pi-goal-x 0.9.0 → 0.10.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/extensions/goal-auditor.ts +94 -12
- package/extensions/widgets/goal-widget.ts +48 -5
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -27,7 +27,7 @@ All core features of [@capyup/pi-goal](https://github.com/capyup/pi-goal) are pr
|
|
|
27
27
|
|
|
28
28
|
### Completion auditor
|
|
29
29
|
|
|
30
|
-
- **Live progress widget** — when the auditor runs, the TUI shows a spinner, the current tool being executed, and recent output lines. No more wondering if anything is happening.
|
|
30
|
+
- **Live progress widget** — when the auditor runs, the TUI shows a spinner, a progress bar (`[████░░░░] 40%`), step labels (`Inspecting files...`, `Verifying success criteria...`), the current tool being executed, and recent output lines. No more wondering if anything is happening.
|
|
31
31
|
- **Escape to skip** — press Escape during an audit to abort it and complete the goal immediately. The skip is recorded in the ledger as `audit_skipped` with reason `user_aborted` and auditor model metadata.
|
|
32
32
|
- **Disable the auditor entirely** — set `disabled: true` in `.pi/goal-auditor.json` (or toggle it via `/goal-settings` → `disabled`). The agent can still bypass with user confirmation by passing `confirmBypassAuditor: true` to `update_goal`.
|
|
33
33
|
- **Skipped audits are recorded** — every skip (whether disabled or Escape-aborted) is logged to the ledger with the reason, provider, model, and thinking level for full traceability.
|
|
@@ -1,10 +1,13 @@
|
|
|
1
1
|
import * as fs from "node:fs";
|
|
2
2
|
import * as path from "node:path";
|
|
3
|
+
import type { Static } from "@earendil-works/pi-ai";
|
|
4
|
+
import { Type } from "@earendil-works/pi-ai";
|
|
3
5
|
import type { ThinkingLevel } from "@earendil-works/pi-agent-core";
|
|
4
6
|
import type { Model } from "@earendil-works/pi-ai";
|
|
5
7
|
import {
|
|
6
8
|
createAgentSession,
|
|
7
9
|
createExtensionRuntime,
|
|
10
|
+
defineTool,
|
|
8
11
|
SessionManager,
|
|
9
12
|
SettingsManager,
|
|
10
13
|
type ExtensionContext,
|
|
@@ -29,9 +32,13 @@ export interface AuditorProgress {
|
|
|
29
32
|
/** Recent text output lines from the auditor's assistant messages */
|
|
30
33
|
recentOutput: string[];
|
|
31
34
|
/** Phase of the audit */
|
|
32
|
-
phase: "running" | "tool_executing" | "producing_report" | "done";
|
|
35
|
+
phase: "running" | "tool_executing" | "producing_report" | "thinking" | "done";
|
|
33
36
|
/** Elapsed ms since audit started */
|
|
34
37
|
elapsedMs: number;
|
|
38
|
+
/** Current step label shown to the user (e.g. "Inspecting files...") */
|
|
39
|
+
label?: string;
|
|
40
|
+
/** Completion percentage from 0 to 100 */
|
|
41
|
+
percentage?: number;
|
|
35
42
|
}
|
|
36
43
|
|
|
37
44
|
export type AuditorProgressCallback = (progress: AuditorProgress) => void;
|
|
@@ -156,9 +163,28 @@ export function buildGoalAuditorPrompt(args: {
|
|
|
156
163
|
"2. Inspect artifacts or command output that can prove or disprove those criteria.",
|
|
157
164
|
"3. Explain missing or weak evidence, especially scaffold-vs-final quality gaps.",
|
|
158
165
|
"4. End with exactly <approved/> only if the objective is truly complete; otherwise end with exactly <disapproved/>.",
|
|
166
|
+
"",
|
|
167
|
+
"Progress reporting:",
|
|
168
|
+
"You have the report_auditor_progress tool available to report your progress to the user.",
