed to the most
 	 * recent successful `rebuildSystemPrompt` call. Used to skip redundant rebuilds
@@ -2083,8 +2085,8 @@ export class AgentSession {
 	getLastAssistantMessage(): AssistantMessage | undefined {
 		return this.#findLastAssistantMessage();
 	}
-	/** Current effective system prompt (includes any per-turn extension modifications) */
-	get systemPrompt(): string {
+	/** Current effective system prompt blocks (includes any per-turn extension modifications) */
+	get systemPrompt(): string[] {
 		return this.agent.state.systemPrompt;
 	}
 
@@ -2281,7 +2283,8 @@ export class AgentSession {
 		if (this.#rebuildSystemPrompt) {
 			const signature = this.#computeAppliedToolSignature(validToolNames, tools);
 			if (signature !== this.#lastAppliedToolSignature) {
-				this.#baseSystemPrompt = await this.#rebuildSystemPrompt(validToolNames, this.#toolRegistry);
+				const built = await this.#rebuildSystemPrompt(validToolNames, this.#toolRegistry);
+				this.#baseSystemPrompt = built.systemPrompt;
 				this.agent.setSystemPrompt(this.#baseSystemPrompt);
 				this.#lastAppliedToolSignature = signature;
 			}
@@ -2324,7 +2327,8 @@ export class AgentSession {
 	async refreshBaseSystemPrompt(): Promise<void> {
 		if (!this.#rebuildSystemPrompt) return;
 		const activeToolNames = this.getActiveToolNames();
-		this.#baseSystemPrompt = await this.#rebuildSystemPrompt(activeToolNames, this.#toolRegistry);
+		const built = await this.#rebuildSystemPrompt(activeToolNames, this.#toolRegistry);
+		this.#baseSystemPrompt = built.systemPrompt;
 		this.agent.setSystemPrompt(this.#baseSystemPrompt);
 		// Refresh the cached signature so a subsequent `#applyActiveToolsByName` with
 		// the same tool set does not re-rebuild on top of the explicit refresh we
@@ -2335,14 +2339,14 @@ export class AgentSession {
 		this.#lastAppliedToolSignature = this.#computeAppliedToolSignature(activeToolNames, activeTools);
 	}
 
-	async #buildSystemPromptForAgentStart(promptText: string): Promise<string> {
+	async #buildSystemPromptForAgentStart(promptText: string): Promise<string[]> {
 		const backend = resolveMemoryBackend(this.settings);
 		if (!backend.beforeAgentStartPrompt) return this.#baseSystemPrompt;
 
 		try {
 			const injected = await backend.beforeAgentStartPrompt(this, promptText);
 			if (!injected) return this.#baseSystemPrompt;
-			return `${this.#baseSystemPrompt}\n\n${injected}`;
+			return [...this.#baseSystemPrompt, injected];
 		} catch (err) {
 			logger.debug("Memory backend beforeAgentStartPrompt failed", {
 				backend: backend.id,
@@ -4215,7 +4219,11 @@ export class AgentSession {
 					apiKey,
 					customInstructions,
 					compactionAbortController.signal,
-					{ promptOverride: hookPrompt, extraContext: hookContext, remoteInstructions: this.#baseSystemPrompt },
+					{
+						promptOverride: hookPrompt,
+						extraContext: hookContext,
+						remoteInstructions: this.#baseSystemPrompt.join("\n\n"),
+					},
 				);
 				summary = result.summary;
 				shortSummary = result.shortSummary;
@@ -5328,7 +5336,7 @@ export class AgentSession {
 							compactResult = await compact(preparation, candidate, apiKey, undefined, autoCompactionSignal, {
 								promptOverride: hookPrompt,
 								extraContext: hookContext,
-								remoteInstructions: this.#baseSystemPrompt,
+								remoteInstructions: this.#baseSystemPrompt.join("\n\n"),
 								initiatorOverride: "agent",
 							});
 							break;

package/src/session/compaction/branch-summarization.ts

@@ -290,7 +290,7 @@ export async function generateBranchSummary(
 	// Call LLM for summarization
 	const response = await completeSimple(
 		model,
-		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
+		{ systemPrompt: [SUMMARIZATION_SYSTEM_PROMPT], messages: summarizationMessages },
 		{ apiKey, signal, maxTokens: 2048 },
 	);
 

package/src/session/compaction/compaction.ts

@@ -1019,7 +1019,7 @@ export async function generateSummary(
 
 	const response = await completeSimple(
 		model,
-		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
+		{ systemPrompt: [SUMMARIZATION_SYSTEM_PROMPT], messages: summarizationMessages },
 		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride: options?.initiatorOverride },
 	);
 
