@mrclrchtr/supi-review 3.0.0 → 3.2.0
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 +41 -20
- package/node_modules/@mrclrchtr/supi-core/package.json +1 -1
- package/package.json +2 -2
- package/src/git-choices.ts +42 -0
- package/src/git-command.ts +172 -2
- package/src/git.ts +126 -291
- package/src/history/collect.ts +15 -6
- package/src/review-input.ts +36 -0
- package/src/review-limits.ts +12 -0
- package/src/review-path.ts +18 -12
- package/src/review-result.ts +26 -5
- package/src/review.ts +201 -76
- package/src/session/review-artifact-store.ts +97 -0
- package/src/session/review-plan-store.ts +93 -12
- package/src/target/change-metadata.ts +125 -0
- package/src/target/diff.ts +43 -0
- package/src/target/file-manifest.ts +40 -0
- package/src/target/packet.ts +17 -6
- package/src/target/resolve.ts +209 -0
- package/src/tool/agent-review-schemas.ts +21 -9
- package/src/tool/agent-review-tools.ts +66 -17
- package/src/tool/child-failure-diagnostics.ts +35 -0
- package/src/tool/child-lifecycle-trace.ts +27 -11
- package/src/tool/child-session-runner.ts +28 -11
- package/src/tool/child-usage.ts +73 -0
- package/src/tool/diagnostic-sanitizer.ts +35 -0
- package/src/tool/output-page.ts +26 -0
- package/src/tool/planner-runner.ts +28 -10
- package/src/tool/review-output-tool.ts +93 -0
- package/src/tool/review-runner.ts +9 -7
- package/src/tool/review-tools.ts +90 -12
- package/src/tool/review-workflow.ts +163 -83
- package/src/tool/schemas.ts +24 -5
- package/src/tool/session-lifecycle.ts +44 -54
- package/src/tool/usage-format.ts +15 -0
- package/src/tui/common.ts +58 -1
- package/src/tui/prepare.ts +57 -31
- package/src/tui/run.ts +13 -3
- package/src/types.ts +62 -19
|
@@ -1,9 +1,11 @@
|
|
|
1
|
+
import type { Usage } from "@earendil-works/pi-ai";
|
|
1
2
|
import { buildSessionContext, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
|
|
2
3
|
import { StatusSpinner } from "@mrclrchtr/supi-core/status-spinner";
|
|
3
4
|
import { loadReviewConfig } from "../config.ts";
|
|
4
5
|
import { summarizeReviewSnapshot } from "../git.ts";
|
|
5
6
|
import { collectPlannerContext } from "../history/collect.ts";
|
|
6
7
|
import { resolveAgentReviewModel } from "../model.ts";
|
|
8
|
+
import type { ReviewArtifactStore } from "../session/review-artifact-store.ts";
|
|
7
9
|
import type { ReviewPlanStore } from "../session/review-plan-store.ts";
|
|
8
10
|
import { renderPrepareCall, renderPrepareResult } from "../tui/prepare.ts";
|
|
9
11
|
import { renderRunCall, renderRunResult } from "../tui/run.ts";
|
|
@@ -15,25 +17,40 @@ import {
|
|
|
15
17
|
prepareReviewSchema,
|
|
16
18
|
runReviewSchema,
|
|
17
19
|
} from "./agent-review-schemas.ts";
|
|
18
|
-
import {
|
|
20
|
+
import { formatChildFailureDiagnostics } from "./child-failure-diagnostics.ts";
|
|
21
|
+
import { createReviewOutput } from "./review-output-tool.ts";
|
|
19
22
|
import { prepareReview, runReview } from "./review-workflow.ts";
|
|
23
|
+
import { formatReviewUsage } from "./usage-format.ts";
|
|
20
24
|
|
|
21
25
|
function target(input: PrepareReviewToolInput["target"]): ReviewTargetSpec {
|
|
22
26
|
return (input ?? { kind: "working-tree" }) as ReviewTargetSpec;
|
|
23
27
|
}
|
|
24
28
|
|
|
29
|
+
function plannerFailureReason(failure: {
|
|
30
|
+
kind: string;
|
|
31
|
+
failureCode?: string;
|
|
32
|
+
timeoutMs?: number;
|
|
33
|
+
}): string {
|
|
34
|
+
if (failure.kind === "failed") return failure.failureCode ?? "failed";
|
|
35
|
+
if (failure.kind === "timeout") return `timeout (${failure.timeoutMs} ms)`;
|
|
36
|
+
return failure.kind;
|
|
37
|
+
}
|
|
38
|
+
|
|
25
39
|
function formatPrepared(plan: {
|
|
26
40
|
id: string;
|
|
27
|
-
snapshot: { title: string;
|
|
41
|
+
snapshot: { title: string; changes: Array<{ path: string }> };
|
|
28
42
|
plannerDraft?: { sharedContext?: string; tasks: Array<{ id: string; instructions: string }> };
|
|
43
|
+
plannerFailure?: { kind: string; failureCode?: string; timeoutMs?: number };
|
|
44
|
+
plannerUsage?: Usage;
|
|
29
45
|
}): string {
|
|
30
46
|
const lines = [
|
|
31
47
|
"# Review Plan Prepared",
|
|
32
48
|
"",
|
|
33
49
|
`Plan ID: ${plan.id}`,
|
|
34
50
|
`Target: ${plan.snapshot.title}`,
|
|
35
|
-
`Files changed: ${plan.snapshot.