|
|
169
|
+
"Please use it at natural phase boundaries:",
|
|
170
|
+
" - When starting: report_auditor_progress(label='Starting audit...', percentage=0)",
|
|
171
|
+
" - When beginning file inspection: report_auditor_progress(label='Inspecting files...', percentage=25)",
|
|
172
|
+
" - When verifying success criteria: report_auditor_progress(label='Verifying success criteria...', percentage=50)",
|
|
173
|
+
" - When evaluating evidence: report_auditor_progress(label='Evaluating evidence...', percentage=75)",
|
|
174
|
+
" - When producing final report: report_auditor_progress(label='Producing report...', percentage=90)",
|
|
175
|
+
"This is purely for user visibility and does not affect the audit outcome.",
|
|
159
176
|
].join("\n");
|
|
160
177
|
}
|
|
161
178
|
|
|
179
|
+
/** Tool name for auditor progress reporting */
|
|
180
|
+
export const REPORT_AUDITOR_PROGRESS_TOOL_NAME = "report_auditor_progress";
|
|
181
|
+
|
|
182
|
+
/** Parameters for the report_auditor_progress tool */
|
|
183
|
+
export const reportAuditorProgressParams = Type.Object({
|
|
184
|
+
label: Type.String({ description: "Current step label describing what the auditor is doing (e.g. 'Inspecting files...', 'Verifying success criteria...', 'Producing report...')" }),
|
|
185
|
+
percentage: Type.Number({ description: "Completion percentage from 0 to 100", minimum: 0, maximum: 100 }),
|
|
186
|
+
});
|
|
187
|
+
|
|
162
188
|
function makeAuditorResourceLoader(): ResourceLoader {
|
|
163
189
|
return {
|
|
164
190
|
getExtensions: () => ({ extensions: [], errors: [], runtime: createExtensionRuntime() }),
|
|
@@ -170,9 +196,14 @@ function makeAuditorResourceLoader(): ResourceLoader {
|
|
|
170
196
|
"You are a read-only completion auditor running in an isolated pi agent session.",
|
|
171
197
|
"Inspect the repository and decide whether the claimed goal completion is genuinely satisfied.",
|
|
172
198
|
"Never modify files. Never approve unless the actual user objective is complete.",
|
|
199
|
+
"",
|
|
200
|
+
"You have the report_auditor_progress tool available. Use it to report your audit progress",
|
|
201
|
+
"to the user at natural phase boundaries (starting, inspecting files, verifying criteria,",
|
|
202
|
+
"producing report). This helps the user understand what the auditor is doing and how far",
|
|
203
|
+
"along it is.",
|
|
173
204
|
].join("\n"),
|
|
174
205
|
getAppendSystemPrompt: () => [],
|
|
175
|
-
|
|
206
|
+
extendResources: () => {},
|
|
176
207
|
reload: async () => {},
|
|
177
208
|
};
|
|
178
209
|
}
|
|
@@ -228,16 +259,6 @@ export async function runGoalCompletionAuditor(args: {
|
|
|
228
259
|
}
|
|
229
260
|
try {
|
|
230
261
|
const createSession = args.createSession ?? createAgentSession;
|
|
231
|
-
const { session } = await createSession({
|
|
232
|
-
cwd: args.ctx.cwd,
|
|
233
|
-
model,
|
|
234
|
-
thinkingLevel,
|
|
235
|
-
modelRegistry: args.ctx.modelRegistry,
|
|
236
|
-
resourceLoader: makeAuditorResourceLoader(),
|
|
237
|
-
sessionManager: SessionManager.inMemory(args.ctx.cwd),
|
|
238
|
-
settingsManager: SettingsManager.inMemory({ compaction: { enabled: false } }),
|
|
239
|
-
tools: ["read", "grep", "find", "ls", "bash"],
|
|
240
|
-
});
|
|
241
262
|
const startedAt = Date.now();
|
|
242
263
|
const progress: AuditorProgress = {
|
|
243
264
|
recentOutput: [],
|
|
