@oh-my-pi/pi-coding-agent 15.7.2 → 15.7.3

package/src/eval/js/tool-bridge.ts

@@ -1,6 +1,8 @@
 import type { AgentTool, AgentToolResult } from "@oh-my-pi/pi-agent-core";
 import type { ToolSession } from "../../tools";
 import { ToolError } from "../../tools/tool-errors";
+import { EVAL_AGENT_BRIDGE_NAME, runEvalAgent } from "../agent-bridge";
+import { EVAL_BUDGET_BRIDGE_NAME, type EvalBudgetResult, runEvalBudget } from "../budget-bridge";
 import { EVAL_LLM_BRIDGE_NAME, runEvalLlm } from "../llm-bridge";
 import type { JsStatusEvent } from "./shared/types";
 
@@ -14,6 +16,7 @@ interface ToolBridgeOptions {
 
 type ToolValue =
 	| string
+	| EvalBudgetResult
 	| {
 			text: string;
 			details?: unknown;
@@ -105,6 +108,12 @@ export async function callSessionTool(name: string, args: unknown, options: Tool
 	if (name === EVAL_LLM_BRIDGE_NAME) {
 		return await runEvalLlm(args, options);
 	}
+	if (name === EVAL_AGENT_BRIDGE_NAME) {
+		return await runEvalAgent(args, options);
+	}
+	if (name === EVAL_BUDGET_BRIDGE_NAME) {
+		return await runEvalBudget(args, options);
+	}
 	const tool = getTool(options.session, name);
 	const normalizedArgs = normalizeArgs(args);
 	const toolCallId = `js-${name}-${crypto.randomUUID()}`;

package/src/eval/py/executor.ts

@@ -25,6 +25,12 @@ export interface PythonExecutorOptions {
 	timeoutMs?: number;
 	/** Absolute wall-clock deadline in milliseconds since epoch */
 	deadlineMs?: number;
+	/**
+	 * Inactivity budget (ms). Used only for timeout-annotation text when the
+	 * caller drives cancellation via an idle-aware `signal` instead of a
+	 * wall-clock `deadlineMs`/`timeoutMs`. Does not arm a timer.
+	 */
+	idleTimeoutMs?: number;
 	/** Callback for streaming output chunks (already sanitized) */
 	onChunk?: (chunk: string) => Promise<void> | void;
 	/** AbortSignal for cancellation */
@@ -57,6 +63,13 @@ export interface PythonExecutorOptions {
 	toolSession?: ToolSession;
 	/** Callback for status events emitted by tool bridge invocations. */
 	emitStatus?: (event: JsStatusEvent) => void;
+	/**
+	 * Live status events streamed as they are emitted (both host-side bridge
+	 * helpers like `agent()` and kernel-side `display`/`log`/`phase`). Mirrors
+	 * what lands in `displayOutputs` so callers can render progress before the
+	 * cell finishes.
+	 */
+	onStatus?: (event: JsStatusEvent) => void;
 	/** @internal Bridge session id, set by `executePython` before delegating. */
 	bridgeSessionId?: string;
 	/** @internal Bridge endpoint info, set by `executePython` before delegating. */
@@ -218,18 +231,20 @@ async function waitForPromiseWithCancellation<T>(
 // Result formatting
 // ---------------------------------------------------------------------------
 
-function formatTimeoutAnnotation(timeoutMs?: number): string | undefined {
+function formatTimeoutAnnotation(timeoutMs?: number, idle = false): string | undefined {
+	const suffix = idle ? " of inactivity" : "";
 	if (timeoutMs === undefined) return "Command timed out";
 	const secs = Math.max(1, Math.round(timeoutMs / 1000));
-	return `Command timed out after ${secs} seconds`;
+	return `Command timed out after ${secs} seconds${suffix}`;
 }
 
-function formatKernelTimeoutAnnotation(timeoutMs: number | undefined, kernelKilled: boolean): string {
+function formatKernelTimeoutAnnotation(timeoutMs: number | undefined, kernelKilled: boolean, idle = false): string {
 	const secs = timeoutMs === undefined ? undefined : Math.max(1, Math.round(timeoutMs / 1000));
+	const suffix = idle ? " of inactivity" : "";
 	if (kernelKilled) {
-		return "eval cell timed out and the kernel was unresponsive to interrupt; the kernel has been killed and will be recreated on the next call.";
+		return `eval cell timed out${suffix} and the kernel was unresponsive to interrupt; the kernel has been killed and will be recreated on the next call.`;
 	}
-	const duration = secs === undefined ? "the configured timeout" : `${secs}s`;
+	const duration = secs === undefined ? "the configured timeout" : `${secs}s${suffix}`;
 	return `eval cell timed out after ${duration}; kernel interrupted but remains running. Reset the kernel via { reset: true } if state appears corrupted.`;
 }
 