|
|
51
|
+
`Files changed: ${plan.snapshot.changes.length}`,
|
|
36
52
|
];
|
|
53
|
+
if (plan.plannerUsage) lines.push(`Planner usage: ${formatReviewUsage(plan.plannerUsage)}`);
|
|
37
54
|
if (plan.plannerDraft) {
|
|
38
55
|
lines.push("", "## Planner Draft");
|
|
39
56
|
if (plan.plannerDraft.sharedContext) lines.push("", plan.plannerDraft.sharedContext);
|
|
@@ -42,6 +59,10 @@ function formatPrepared(plan: {
|
|
|
42
59
|
}
|
|
43
60
|
lines.push("", "Call supi_review_run with an explicit accept-draft or use-review decision.");
|
|
44
61
|
} else {
|
|
62
|
+
if (plan.plannerFailure) {
|
|
63
|
+
const reason = plannerFailureReason(plan.plannerFailure);
|
|
64
|
+
lines.push("", `Planner unavailable: ${reason}. The plan remains usable without a draft.`);
|
|
65
|
+
}
|
|
45
66
|
lines.push("", "Call supi_review_run with a use-review decision containing one to four tasks.");
|
|
46
67
|
}
|
|
47
68
|
return lines.join("\n");
|
|
@@ -54,10 +75,27 @@ function formatTaskResult(result: ReviewTaskResult): string[] {
|
|
|
54
75
|
`Model: ${result.modelId}`,
|
|
55
76
|
`Packet SHA-256: ${result.packetHash}`,
|
|
56
77
|
];
|
|
57
|
-
if (result.
|
|
58
|
-
if (result.status === "
|
|
78
|
+
if (result.usage) lines.push(`Usage: ${formatReviewUsage(result.usage)}`);
|
|
79
|
+
if (result.status === "failed") {
|
|
80
|
+
lines.push(`Status: failed (${result.failureCode})`);
|
|
81
|
+
if (result.diagnostics) {
|
|
82
|
+
lines.push("", ...formatChildFailureDiagnostics(result.diagnostics));
|
|
83
|
+
}
|
|
84
|
+
return lines;
|
|
85
|
+
}
|
|
86
|
+
if (result.status === "canceled") {
|
|
87
|
+
lines.push("Status: canceled");
|
|
88
|
+
if (result.diagnostics) {
|
|
89
|
+
lines.push("", ...formatChildFailureDiagnostics(result.diagnostics));
|
|
90
|
+
}
|
|
91
|
+
return lines;
|
|
92
|
+
}
|
|
59
93
|
if (result.status === "timeout") {
|
|
60
|
-
|
|
94
|
+
lines.push(`Status: timeout (${result.timeoutMs} ms)`);
|
|
95
|
+
if (result.diagnostics) {
|
|
96
|
+
lines.push("", ...formatChildFailureDiagnostics(result.diagnostics));
|
|
97
|
+
}
|
|
98
|
+
return lines;
|
|
61
99
|
}
|
|
62
100
|
lines.push(`Verdict: ${result.verdict.toUpperCase()}`, "", result.summary);
|
|
63
101
|
for (const finding of result.findings) {
|
|
@@ -77,7 +115,7 @@ function formatTaskResult(result: ReviewTaskResult): string[] {
|
|
|
77
115
|
|
|
78
116
|
export function formatReviewBatch(details: ReviewBatchDetails): string {
|
|
79
117
|
const lines = [
|
|
80
|
-
"# Review
|
|
118
|
+
"# Review Finished",
|
|
81
119
|
"",
|
|
82
120
|
`Mode: ${details.mode}`,
|
|
83
121
|
`Provenance: ${details.provenance}`,
|
|
@@ -87,6 +125,9 @@ export function formatReviewBatch(details: ReviewBatchDetails): string {
|
|
|
87
125
|
lines.push(
|
|
88
126
|
`Planner: ${details.planning.modelId} (protocol ${details.planning.promptVersion})`,
|
|
89
127
|
`Planner decision: ${details.planning.decision}`,