@@ -248,6 +269,49 @@ export async function runGoalCompletionAuditor(args: {
|
|
|
248
269
|
progress.elapsedMs = Date.now() - startedAt;
|
|
249
270
|
args.onProgress?.({ ...progress });
|
|
250
271
|
}
|
|
272
|
+
|
|
273
|
+
// Build the report_auditor_progress tool, capturing the progress state
|
|
274
|
+
const reportProgressTool = defineTool({
|
|
275
|
+
name: REPORT_AUDITOR_PROGRESS_TOOL_NAME,
|
|
276
|
+
label: "Report Auditor Progress",
|
|
277
|
+
description: "Report current progress of the audit to the user. Call this at natural phase boundaries (starting, inspecting files, verifying criteria, producing report) to keep the user informed.",
|
|
278
|
+
promptSnippet: "Report current audit progress (step label and completion percentage) to the user.",
|
|
279
|
+
promptGuidelines: [
|
|
280
|
+
"Use report_auditor_progress at natural phase boundaries during the audit:",
|
|
281
|
+
" - When starting the audit: label='Starting audit...' percentage=0",
|
|
282
|
+
" - When beginning file inspection: label='Inspecting files...' percentage=25",
|
|
283
|
+
" - When verifying success criteria: label='Verifying success criteria...' percentage=50",
|
|
284
|
+
" - When evaluating evidence: label='Evaluating evidence...' percentage=75",
|
|
285
|
+
" - When producing final report: label='Producing report...' percentage=90",
|
|
286
|
+
"This is purely for user visibility — it does not affect the audit outcome.",
|
|
287
|
+
"Do not call this tool more than once every few seconds to avoid flooding.",
|
|
288
|
+
],
|
|
289
|
+
parameters: reportAuditorProgressParams,
|
|
290
|
+
executionMode: "sequential",
|
|
291
|
+
async execute(_toolCallId, params) {
|
|
292
|
+
const { label, percentage } = params as Static<typeof reportAuditorProgressParams>;
|
|
293
|
+
progress.label = label;
|
|
294
|
+
progress.percentage = percentage;
|
|
295
|
+
progress.phase = "running";
|
|
296
|
+
emitProgress();
|
|
297
|
+
return {
|
|
298
|
+
content: [{ type: "text", text: `Progress reported: ${label} (${percentage}%)` }],
|
|
299
|
+
details: {},
|
|
300
|
+
};
|
|
301
|
+
},
|
|
302
|
+
});
|
|
303
|
+
|
|
304
|
+
const { session } = await createSession({
|
|
305
|
+
cwd: args.ctx.cwd,
|
|
306
|
+
model,
|
|
307
|
+
thinkingLevel,
|
|
308
|
+
modelRegistry: args.ctx.modelRegistry,
|
|
309
|
+
resourceLoader: makeAuditorResourceLoader(),
|
|
310
|
+
sessionManager: SessionManager.inMemory(args.ctx.cwd),
|
|
311
|
+
settingsManager: SettingsManager.inMemory({ compaction: { enabled: false } }),
|
|
312
|
+
tools: ["read", "grep", "find", "ls", "bash", REPORT_AUDITOR_PROGRESS_TOOL_NAME],
|
|
313
|
+
customTools: [reportProgressTool],
|
|
314
|
+
});
|
|
251
315
|
const unsubscribe = session.subscribe((event) => {
|
|
252
316
|
if (event.type === "tool_execution_start") {
|
|
253
317
|
progress.currentTool = event.toolName;
|
|
@@ -268,6 +332,20 @@ export async function runGoalCompletionAuditor(args: {
|
|
|
268
332
|
return;
|
|
269
333
|
}
|
|
270
334
|
if (event.type === "message_update") {
|
|
335
|
+
// Check for thinking events from the assistant stream
|
|
336
|
+
const streamEvent = (event as any).assistantMessageEvent;
|
|
337
|
+
if (streamEvent?.type === "thinking_start") {
|
|
338
|
+
progress.phase = "thinking";
|
|
339
|
+
if (!progress.label) progress.label = "Analyzing goal...";
|
|
340
|
+
emitProgress();
|
|
341
|
+
return;
|
|
342
|
+
}
|
|
343
|
+