@@ -473,12 +488,18 @@ async function executeWithKernel(
 	const displayOutputs: KernelDisplayOutput[] = [];
 	const deadlineMs = getExecutionDeadlineMs(options);
 	let executionTimeoutMs: number | undefined;
-
-	const emitStatus =
-		options?.emitStatus ??
-		((event: JsStatusEvent) => {
-			displayOutputs.push({ type: "status", event });
-		});
+	// Idle mode: the caller (eval tool) drives cancellation via an idle-aware
+	// signal and passes no wall-clock deadline, so annotate timeouts with the
+	// configured inactivity budget rather than a remaining-deadline figure.
+	const idleMode = deadlineMs === undefined && options?.idleTimeoutMs !== undefined;
+
+	// Collect every display output and, for status events, stream them live so
+	// long-running bridge helpers (e.g. `agent()`) surface progress mid-cell.
+	const collectDisplay = (output: KernelDisplayOutput) => {
+		displayOutputs.push(output);
+		if (output.type === "status") options?.onStatus?.(output.event);
+	};
+	const emitStatus = options?.emitStatus ?? ((event: JsStatusEvent) => collectDisplay({ type: "status", event }));
 	const runId = `py-${crypto.randomUUID()}`;
 	const unregisterBridge =
 		options?.toolSession && options?.bridgeSessionId
@@ -498,12 +519,16 @@ async function executeWithKernel(
 			signal: options?.signal,
 			timeoutMs: executionTimeoutMs,
 			onChunk: text => sink.push(text),
-			onDisplay: output => void displayOutputs.push(output),
+			onDisplay: output => collectDisplay(output),
 		});
 
 		if (result.cancelled) {
 			const annotation = result.timedOut
-				? formatKernelTimeoutAnnotation(executionTimeoutMs, result.kernelKilled ?? false)
+				? formatKernelTimeoutAnnotation(
+						executionTimeoutMs ?? options?.idleTimeoutMs,
+						result.kernelKilled ?? false,
+						idleMode,
+					)
 				: undefined;
 			return {
 				exitCode: undefined,
@@ -540,7 +565,9 @@ async function executeWithKernel(
 				cancelled: true,
 				displayOutputs,
 				stdinRequested: false,
-				...(await sink.dump(timedOut ? formatTimeoutAnnotation(executionTimeoutMs) : undefined)),
+				...(await sink.dump(
+					timedOut ? formatTimeoutAnnotation(executionTimeoutMs ?? options?.idleTimeoutMs, idleMode) : undefined,
+				)),
 			};
 		}
 		const error = err instanceof Error ? err : new Error(String(err));

package/src/eval/py/index.ts

@@ -28,7 +28,7 @@ export default {
 		const kernelMode = readSetting<PythonExecutorOptions["kernelMode"]>(opts.session, "python.kernelMode");
 		const executorOptions: PythonExecutorOptions = {
 			cwd: opts.cwd,
-			deadlineMs: opts.deadlineMs,
+			idleTimeoutMs: opts.idleTimeoutMs,
 			signal: opts.signal,
 			sessionId: namespaceSessionId(opts.sessionId),
 			kernelMode,
@@ -39,6 +39,7 @@ export default {
 			artifactPath: opts.artifactPath,
 			artifactId: opts.artifactId,
 			onChunk: opts.onChunk,
+			onStatus: opts.onStatus,
 			toolSession: opts.session,
 		};
 		const result = await executePython(code, executorOptions);

package/src/eval/py/prelude.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 if "__omp_prelude_loaded__" not in globals():
     __omp_prelude_loaded__ = True
     from pathlib import Path
-    import os, json
+    import os, json, math
 