|
|
128
|
+
...(details.planning.usage
|
|
129
|
+
? [`Planner usage: ${formatReviewUsage(details.planning.usage)}`]
|
|
130
|
+
: []),
|
|
90
131
|
);
|
|
91
132
|
}
|
|
92
133
|
for (const result of details.results) lines.push(...formatTaskResult(result));
|
|
@@ -99,7 +140,7 @@ function resolveModels(ctx: Parameters<Parameters<ExtensionAPI["registerTool"]>[
|
|
|
99
140
|
const planner = resolveAgentReviewModel(ctx, config.plannerModel);
|
|
100
141
|
if (!reviewer)
|
|
101
142
|
throw new Error(`Configured reviewer model "${config.agentModel}" is unavailable.`);
|
|
102
|
-
return { reviewer, planner
|
|
143
|
+
return { reviewer, planner };
|
|
103
144
|
}
|
|
104
145
|
|
|
105
146
|
/**
|
|
@@ -157,6 +198,7 @@ function initialTaskCount(
|
|
|
157
198
|
/** Factory for the supi_review_run execute function with animated status-bar spinner. */
|
|
158
199
|
function makeRunReviewExecute(
|
|
159
200
|
planStore: ReviewPlanStore,
|
|
201
|
+
artifactStore: ReviewArtifactStore,
|
|
160
202
|
): NonNullable<Parameters<ExtensionAPI["registerTool"]>[0]["execute"]> {
|
|
161
203
|
// biome-ignore lint/complexity/useMaxParams: Pi ToolDefinition execute signature
|
|
162
204
|
return async (_id, params, signal, onUpdate, ctx) => {
|
|
@@ -192,9 +234,11 @@ function makeRunReviewExecute(
|
|
|
192
234
|
onUpdate: wrappedUpdate,
|
|
193
235
|
});
|
|
194
236
|
if (outcome.kind !== "completed") throw new Error(outcome.reason);
|
|
237
|
+
const output = createReviewOutput(artifactStore, formatReviewBatch(outcome.details));
|
|
195
238
|
return {
|
|
196
|
-
content: [{ type: "text", text:
|
|
197
|
-
details: outcome.details,
|
|
239
|
+
content: [{ type: "text", text: output.text }],
|
|
240
|
+
details: { ...outcome.details, output: output.reference },
|
|
241
|
+
...(outcome.usage ? { usage: outcome.usage } : {}),
|
|
198
242
|
};
|
|
199
243
|
} finally {
|
|
200
244
|
statusSpinner.stop();
|
|
@@ -219,7 +263,11 @@ function throwPrepareFailure(
|
|
|
219
263
|
}
|
|
220
264
|
|
|
221
265
|
/** Register optional preparation and universal direct/prepared execution tools. */
|
|
222
|
-
export function registerAgentReviewTools(
|
|
266
|
+
export function registerAgentReviewTools(
|
|
267
|
+
pi: ExtensionAPI,
|
|
268
|
+
planStore: ReviewPlanStore,
|
|
269
|
+
artifactStore: ReviewArtifactStore,
|
|
270
|
+
): void {
|
|
223
271
|
pi.registerTool({
|
|
224
272
|
name: "supi_review_prepare",
|
|
225
273
|
label: "Prepare Review",
|
|
@@ -236,10 +284,6 @@ export function registerAgentReviewTools(pi: ExtensionAPI, planStore: ReviewPlan
|
|
|
236
284
|
async execute(_id, params, signal, onUpdate, ctx) {
|
|
237
285
|
const input = parsePrepareReviewToolInput(params);
|
|
238
286
|
const models = resolveModels(ctx);
|
|
239
|
-
if ((input.planning ?? "none") === "suggest" && !models.planner) {
|
|
240
|
-
throw new Error(`Configured Planner model "${models.config.plannerModel}" is unavailable.`);