if (streamEvent?.type === "thinking_end") {
|
|
344
|
+
progress.phase = "running";
|
|
345
|
+
emitProgress();
|
|
346
|
+
return;
|
|
347
|
+
}
|
|
348
|
+
// For text content, show producing_report phase
|
|
271
349
|
progress.phase = "producing_report";
|
|
272
350
|
const message = event.message as any;
|
|
273
351
|
if (message?.role === "assistant") {
|
|
@@ -300,6 +378,8 @@ export async function runGoalCompletionAuditor(args: {
|
|
|
300
378
|
args.signal?.addEventListener("abort", abortSession, { once: true });
|
|
301
379
|
|
|
302
380
|
// Emit initial progress
|
|
381
|
+
progress.label = "Starting audit...";
|
|
382
|
+
progress.percentage = 0;
|
|
303
383
|
emitProgress();
|
|
304
384
|
try {
|
|
305
385
|
if (args.signal?.aborted) return { approved: false, disapproved: true, output: "", model: modelLabel(model), thinkingLevel, error: "Auditor aborted." };
|
|
@@ -307,6 +387,8 @@ export async function runGoalCompletionAuditor(args: {
|
|
|
307
387
|
} finally {
|
|
308
388
|
args.signal?.removeEventListener("abort", abortSession);
|
|
309
389
|
progress.phase = "done";
|
|
390
|
+
progress.label = "Audit complete.";
|
|
391
|
+
progress.percentage = 100;
|
|
310
392
|
emitProgress();
|
|
311
393
|
unsubscribe();
|
|
312
394
|
}
|
|
@@ -24,8 +24,12 @@ export interface AuditorWidgetProgress {
|
|
|
24
24
|
currentToolArgs?: string;
|
|
25
25
|
currentToolStartedAt?: number;
|
|
26
26
|
recentOutput: string[];
|
|
27
|
-
phase: "running" | "tool_executing" | "producing_report" | "done";
|
|
27
|
+
phase: "running" | "tool_executing" | "producing_report" | "thinking" | "done";
|
|
28
28
|
elapsedMs: number;
|
|
29
|
+
/** Current step label shown to the user */
|
|
30
|
+
label?: string;
|
|
31
|
+
/** Completion percentage from 0 to 100 */
|
|
32
|
+
percentage?: number;
|
|
29
33
|
}
|
|
30
34
|
|
|
31
35
|
export interface GoalWidgetOptions {
|
|
@@ -52,6 +56,13 @@ function branchLine(theme: Theme, width: number, isLast: boolean, content: strin
|
|
|
52
56
|
return fit(`${theme.fg("dim", prefix)} ${content}`, width);
|
|
53
57
|
}
|
|
54
58
|
|
|
59
|
+
function progressBar(pct: number, barWidth: number, theme: Theme): string {
|
|
60
|
+
const safeBar = Math.max(3, barWidth);
|
|
61
|
+
const filled = Math.min(safeBar, Math.max(0, Math.round((pct / 100) * safeBar)));
|
|
62
|
+
const empty = safeBar - filled;
|
|
63
|
+
return `[${theme.fg("accent", "█".repeat(filled))}${theme.fg("dim", "░".repeat(empty))}]`;
|
|
64
|
+
}
|
|
65
|
+
|
|
55
66
|
function displayIcon(goal: GoalWidgetRecord): { icon: string; color: GoalWidgetColor; label: string } {
|
|
56
67
|
if (goal.status === "complete") return { icon: "✓", color: "success", label: "complete" };
|
|
57
68
|
if (goal.status === "paused") {
|
|
@@ -81,8 +92,17 @@ function spinnerFrame(): string {
|
|
|
81
92
|
export function renderAuditorWidgetLines(progress: AuditorWidgetProgress, theme: Theme, width: number): string[] {
|
|
82
93
|
const safeWidth = Math.max(1, width);
|
|
83
94
|
const isActive = progress.phase !== "done";
|
|
84
|
-
const
|
|
85
|
-
const
|
|
95
|
+
const isThinking = progress.phase === "thinking";
|
|
96
|
+
const icon = isActive
|
|
97
|
+
? isThinking
|
|
98
|
+
? theme.fg("muted", "⟡")
|
|
99
|
+
: theme.fg("accent", spinnerFrame())
|
|
100
|
+
: theme.fg("success", "✓");
|
|
101
|
+
const label = isActive
|
|
102
|
+
? isThinking
|
|
103
|
+
? "thinking..."