     # __omp_display is injected by runner.py before the prelude executes; it
     # mirrors IPython's display() semantics with the same MIME bundle output.
@@ -479,3 +479,134 @@ if "__omp_prelude_loaded__" not in globals():
         res = _bridge_call("__llm__", args)
         text = res.get("text") if isinstance(res, dict) else res
         return json.loads(text) if schema is not None else text
+
+    def agent(prompt, *, agent_type="task", model=None, context=None, label=None, schema=None):
+        """Run a subagent and return its final output.
+
+        `agent_type` selects the subagent definition (default "task"). Pass
+        `model` to override that agent's model, `context` for shared background,
+        `label` for the output artifact id, and `schema` to request structured
+        JSON output; when `schema` is supplied the parsed object is returned.
+        """
+        args = {"prompt": prompt}
+        if agent_type is not None:
+            args["agentType"] = agent_type
+        if model is not None:
+            args["model"] = model
+        if context is not None:
+            args["context"] = context
+        if label is not None:
+            args["label"] = label
+        if schema is not None:
+            args["schema"] = schema
+        res = _bridge_call("__agent__", args)
+        text = res.get("text") if isinstance(res, dict) else res
+        return json.loads(text) if schema is not None else text
+
+    def _normalize_concurrency(value):
+        """Clamp a concurrency hint to [1, 16], defaulting to 4 on bad input."""
+        try:
+            n = int(value)
+        except (TypeError, ValueError):
+            n = 4
+        return max(1, min(16, n))
+
+    def _pool_map(items, fn, concurrency):
+        """Run ``fn`` over ``items`` through a bounded thread pool.
+
+        Preserves input order, barriers until every task settles, and raises the
+        lowest-index exception if any task failed. Each task runs inside a copy
+        of the submitting thread's context so the ``_CURRENT_RID`` ContextVar
+        propagates and bridge calls (agent(), tool.*, etc.) keep working.
+        """
+        import concurrent.futures, contextvars
+        items = list(items)
+        if not items:
+            return []
+        workers = min(_normalize_concurrency(concurrency), len(items))
+        results = [None] * len(items)
+        errors = {}
+        with concurrent.futures.ThreadPoolExecutor(max_workers=workers) as pool:
+            futures = {}
+            for i, item in enumerate(items):
+                ctx = contextvars.copy_context()
+                futures[pool.submit(ctx.run, fn, item)] = i
+            for fut in concurrent.futures.as_completed(futures):
+                i = futures[fut]
+                try:
+                    results[i] = fut.result()
+                except BaseException as exc:  # noqa: BLE001 - propagate to caller
+                    errors[i] = exc
+        if errors:
+            raise errors[min(errors)]
+        return results
+
+    def parallel(thunks, *, concurrency=4):
+        """Run zero-arg callables through a bounded pool, preserving input order.
+
+        Barriers until all finish; re-raises the lowest-index exception if any
+        thunk raised.
+        """
+        thunks = list(thunks)
+        for t in thunks:
+            if not callable(t):
+                raise TypeError("parallel() expects an iterable of zero-arg callables")
+        return _pool_map(thunks, lambda t: t(), concurrency)
+
+    def pipeline(items, *stages, concurrency=4):
+        """Map items left-to-right through one-arg stage callables.
+
+        Every item clears stage N before any item enters stage N+1 (barrier per
+        stage). Stage 1 receives the original item; later stages receive the
+        previous stage's result.
+        """
+        current = list(items)
+        for stage in stages:
+            if not callable(stage):
+                raise TypeError("pipeline() stages must be callables")
+            current = _pool_map(current, stage, concurrency)
+        return current
+
+    def log(message):
+        """Emit a status ``log`` event for TUI rendering."""
+        _emit_status("log", message=str(message))
+        return None
+
+    def phase(title):
+        """Record the current readable phase and emit a status ``phase`` event."""
+        globals()["__omp_current_phase__"] = str(title)
+        _emit_status("phase", title=str(title))
+        return None
+
+    class _Budget:
+        """Live view of the host Goal Mode token budget via the host bridge."""
+
+        @property
+        def total(self):
+            snap = _bridge_call("__budget__", {})
+            return (snap or {}).get("total")
+
+        @property
+        def hard(self):
+            snap = _bridge_call("__budget__", {})
+            return bool((snap or {}).get("hard"))
+
+        def spent(self):
+            snap = _bridge_call("__budget__", {})
+            return int((snap or {}).get("spent") or 0)
+
+        def remaining(self):
+            snap = _bridge_call("__budget__", {}) or {}
+            total = snap.get("total")
+            if total is None:
+                return math.inf
+            return max(0, total - int(snap.get("spent") or 0))
+
+        def __repr__(self):
+            try:
+                snap = _bridge_call("__budget__", {}) or {}
+                return f"<budget total={snap.get('total')} spent={snap.get('spent')}>"
+            except Exception:
+                return "<budget unavailable>"
+
+    budget = _Budget()