|
|
241
|
-
}
|
|
242
|
-
|
|
243
287
|
const { statusSpinner, wrappedUpdate } = wireSpinnerToProgress(
|
|
244
288
|
ctx,
|
|
245
289
|
onUpdate,
|
|
@@ -263,8 +307,9 @@ export function registerAgentReviewTools(pi: ExtensionAPI, planStore: ReviewPlan
|
|
|
263
307
|
onUpdate: wrappedUpdate,
|
|
264
308
|
});
|
|
265
309
|
if (outcome.kind !== "prepared") throwPrepareFailure(outcome);
|
|
310
|
+
const output = createReviewOutput(artifactStore, formatPrepared(outcome.plan));
|
|
266
311
|
return {
|
|
267
|
-
content: [{ type: "text", text:
|
|
312
|
+
content: [{ type: "text", text: output.text }],
|
|
268
313
|
details: {
|
|
269
314
|
kind: "review-prepared",
|
|
270
315
|
planId: outcome.plan.id,
|
|
@@ -273,7 +318,11 @@ export function registerAgentReviewTools(pi: ExtensionAPI, planStore: ReviewPlan
|
|
|
273
318
|
plannerDraft: outcome.plan.plannerDraft,
|
|
274
319
|
plannerModelId: outcome.plan.plannerModelId,
|
|
275
320
|
plannerPromptVersion: outcome.plan.plannerPromptVersion,
|
|
321
|
+
plannerUsage: outcome.plan.plannerUsage,
|
|
322
|
+
plannerFailure: outcome.plan.plannerFailure,
|
|
323
|
+
output: output.reference,
|
|
276
324
|
},
|
|
325
|
+
...(outcome.usage ? { usage: outcome.usage } : {}),
|
|
277
326
|
};
|
|
278
327
|
} finally {
|
|
279
328
|
statusSpinner.stop();
|
|
@@ -293,7 +342,7 @@ export function registerAgentReviewTools(pi: ExtensionAPI, planStore: ReviewPlan
|
|
|
293
342
|
parameters: runReviewSchema,
|
|
294
343
|
renderCall: renderRunCall,
|
|
295
344
|
renderResult: renderRunResult,
|
|
296
|
-
execute: makeRunReviewExecute(planStore),
|
|
345
|
+
execute: makeRunReviewExecute(planStore, artifactStore),
|
|
297
346
|
});
|
|
298
347
|
|
|
299
348
|
pi.on("session_start", () => planStore.clear());
|
|
@@ -8,10 +8,12 @@ import type {
|
|
|
8
8
|
import {
|
|
9
9
|
type ChildLifecycleTrace,
|
|
10
10
|
ChildLifecycleTraceCollector,
|
|
11
|
+
extractLastLifecycleErrorText,
|
|
11
12
|
formatChildLifecycleTrace,
|
|
12
13
|
getRegisteredToolNames,
|
|
13
14
|
toSafeAssistantStopReason,
|
|
14
15
|
} from "./child-lifecycle-trace.ts";
|
|
16
|
+
import { sanitizeChildErrorText } from "./diagnostic-sanitizer.ts";
|
|
15
17
|
import { buildProgressTokens } from "./runner-helpers.ts";
|
|
16
18
|
|
|
17
19
|
/** Inputs used to create one safe non-success child-run diagnostic artifact. */
|
|
@@ -40,6 +42,8 @@ export function buildChildFailureDiagnostics(
|
|
|
40
42
|
recentActivity: input.recentActivity.length > 0 ? [...input.recentActivity] : undefined,
|
|
41
43
|
lastAssistantStopReason: lastAssistant?.stopReason,
|
|
42
44
|
lastAssistantToolCalls: lastAssistant?.toolCalls,
|
|
45
|
+
lastAssistantErrorText: lastAssistant?.errorText,
|
|
46
|
+
lastLifecycleErrorText: extractLastLifecycleErrorText(input.lifecycleTrace),
|
|
43
47
|
};
|
|
44
48
|
}
|
|
45
49
|
|
|
