omp-auto-loop 0.1.1 → 0.2.1

package/README.md

@@ -1,50 +1,78 @@
-# 🔄 Agent Loop
+# omp-auto-loop
 
-General-purpose loop extension for pi. Repeats agent turns automatically in three modes: until a goal is met, for a fixed number of passes, or through a sequence of named pipeline stages.
+General-purpose agent loop extension for the Oh-My-Pi coding agent. When you send a normal (non-slash) message it becomes the loop goal and the extension runs repeated agent turns until the goal is finished or you stop the loop.
 
 ## Install
 
 ```bash
-pi install npm:pi-agent-loop
+omp plugin install omp-auto-loop
 ```
 
-## What it does
+## Overview
 
-Registers a `/loop` command and a `loop_control` tool. On each iteration, the agent does its work then calls `loop_control` to either advance to the next iteration (`next`) or declare the goal complete (`done`). The loop context is injected into the system prompt automatically so the agent always knows where it is.
+- Starts a loop when the user sends a normal message (one that does NOT start with `/`). That message becomes the loop `goal`.
+- Injects a short loop-focused system prompt so the agent knows the current iteration and goal.
+- Provides a tool named `loop_control` which the agent must call to indicate progress: use `status: "next"` to continue, or `status: "done"` to finish.
+- Registers a `loop-stop` command and a `Ctrl+Shift+X` shortcut to stop the active loop from the UI.
 
-## Commands
+## Usage
 
-| Command | Description |
-|---|---|
-| `/loop goal <description>` | Repeat until the LLM declares the goal met (open-ended) |
-| `/loop passes <N> <task>` | Run exactly N passes |
-| `/loop pipeline <s1\|s2\|s3> <goal>` | Run named stages sequentially, stop after the last |
-| `/loop-stop` | Stop the active loop immediately |
+- Start a loop: send a normal message like `Refactor all test files to use the new assertion API` (do NOT prefix with `/`). The extension will begin iterating on that goal.
+- One-off message: prefix with `/once ` to send a single, non-looping turn (e.g. `/once Quick status check`).
+- Stop loop: run the `loop-stop` command or press `Ctrl+Shift+X` to abort the active loop immediately.
 
-## Shortcut
+## Tool: `loop_control`
 
-| Shortcut | Description |
-|---|---|
-| `Ctrl+Shift+X` | Emergency abort — stops the loop and cancels the current turn |
+Call this tool from the assistant to signal what happened during the iteration. The tool is only meaningful while a loop is active.
 
-## Tool: `loop_control`
+Parameters:
+
+- `status` — string: either `"next"` or `"done"`.
+- `summary` — string: brief summary of what was accomplished this iteration.
+- `reason` — optional string: a reason explaining why the goal is complete (used with `"done"`).
+
+Behavior notes:
 
-The LLM calls this tool to signal progress. It is only active during a loop.
+- When the model calls `loop_control` with `status: "next"`, the extension advances the iteration count and schedules the next iteration (it sends a steer message with the updated loop prompt).
+- When the model calls `loop_control` with `status: "done"`, the extension asks for confirmation (it sets a `confirmingDone` flag and returns a prompt asking the model to either call `loop_control` again or finish the response). The loop is finalized if the model finishes the response (skips calling `loop_control` again) or explicitly confirms by calling `loop_control` with `done` again.
 
-| Parameter | Type | Description |
-|---|---|---|
-| `status` | `"next" \| "done"` | Advance to next iteration or declare completion |
-| `summary` | `string` | Brief summary of what was accomplished this iteration |
-| `reason` | `string?` | Why the goal is met (used with `"done"`) |
+Fallbacks:
+
+- If the agent ends its turn without calling `loop_control` and without scheduling the next step, the extension increments the iteration count and sends a steer message reminding the agent to call `loop_control` (so work won't silently stop).
 
 ## Examples
 
+- Start by sending a goal (normal message):
+
 ```
-/loop goal Refactor all test files to use the new assertion API
-/loop passes 3 Review and improve the README
-/loop pipeline analyze|implement|test Write and test the new auth module
+Refactor all test files to use the new assertion API
 ```
 
-## TUI widget
+- Example tool call (assistant -> tool):
+
+```json
+{
+	"toolName": "loop_control",
+	"params": { "status": "next", "summary": "Refactored 3 files; more work remains" }
+}
+```
+
+- Marking completion (assistant -> tool):
+
+```json
+{
+	"toolName": "loop_control",
+	"params": { "status": "done", "summary": "All tests updated and passing" }
+}
+```
+
+If the extension asks for confirmation after a `done` call, finish the response (or call `loop_control` with `done` again) to finalize the loop.
+
+## UI
+
+- While a loop is active the status bar and a small widget show the current iteration and provide the `Ctrl+Shift+X` stop shortcut.
+
+## Notes
 
-While a loop is active, a status bar entry and widget show the current mode, goal, and iteration count.
+- Package name: `omp-auto-loop` (see `package.json`).
+- Peer-dep: this extension expects to run within `@oh-my-pi/pi-coding-agent`.

package/index.ts

@@ -1,7 +1,7 @@
 import type { ExtensionAPI, ExtensionContext } from "@oh-my-pi/pi-coding-agent";
 import { Key } from "@oh-my-pi/pi-tui";
-import { buildPrompt, emptyState, getSystemPromptAddition, type LoopState, updateWidget } from "./state.js";
-import { getLoopControlToolDefinition, handleLoopControlTool, renderLoopControlCall, renderLoopControlResult } from "./tool.js";
+import { buildPrompt, emptyState, getSystemPromptAddition, type LoopState, updateWidget } from "./state.ts";
+import { getLoopControlToolDefinition, handleLoopControlTool, renderLoopControlCall, renderLoopControlResult } from "./tool.ts";
 
 export default function (pi: ExtensionAPI) {
 	let state = emptyState();

package/package.json

@@ -1,14 +1,21 @@
 {
 	"name": "omp-auto-loop",
-	"version": "0.1.1",
-	"description": "General-purpose agent loop extension for pi. Supports goal loops (repeat until done), fixed-pass loops, and multi-stage pipelines via /loop goal, /loop passes, and /loop pipeline commands.",
+	"version": "0.2.1",
+	"description": "General-purpose auto agent loop plugin for omp.",
+	"type": "module",
+	"main": "index.ts",
 	"keywords": [
-		"pi-package"
+		"omp-plugin"
 	],
 	"license": "MIT",
 	"pi": {
 		"extensions": [
-			"."
+			"index.ts"
+		]
+	},
+	"omp": {
+		"extensions": [
+			"index.ts"
 		]
 	},
 	"peerDependencies": {

package/tool.ts

@@ -2,7 +2,7 @@ import { StringEnum } from "@oh-my-pi/pi-ai";
 import type { ExtensionAPI, ExtensionContext } from "@oh-my-pi/pi-coding-agent";
 import { Text } from "@oh-my-pi/pi-tui";
 import { Type } from "@sinclair/typebox";
-import { buildPrompt, type LoopState } from "./state.js";
+import { buildPrompt, type LoopState } from "./state.ts";
 
 export function handleLoopControlTool(opts: {
         params: { status: "next" | "done"; summary: string; reason?: string };