package/src/exec/bash-executor.ts

@@ -16,6 +16,7 @@ export interface BashExecutorOptions {
 	cwd?: string;
 	timeout?: number;
 	onChunk?: (chunk: string) => void;
+	chunkThrottleMs?: number;
 	signal?: AbortSignal;
 	/** Session key suffix to isolate shell sessions per agent */
 	sessionKey?: string;
@@ -97,9 +98,7 @@ export async function executeBash(command: string, options?: BashExecutorOptions
 		artifactId: options?.artifactId,
 		headBytes: resolveOutputSinkHeadBytes(settings),
 		maxColumns: resolveOutputMaxColumns(settings),
-		// Throttle the streaming preview callback to avoid saturating the
-		// event loop when commands produce massive output (e.g. seq 1 50M).
-		chunkThrottleMs: options?.onChunk ? 50 : 0,
+		chunkThrottleMs: options?.onChunk ? (options.chunkThrottleMs ?? 50) : 0,
 	});
 
 	// sink.push() is synchronous — buffer management, counters, and onChunk

package/src/extensibility/custom-tools/types.ts

@@ -101,11 +101,11 @@ export type CustomToolSessionEvent =
 	| {
 			reason: "auto_compaction_start";
 			trigger: "threshold" | "overflow" | "idle" | "incomplete";
-			action: "context-full" | "handoff";
+			action: "context-full" | "handoff" | "shake" | "shake-summary";
 	  }
 	| {
 			reason: "auto_compaction_end";
-			action: "context-full" | "handoff";
+			action: "context-full" | "handoff" | "shake" | "shake-summary";
 			result: CompactionResult | undefined;
 			aborted: boolean;
 			willRetry: boolean;

package/src/extensibility/extensions/runner.ts

@@ -315,9 +315,15 @@ export class ExtensionRunner {
 		return tools;
 	}
 
-	getFlags(): Map<string, ExtensionFlag> {
+	/**
+	 * Aggregate the registered CLI flags across a set of extensions (last write
+	 * wins on name collision). Static so callers that need the flag set before a
+	 * runner exists — e.g. the CLI resolving `@file`/flag args before session
+	 * creation — share this exact logic instead of duplicating it.
+	 */
+	static aggregateFlags(extensions: readonly Extension[]): Map<string, ExtensionFlag> {
 		const allFlags = new Map<string, ExtensionFlag>();
-		for (const ext of this.extensions) {
+		for (const ext of extensions) {
 			for (const [name, flag] of ext.flags) {
 				allFlags.set(name, flag);
 			}
@@ -325,6 +331,10 @@ export class ExtensionRunner {
 		return allFlags;
 	}
 
+	getFlags(): Map<string, ExtensionFlag> {
+		return ExtensionRunner.aggregateFlags(this.extensions);
+	}
+
 	getFlagValues(): Map<string, boolean | string> {
 		return new Map(this.runtime.flagValues);
 	}

package/src/extensibility/plugins/git-url.ts

@@ -25,6 +25,29 @@ const KNOWN_HOSTS: Record<string, (pathname: string, hash: string) => { user: st
 	"codeberg.org": extractStandard,
 };
 