@@ -78,6 +82,28 @@ export function formatChildFailureCopy(stage: ChildStage, code: ChildFailureCode
|
|
|
78
82
|
}
|
|
79
83
|
}
|
|
80
84
|
|
|
85
|
+
export interface ChildDiagnosticErrorRow {
|
|
86
|
+
label: "Last assistant error" | "Lifecycle error";
|
|
87
|
+
text: string;
|
|
88
|
+
}
|
|
89
|
+
|
|
90
|
+
/** Return the distinct bounded provider error summaries shared by text and TUI rendering. */
|
|
91
|
+
export function getChildDiagnosticErrorRows(
|
|
92
|
+
diagnostics: ChildFailureDiagnostics,
|
|
93
|
+
): ChildDiagnosticErrorRow[] {
|
|
94
|
+
const rows: ChildDiagnosticErrorRow[] = [];
|
|
95
|
+
if (diagnostics.lastAssistantErrorText) {
|
|
96
|
+
rows.push({ label: "Last assistant error", text: diagnostics.lastAssistantErrorText });
|
|
97
|
+
}
|
|
98
|
+
if (
|
|
99
|
+
diagnostics.lastLifecycleErrorText &&
|
|
100
|
+
diagnostics.lastLifecycleErrorText !== diagnostics.lastAssistantErrorText
|
|
101
|
+
) {
|
|
102
|
+
rows.push({ label: "Lifecycle error", text: diagnostics.lastLifecycleErrorText });
|
|
103
|
+
}
|
|
104
|
+
return rows;
|
|
105
|
+
}
|
|
106
|
+
|
|
81
107
|
/** Format the complete retained safe diagnostic artifact for parent-facing text. */
|
|
82
108
|
export function formatChildFailureDiagnostics(diagnostics: ChildFailureDiagnostics): string[] {
|
|
83
109
|
const lines = [
|
|
@@ -95,6 +121,7 @@ export function formatChildFailureDiagnostics(diagnostics: ChildFailureDiagnosti
|
|
|
95
121
|
diagnostics.lastAssistantToolCalls && diagnostics.lastAssistantToolCalls.length > 0
|
|
96
122
|
? `- Last assistant tools: ${diagnostics.lastAssistantToolCalls.join(", ")}`
|
|
97
123
|
: undefined,
|
|
124
|
+
...getChildDiagnosticErrorRows(diagnostics).map((row) => `- ${row.label}: ${row.text}`),
|
|
98
125
|
`- ${formatChildLifecycleTrace(diagnostics.lifecycleTrace)}`,
|
|
99
126
|
];
|
|
100
127
|
return lines.filter((line): line is string => !!line);
|
|
@@ -103,6 +130,7 @@ export function formatChildFailureDiagnostics(diagnostics: ChildFailureDiagnosti
|
|
|
103
130
|
interface LastAssistantMetadata {
|
|
104
131
|
stopReason?: string;
|
|
105
132
|
toolCalls?: string[];
|
|
133
|
+
errorText?: string;
|
|
106
134
|
}
|
|
107
135
|
|
|
108
136
|
function extractLastAssistantMetadata(session: AgentSession): LastAssistantMetadata | undefined {
|
|
@@ -117,9 +145,11 @@ function extractLastAssistantMetadata(session: AgentSession): LastAssistantMetad
|
|
|
117
145
|
|
|
118
146
|
const stopReason = toSafeAssistantStopReason(message.stopReason);
|
|
119
147
|
const toolCalls = extractAssistantToolCalls(message.content, registeredToolNames);
|
|
148
|
+
const errorText = stopReason === "error" ? extractAssistantErrorText(message) : undefined;
|
|
120
149
|
return {
|
|
121
150
|
stopReason,
|
|
122
151
|
toolCalls: toolCalls.length > 0 ? toolCalls : undefined,
|
|
152
|
+
errorText,
|
|
123
153
|
};
|
|
124
154
|
}
|
|
125
155
|
return undefined;