|
|
104
|
+
: "auditing"
|
|
105
|
+
: "audit complete";
|
|
86
106
|
// formatDuration expects seconds, progress.elapsedMs is in milliseconds
|
|
87
107
|
const duration = formatDuration(Math.floor(progress.elapsedMs / 1000));
|
|
88
108
|
const lines: string[] = [
|
|
@@ -94,7 +114,30 @@ export function renderAuditorWidgetLines(progress: AuditorWidgetProgress, theme:
|
|
|
94
114
|
),
|
|
95
115
|
];
|
|
96
116
|
|
|
97
|
-
|
|
117
|
+
// Show step label when available
|
|
118
|
+
if (progress.label) {
|
|
119
|
+
lines.push(branchLine(
|
|
120
|
+
theme,
|
|
121
|
+
safeWidth,
|
|
122
|
+
false,
|
|
123
|
+
`${theme.fg("text", truncateText(progress.label, Math.max(8, safeWidth - 6)))}`,
|
|
124
|
+
));
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
// Show progress bar when percentage is available
|
|
128
|
+
if (typeof progress.percentage === "number") {
|
|
129
|
+
const barWidth = Math.max(6, Math.min(safeWidth - 10, 30));
|
|
130
|
+
const bar = progressBar(progress.percentage, barWidth, theme);
|
|
131
|
+
const pct = `${theme.fg("muted", `${Math.round(progress.percentage)}%`)}`;
|
|
132
|
+
lines.push(branchLine(
|
|
133
|
+
theme,
|
|
134
|
+
safeWidth,
|
|
135
|
+
isActive && !progress.currentTool && progress.recentOutput.length === 0 && !isThinking,
|
|
136
|
+
`${bar} ${pct}`,
|
|
137
|
+
));
|
|
138
|
+
}
|
|
139
|
+
|
|
140
|
+
if (isActive && !isThinking && progress.currentTool) {
|
|
98
141
|
const argText = progress.currentToolArgs
|
|
99
142
|
? truncateText(progress.currentToolArgs, Math.max(10, safeWidth - 24))
|
|
100
143
|
: "";
|
|
@@ -129,7 +172,7 @@ export function renderAuditorWidgetLines(progress: AuditorWidgetProgress, theme:
|
|
|
129
172
|
}
|
|
130
173
|
|
|
131
174
|
// Show skip hint when audit is actively running
|
|
132
|
-
if (isActive) {
|
|
175
|
+
if (isActive && !isThinking) {
|
|
133
176
|
lines.push(branchLine(
|
|
134
177
|
theme,
|
|
135
178
|
safeWidth,
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pi-goal-x",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.10.0",
|
|
4
4
|
"description": "Goal mode extension for pi: persistent long-running objectives, /goal-set drafting, Sisyphus prompt style, autoContinue, and an above-editor status overlay. Fork of @capyup/pi-goal.",
|
|
5
5
|
"license": "MIT",
|
|
6
6
|
"author": "pi-goal-x contributors",
|