+/**
+ * Namespaced shorthand prefixes accepted by `omp plugin install`, mapped to
+ * their canonical host. `PluginManager.install` normalizes non-GitHub prefixes
+ * before invoking bun because bun only treats `github:` as a hosted shorthand.
+ */
+const SHORTHAND_PREFIXES: Record<string, string> = {
+	github: "github.com",
+	gitlab: "gitlab.com",
+	bitbucket: "bitbucket.org",
+	codeberg: "codeberg.org",
+	sourcehut: "git.sr.ht",
+	srht: "git.sr.ht",
+};
+
+/**
+ * `<prefix>:<user>/<repo>[.git][#<ref>]` shape. `<repo>` is non-greedy so the
+ * optional `.git` suffix and `#ref` tail bind tightly; `<repo>` may itself
+ * contain `/` to support nested GitLab groups (`gitlab:group/sub/project`).
+ * `<user>` rejects `/`, `:`, `#` to keep protocol URLs (`https://…`) and
+ * scp-like SSH (`git@github.com:user/repo`) out of this path.
+ */
+const SHORTHAND_RE = /^([a-z]+):([^/:#]+)\/([^#]+?)(?:\.git)?(?:#(.+))?$/i;
+
 function stripUrlCredentials(url: string): string {
 	if (!url.includes("://")) return url;
 	try {
@@ -209,14 +232,55 @@ function parseGenericGitUrl(url: string): GitSource | null {
 	return { type: "git", repo, host, path: normalizedPath, ref, pinned: Boolean(ref) };
 }
 
+/**
+ * Match an npm/bun-style namespaced shorthand (`github:user/repo`, optionally
+ * `…#ref` or `….git`). Returns null for protocol URLs and any prefix not in
+ * `SHORTHAND_PREFIXES` so the caller can fall through to the generic paths.
+ */
+function tryNamespacedShorthand(trimmed: string): GitSource | null {
+	// Cheap gate: bail out before touching protocol URLs (`https://`, `ssh://`,
+	// `git://`) where the char after the colon is always `/`. The shorthand we
+	// care about never starts with `<scheme>://`.
+	if (!/^[a-z]+:[^/]/i.test(trimmed)) return null;
+	const match = trimmed.match(SHORTHAND_RE);
+	if (!match) return null;
+	const prefix = (match[1] ?? "").toLowerCase();
+	const host = SHORTHAND_PREFIXES[prefix];
+	if (!host) return null;
+	const user = match[2] ?? "";
+	const repoPath = match[3] ?? "";
+	if (!user || !repoPath) return null;
+	const ref = match[4];
+	if (ref) {
+		try {
+			decodeURIComponent(ref);
+		} catch {
+			return null;
+		}
+	}
+	const fullPath = `${user}/${repoPath}`;
+	return {
+		type: "git",
+		repo: `https://${host}/${fullPath}`,
+		host,
+		path: fullPath,
+		ref: ref || undefined,
+		pinned: Boolean(ref),
+	};
+}
+
 /**
  * Parse git source into a GitSource.
  *
  * Rules:
- * - With `git:` prefix, accept shorthand forms.
+ * - Namespaced shorthand (`github:user/repo`, `gitlab:`, `bitbucket:`,
+ *   `codeberg:`, `sourcehut:`/`srht:`) is accepted directly; installers should
+ *   normalize entries that bun does not understand natively.
+ * - With `git:` prefix, accept generic shorthand forms.
  * - Without `git:` prefix, only accept explicit protocol URLs.
  *
  * Handles:
+ * - `github:user/repo[#ref]`-style namespaced shorthand
  * - `git:` prefixed URLs (`git:github.com/user/repo`)
  * - SSH SCP-like URLs (`git:git@github.com:user/repo`)
  * - HTTPS/HTTP/SSH/git protocol URLs
@@ -227,10 +291,23 @@ function parseGenericGitUrl(url: string): GitSource | null {
  */
 export function parseGitUrl(source: string): GitSource | null {
 	const trimmed = source.trim();
-	const hasGitPrefix = /^git:(?!\/\/)/i.test(trimmed);
-	const url = hasGitPrefix ? trimmed.slice(4).trim() : trimmed;
 
-	if (!hasGitPrefix && !/^(https?|ssh|git):\/\//i.test(url)) {
+	const shorthand = tryNamespacedShorthand(trimmed);
+	if (shorthand) return shorthand;
+
+	// Strip the `git+` URL prefix that npm/bun accept (`git+https://…`,
+	// `git+ssh://…`, `git+git://…`). The rest of the pipeline only deals with
+	// bare schemes.
+	const stripped = /^git\+/i.test(trimmed) ? trimmed.slice(4) : trimmed;
+
+	const hasGitPrefix = /^git:(?!\/\/)/i.test(stripped);
+	const url = hasGitPrefix ? stripped.slice(4).trim() : stripped;
+
+	// Accept: explicit protocol URL, `git:` shorthand, or scp-like SSH
+	// (`git@host:user/repo`). The scp form is unambiguous — no local path
+	// starts with `git@` — and matches the syntax that `git clone` itself
+	// accepts, which `bun install` forwards through.
+	if (!hasGitPrefix && !/^(https?|ssh|git):\/\//i.test(url) && !/^git@[^:]+:.+\/.+/i.test(url)) {
 		return null;
 	}
 