|
|
@@ -143,3 +173,8 @@ function extractAssistantToolCalls(content: unknown, registeredToolNames: Set<st
|
|
|
143
173
|
})
|
|
144
174
|
.filter((name): name is string => !!name);
|
|
145
175
|
}
|
|
176
|
+
|
|
177
|
+
/** Extract only Pi's canonical provider error field from an errored assistant message. */
|
|
178
|
+
function extractAssistantErrorText(message: Record<string, unknown>): string | undefined {
|
|
179
|
+
return sanitizeChildErrorText(message.errorMessage);
|
|
180
|
+
}
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import type { AgentSession, AgentSessionEvent } from "@earendil-works/pi-coding-agent";
|
|
2
|
+
import { sanitizeChildErrorText } from "./diagnostic-sanitizer.ts";
|
|
2
3
|
|
|
3
4
|
/** Maximum number of observed Child Lifecycle Trace entries retained per child run. */
|
|
4
5
|
export const CHILD_LIFECYCLE_TRACE_MAX = 32;
|
|
@@ -53,9 +54,16 @@ export type ChildLifecycleTraceEntry =
|
|
|
53
54
|
willRetry: boolean;
|
|
54
55
|
hasResult: boolean;
|
|
55
56
|
hasError: boolean;
|
|
57
|
+
errorText?: string;
|
|
56
58
|
}
|
|
57
59
|
| { type: "auto_retry_start"; attempt: number; maxAttempts: number; delayMs: number }
|
|
58
|
-
| {
|
|
60
|
+
| {
|
|
61
|
+
type: "auto_retry_end";
|
|
62
|
+
success: boolean;
|
|
63
|
+
attempt: number;
|
|
64
|
+
hasFinalError: boolean;
|
|
65
|
+
finalErrorText?: string;
|
|
66
|
+
}
|
|
59
67
|
| {
|
|
60
68
|
type: "summarization_retry_scheduled";
|
|
61
69
|
attempt: number;
|
|
@@ -73,7 +81,6 @@ export type ChildLifecycleTraceEntry =
|
|
|
73
81
|
|
|
74
82
|
/** Minimal runner-control transitions that explain host-originated termination. */
|
|
75
83
|
export type ChildLifecycleHostMarker =
|
|
76
|
-
| { type: "steer_requested"; reason: "submit" | "timeout" }
|
|
77
84
|
| { type: "timeout_expired" }
|
|
78
85
|
| { type: "abort_requested"; reason: "canceled" | "timeout" }
|
|
79
86
|
| { type: "prompt_rejected" };
|
|
@@ -115,11 +122,6 @@ export class ChildLifecycleTraceCollector {
|
|
|
115
122
|
/** Record an allowlisted runner-control transition. */
|
|
116
123
|
recordHostMarker(marker: ChildLifecycleHostMarker): void {
|
|
117
124
|
switch (marker.type) {
|
|
118
|
-
case "steer_requested":
|
|
119
|
-
if (marker.reason === "submit" || marker.reason === "timeout") {
|
|
120
|
-
this.push({ type: "steer_requested", reason: marker.reason });
|
|
121
|
-
}
|
|
122
|
-
break;
|
|
123
125
|
case "abort_requested":
|
|
124
126
|
if (marker.reason === "canceled" || marker.reason === "timeout") {
|
|
125
127
|
this.push({ type: "abort_requested", reason: marker.reason });
|
|
@@ -161,6 +163,20 @@ export class ChildLifecycleTraceCollector {
|
|
|
161
163
|
}
|
|
162
164
|
}
|
|
163
165
|
|
|
166
|
+
/**
|
|
167
|
+
* Extract error text from the most recent lifecycle entry that carries an error.
|
|
168
|
+
* Checks compaction_end.errorText and auto_retry_end.finalErrorText.