@@ -1066,7 +1066,7 @@ async function generateShortSummary(
 	const response = await completeSimple(
 		model,
 		{
-			systemPrompt: SUMMARIZATION_SYSTEM_PROMPT,
+			systemPrompt: [SUMMARIZATION_SYSTEM_PROMPT],
 			messages: [{ role: "user", content: [{ type: "text", text: promptText }], timestamp: Date.now() }],
 		},
 		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride: options?.initiatorOverride },
@@ -1386,7 +1386,7 @@ async function generateTurnPrefixSummary(
 
 	const response = await completeSimple(
 		model,
-		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
+		{ systemPrompt: [SUMMARIZATION_SYSTEM_PROMPT], messages: summarizationMessages },
 		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride },
 	);
 

package/src/session/session-dump-format.ts

@@ -25,7 +25,7 @@ export interface SessionDumpToolInfo {
 
 export interface FormatSessionDumpTextOptions {
 	messages: readonly AgentMessage[];
-	systemPrompt?: string | null;
+	systemPrompt?: readonly string[] | null;
 	model?: Model | null;
 	thinkingLevel?: ThinkingLevel | string | null;
 	tools?: readonly SessionDumpToolInfo[];
@@ -64,11 +64,16 @@ function formatArgsAsXml(args: Record<string, unknown>, indent = "\t"): string {
 export function formatSessionDumpText(options: FormatSessionDumpTextOptions): string {
 	const lines: string[] = [];
 
-	const systemPrompt = options.systemPrompt;
-	if (systemPrompt) {
+	const systemPrompt = options.systemPrompt?.filter(prompt => prompt.length > 0) ?? [];
+	if (systemPrompt.length > 0) {
 		lines.push("## System Prompt\n");
-		lines.push(systemPrompt);
-		lines.push("\n");
+		for (let index = 0; index < systemPrompt.length; index++) {
+			if (systemPrompt.length > 1) {
+				lines.push(`### System Prompt ${index + 1}\n`);
+			}
+			lines.push(systemPrompt[index]);
+			lines.push("\n");
+		}
 	}
 
 	const model = options.model;

package/src/session/session-manager.ts

@@ -2182,6 +2182,63 @@ export class SessionManager {
 		return manager.getPath(id);
 	}
 
+	/**
+	 * Path to the unsent-input draft sidecar for the current session. Lives inside
+	 * the artifacts directory so it is removed together with the session on
+	 * `dropSession`. Returns null when the session has no on-disk identity.
+	 */
+	#getDraftPath(): string | null {
+		const dir = this.getArtifactsDir();
+		return dir ? path.join(dir, "draft.txt") : null;
+	}
+
+	/**
+	 * Persist (or clear) the current editor draft so the next resume of this
+	 * session can restore it. Empty text deletes any stale draft. No-op when the
+	 * session is not persisted.
+	 */
+	async saveDraft(text: string): Promise<void> {
+		const draftPath = this.#getDraftPath();
+		if (!draftPath || !this.persist) return;
+		if (text.length === 0) {
+			try {
+				await this.storage.unlink(draftPath);
+			} catch (err) {
+				if (!isEnoent(err)) throw err;
+			}
+			return;
+		}
+		// Force the session header onto disk so resume can find the file we are
+		// attaching this draft to. Without this, a session whose first message
+		// never produced an assistant reply would persist a draft next to a
+		// session file that does not exist on disk.
+		await this.ensureOnDisk();
+		await this.storage.writeText(draftPath, text);
+	}
+
+	/**
+	 * Read and remove the saved draft. Returns the previously-saved text, or
+	 * null when no draft is pending. Single-shot: a successful read removes the
+	 * sidecar so a subsequent resume does not re-restore the same text.
+	 */
+	async consumeDraft(): Promise<string | null> {
+		const draftPath = this.#getDraftPath();
+		if (!draftPath) return null;
+		let text: string;
+		try {
+			text = await this.storage.readText(draftPath);
+		} catch (err) {
+			if (isEnoent(err)) return null;
+			throw err;
+		}
+		try {
+			await this.storage.unlink(draftPath);
+		} catch (err) {
+			if (!isEnoent(err)) throw err;
+		}
+		return text;
+	}
+
 	/** The source that set the session name: "user" (manual /rename or RPC) or "auto" (generated title). */
 	get titleSource(): "auto" | "user" | undefined {
 		return this.#titleSource;

package/src/session/streaming-output.ts

@@ -759,7 +759,7 @@ export function formatHeadTruncationNotice(
 	const totalFileLines = options.totalFileLines ?? truncation.totalLines;
 	const endLineDisplay = startLineDisplay + (truncation.outputLines ?? truncation.totalLines) - 1;
 	const nextOffset = endLineDisplay + 1;
-	const notice = `[Showing lines ${startLineDisplay}-${endLineDisplay} of ${totalFileLines}. Use sel=${nextOffset} to continue]`;
+	const notice = `[Showing lines ${startLineDisplay}-${endLineDisplay} of ${totalFileLines}. Use :${nextOffset} to continue]`;
 	return `\n\n${notice}`;
 }