@@ -279,3 +356,12 @@ export function parseGitUrl(source: string): GitSource | null {
 
 	return parseGenericGitUrl(url);
 }
+
+/**
+ * Returns true if the spec is parseable as a git source (protocol URL,
+ * scp-like SSH wrapped in `git:`, plain `git:` shorthand, or namespaced
+ * shorthand like `github:user/repo`). The inverse of "this is an npm spec".
+ */
+export function isGitSpec(spec: string): boolean {
+	return parseGitUrl(spec) !== null;
+}

package/src/extensibility/plugins/manager.ts

@@ -10,6 +10,7 @@ import {
 	isEnoent,
 	logger,
 } from "@oh-my-pi/pi-utils";
+import { type GitSource, parseGitUrl } from "./git-url";
 import { extractPackageName, parsePluginSpec } from "./parser";
 import type {
 	DoctorCheck,
@@ -29,8 +30,14 @@ import type {
 /** Valid npm package name pattern (scoped and unscoped, with optional version) */
 const VALID_PACKAGE_NAME = /^(@[a-z0-9-~][a-z0-9-._~]*\/)?[a-z0-9-~][a-z0-9-._~]*(@[a-z0-9-._^~>=<]+)?$/i;
 
+/** Characters that are never valid in any plugin install spec — git or npm. */
+const SHELL_METACHARS = /[;&|`$(){}<>\\\n\r\t]/;
+
 /**
- * Validate package name to prevent command injection.
+ * Validate package name to prevent command injection. npm specs only — git
+ * specs (`github:user/repo`, `https://github.com/...`, ...) MUST go through
+ * {@link validateGitSpec} instead because they contain characters npm rejects
+ * (`:`, `/`, `#`, `+`, `@` in non-version positions).
  */
 function validatePackageName(name: string): void {
 	// Remove version specifier for validation
@@ -44,6 +51,29 @@ function validatePackageName(name: string): void {
 	}
 }
 
+/**
+ * Validate a git install spec — accepts `:`, `/`, `#`, `+`, `.`, `-`, `_`,
+ * `~`, `@` (which would all fail {@link validatePackageName}) but rejects
+ * shell metacharacters so the spec stays safe when forwarded to bun install.
+ * `Bun.spawn` does not invoke a shell, but defense-in-depth keeps things
+ * obvious for future readers.
+ */
+function validateGitSpec(spec: string): void {
+	if (SHELL_METACHARS.test(spec)) {
+		throw new Error(`Invalid characters in plugin source: ${spec}`);
+	}
+}
+
+function gitInstallSpec(original: string, source: GitSource): string {
+	if (/^github:/i.test(original) || !/^[a-z]+:[^/]/i.test(original)) {
+		return original;
+	}
+	if (!source.ref || source.repo.includes("#")) {
+		return source.repo;
+	}
+	return `${source.repo}#${source.ref}`;
+}
+
 // =============================================================================
 // Plugin Manager
 // =============================================================================
@@ -127,12 +157,39 @@ export class PluginManager {
 		}
 	}
 
+	/**
+	 * Read the `dependencies` map from `plugins/package.json`. Returns an empty
+	 * object when the file does not exist yet so callers can diff `before`
+	 * against `after` to discover the package bun just installed under its
+	 * real name (git specs do not encode the package name in the spec itself).
+	 */
+	async #readDeps(pkgJsonPath: string): Promise<Record<string, string>> {
+		try {
+			const json = await Bun.file(pkgJsonPath).json();
+			return (json.dependencies as Record<string, string>) ?? {};
+		} catch (err) {
+			if (isEnoent(err)) return {};
+			throw err;
+		}
+	}
+
 	// ==========================================================================
 	// Install / Uninstall
 	// ==========================================================================
 