|
|
169
|
+
*/
|
|
170
|
+
export function extractLastLifecycleErrorText(trace: ChildLifecycleTrace): string | undefined {
|
|
171
|
+
// Walk backwards to find the most recent error
|
|
172
|
+
for (let i = trace.entries.length - 1; i >= 0; i--) {
|
|
173
|
+
const entry = trace.entries[i];
|
|
174
|
+
if (entry.type === "compaction_end" && entry.errorText) return entry.errorText;
|
|
175
|
+
if (entry.type === "auto_retry_end" && entry.finalErrorText) return entry.finalErrorText;
|
|
176
|
+
}
|
|
177
|
+
return undefined;
|
|
178
|
+
}
|
|
179
|
+
|
|
164
180
|
/** Format the complete retained Child Lifecycle Trace for parent-facing diagnostics. */
|
|
165
181
|
export function formatChildLifecycleTrace(trace: ChildLifecycleTrace): string {
|
|
166
182
|
const tail =
|
|
@@ -185,11 +201,11 @@ function formatChildLifecycleTraceEntry(entry: ChildLifecycleTraceEntry): string
|
|
|
185
201
|
case "compaction_start":
|
|
186
202
|
return `compaction_start(reason=${entry.reason})`;
|
|
187
203
|
case "compaction_end":
|
|
188
|
-
return `compaction_end(reason=${entry.reason}, aborted=${entry.aborted}, willRetry=${entry.willRetry}, hasResult=${entry.hasResult}, hasError=${entry.hasError})`;
|
|
204
|
+
return `compaction_end(reason=${entry.reason}, aborted=${entry.aborted}, willRetry=${entry.willRetry}, hasResult=${entry.hasResult}, hasError=${entry.hasError}${entry.errorText ? `, error=${JSON.stringify(entry.errorText)}` : ""})`;
|
|
189
205
|
case "auto_retry_start":
|
|
190
206
|
return `auto_retry_start(attempt=${entry.attempt}/${entry.maxAttempts}, delayMs=${entry.delayMs})`;
|
|
191
207
|
case "auto_retry_end":
|
|
192
|
-
return `auto_retry_end(success=${entry.success}, attempt=${entry.attempt}, hasFinalError=${entry.hasFinalError})`;
|
|
208
|
+
return `auto_retry_end(success=${entry.success}, attempt=${entry.attempt}, hasFinalError=${entry.hasFinalError}${entry.finalErrorText ? `, error=${JSON.stringify(entry.finalErrorText)}` : ""})`;
|
|
193
209
|
case "summarization_retry_scheduled":
|
|
194
210
|
return `summarization_retry_scheduled(attempt=${entry.attempt}/${entry.maxAttempts}, delayMs=${entry.delayMs})`;
|
|
195
211
|
case "summarization_retry_attempt_start":
|
|
@@ -198,8 +214,6 @@ function formatChildLifecycleTraceEntry(entry: ChildLifecycleTraceEntry): string
|
|
|
198
214
|
: `summarization_retry_attempt_start(source=${entry.source})`;
|
|
199
215
|
case "queue_update":
|
|
200
216
|
return `queue_update(steering=${entry.steeringCount}, followUp=${entry.followUpCount})`;
|
|
201
|
-
case "steer_requested":
|
|
202
|
-
return `steer_requested(reason=${entry.reason})`;
|
|
203
217
|
case "abort_requested":
|
|
204
218
|
return `abort_requested(reason=${entry.reason})`;
|
|
205
219
|
}
|
|
@@ -269,6 +283,7 @@ function mapLifecycleEvent(event: AgentSessionEvent): ChildLifecycleTraceEntry |
|
|
|
269
283
|
willRetry: event.willRetry,
|
|
270
284
|
hasResult: event.result !== undefined,
|
|
271
285
|
hasError: event.errorMessage !== undefined,
|
|
286
|
+
errorText: sanitizeChildErrorText(event.errorMessage),
|
|
272
287
|
}
|
|
273
288
|
: undefined;
|
|
274
289
|
case "auto_retry_start":
|
|
@@ -289,6 +304,7 @@ function mapLifecycleEvent(event: AgentSessionEvent): ChildLifecycleTraceEntry |
|
|
|
289
304
|
success: event.success,
|
|
290
305
|
attempt: event.attempt,
|
|
291
306
|
hasFinalError: event.finalError !== undefined,
|
|
307
|
+
finalErrorText: sanitizeChildErrorText(event.finalError),
|
|
292
308
|
}
|
|
293
309
|
: undefined;
|
|
294
310
|
case "summarization_retry_scheduled":
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type { clampThinkingLevel, Model } from "@earendil-works/pi-ai";
|
|
1
|
+
import type { clampThinkingLevel, Model, Usage } from "@earendil-works/pi-ai";