 	/**
-	 * Install a plugin from npm with optional feature selection.
+	 * Install a plugin with optional feature selection.
+	 *
+	 * Accepts:
+	 * - npm specs: `pkg`, `pkg@1.2.3`, `@scope/pkg`, `pkg[features]`
+	 * - namespaced git shorthand: `github:user/repo[#ref]`, `gitlab:`, `bitbucket:`,
+	 *   `codeberg:`, `sourcehut:`/`srht:`
+	 * - full git URLs: `https://github.com/user/repo`, `git@github.com:user/repo`,
+	 *   `ssh://…`, `git+https://…`
+	 *
+	 * For git specs the package name is not knowable from the spec, so the
+	 * installer diffs `plugins/package.json` `dependencies` before and after
+	 * to find the newly added key.
 	 *
 	 * @param specString - Package specifier with optional features: "pkg", "pkg[feat]", "pkg[*]", "pkg[]"
 	 * @param options - Install options
@@ -140,7 +197,12 @@ export class PluginManager {
 	 */
 	async install(specString: string, options: InstallOptions = {}): Promise<InstalledPlugin> {
 		const spec = parsePluginSpec(specString);
-		validatePackageName(spec.packageName);
+		const gitSource = parseGitUrl(spec.packageName);
+		if (gitSource) {
+			validateGitSpec(spec.packageName);
+		} else {
+			validatePackageName(spec.packageName);
+		}
 
 		await this.#ensurePackageJson();
 
@@ -154,9 +216,12 @@ export class PluginManager {
 				enabled: true,
 			};
 		}
+		const pkgJsonPath = getPluginsPackageJson();
+		const depsBefore = gitSource ? await this.#readDeps(pkgJsonPath) : {};
+		const packageInstallSpec = gitSource ? gitInstallSpec(spec.packageName, gitSource) : spec.packageName;
 
 		// Run npm install
-		const proc = Bun.spawn(["bun", "install", spec.packageName], {
+		const proc = Bun.spawn(["bun", "install", packageInstallSpec], {
 			cwd: getPluginsDir(),
 			stdin: "ignore",
 			stdout: "pipe",
@@ -169,9 +234,40 @@ export class PluginManager {
 			const stderr = await new Response(proc.stderr).text();
 			throw new Error(`npm install failed: ${stderr}`);
 		}
-
-		// Resolve actual package name (strip version specifier)
-		const actualName = extractPackageName(spec.packageName);
+		// Resolve actual package name. npm specs encode the name (strip version);
+		// git specs do not, so diff plugins/package.json deps to find the new entry.
+		let actualName: string;
+		if (gitSource) {
+			const depsAfter = await this.#readDeps(pkgJsonPath);
+			let resolved: string | undefined;
+			for (const key of Object.keys(depsAfter)) {
+				if (!(key in depsBefore)) {
+					resolved = key;
+					break;
+				}
+			}
+			// Fallback: a force-reinstall of an already-present git plugin will not
+			// add a new key, just rewrite the existing one to the new spec value.
+			// Match by the install value for force-reinstalls where no new key is
+			// added (non-GitHub shorthands are normalized before bun sees them).
+			if (!resolved) {
+				const needle = packageInstallSpec.replace(/^git\+/i, "");
+				for (const [key, value] of Object.entries(depsAfter)) {
+					if (typeof value === "string" && value.includes(needle)) {
+						resolved = key;
+						break;
+					}
+				}
+			}
+			if (!resolved) {
+				throw new Error(
+					`Installed ${spec.packageName} but could not determine package name from plugins/package.json`,
+				);
+			}
+			actualName = resolved;
+		} else {
+			actualName = extractPackageName(spec.packageName);
+		}
 		const pkgPath = path.join(getPluginsNodeModules(), actualName, "package.json");
 
 		let pkg: { name: string; version: string; omp?: PluginManifest; pi?: PluginManifest };

package/src/extensibility/shared-events.ts

@@ -204,13 +204,13 @@ export interface TurnEndEvent {
 export interface AutoCompactionStartEvent {
 	type: "auto_compaction_start";
 	reason: "threshold" | "overflow" | "idle" | "incomplete";
-	action: "context-full" | "handoff";
+	action: "context-full" | "handoff" | "shake" | "shake-summary";
 }
 
 /** Fired when auto-compaction ends */
 export interface AutoCompactionEndEvent {
 	type: "auto_compaction_end";
-	action: "context-full" | "handoff";
+	action: "context-full" | "handoff" | "shake" | "shake-summary";
 	result: CompactionResult | undefined;
 	aborted: boolean;
 	willRetry: boolean;