|
|
2
2
|
import {
|
|
3
3
|
createAgentSession,
|
|
4
4
|
SessionManager,
|
|
@@ -6,6 +6,7 @@ import {
|
|
|
6
6
|
} from "@earendil-works/pi-coding-agent";
|
|
7
7
|
import type { ChildFailureDiagnostics, ReviewProgress } from "../types.ts";
|
|
8
8
|
import { createIsolatedChildResources } from "./child-resource-loader.ts";
|
|
9
|
+
import { buildProgressTokens } from "./runner-helpers.ts";
|
|
9
10
|
import { runWithLifecycle } from "./session-lifecycle.ts";
|
|
10
11
|
|
|
11
12
|
/** Configuration for one isolated child run — resource loading, session, and lifecycle. */
|
|
@@ -15,19 +16,24 @@ export interface IsolatedRunConfig<THolder, TResult> {
|
|
|
15
16
|
// biome-ignore lint/suspicious/noExplicitAny: Model<any> is Pi's canonical type
|
|
16
17
|
model: Model<any>;
|
|
17
18
|
thinkingLevel: ReturnType<typeof clampThinkingLevel>;
|
|
18
|
-
timeoutMs
|
|
19
|
+
timeoutMs?: number;
|
|
19
20
|
prompt: string;
|
|
20
21
|
signal?: AbortSignal;
|
|
21
22
|
tools: string[];
|
|
22
23
|
customTools: ToolDefinition[];
|
|
23
24
|
holder: { value?: THolder };
|
|
24
|
-
successResult: (value: THolder) => TResult;
|
|
25
|
-
canceledResult: (diagnostics: ChildFailureDiagnostics) => TResult;
|
|
25
|
+
successResult: (value: THolder, usage?: Usage) => TResult;
|
|
26
|
+
canceledResult: (diagnostics: ChildFailureDiagnostics, usage?: Usage) => TResult;
|
|
26
27
|
failedResult: (
|
|
27
28
|
failureCode: "prompt-rejected" | "missing-structured-output" | "unexpected-runner-failure",
|
|
28
29
|
diagnostics: ChildFailureDiagnostics,
|
|
30
|
+
usage?: Usage,
|
|
31
|
+
) => TResult;
|
|
32
|
+
timeoutResult: (
|
|
33
|
+
timeoutMs: number,
|
|
34
|
+
diagnostics: ChildFailureDiagnostics,
|
|
35
|
+
usage?: Usage,
|
|
29
36
|
) => TResult;
|
|
30
|
-
timeoutResult: (timeoutMs: number, diagnostics: ChildFailureDiagnostics) => TResult;
|
|
31
37
|
sessionFailedResult: TResult;
|
|
32
38
|
onProgress?: (progress: ReviewProgress) => void;
|
|
33
39
|
}
|
|
@@ -62,20 +68,31 @@ export async function runIsolatedChild<THolder, TResult>(
|
|
|
62
68
|
signal: config.signal,
|
|
63
69
|
timeoutMs: config.timeoutMs,
|
|
64
70
|
onEvent: (event, ctx) => {
|
|
71
|
+
const reportsProgress =
|
|
72
|
+
event.type === "turn_end" ||
|
|
73
|
+
event.type === "tool_execution_start" ||
|
|
74
|
+
event.type === "agent_settled";
|
|
65
75
|
if (event.type === "turn_end") ctx.progress.turns++;
|
|
66
76
|
if (event.type === "tool_execution_start") ctx.progress.toolUses++;
|
|
67
|
-
|
|
77
|
+
if (reportsProgress) {
|
|
78
|
+
ctx.progress.tokens = buildProgressTokens(() => ctx.session.getSessionStats());
|
|
79
|
+
config.onProgress?.({ ...ctx.progress });
|
|
80
|
+
}
|
|
68
81
|
if (event.type !== "agent_settled") return;
|
|
69
82
|
const result = config.holder.value
|
|
70
|
-
? config.successResult(config.holder.value)
|
|
71
|
-
: config.failedResult(
|
|
83
|
+
? config.successResult(config.holder.value, ctx.getUsage())
|
|
84
|
+
: config.failedResult(
|
|
85
|
+
"missing-structured-output",
|
|
86
|
+
ctx.getFailureDiagnostics(),
|
|
87
|
+
ctx.getUsage(),
|
|
88
|
+
);
|
|
72
89
|
ctx.resolve(ctx.cleanup(result));
|
|
73
90
|
},
|
|
74
|
-
canceledResult: (ctx) => config.canceledResult(ctx.getFailureDiagnostics()),
|
|
91
|
+
canceledResult: (ctx) => config.canceledResult(ctx.getFailureDiagnostics(), ctx.getUsage()),
|
|
75
92
|
failedResult: (failureCode, ctx) =>
|
|
76
|
-
config.failedResult(failureCode, ctx.getFailureDiagnostics()),
|
|
93
|
+
config.failedResult(failureCode, ctx.getFailureDiagnostics(), ctx.getUsage()),
|
|
77
94
|
timeoutResult: (timeoutMs, ctx) =>
|
|
78
|
-
config.timeoutResult(timeoutMs, ctx.getFailureDiagnostics()),
|
|
95
|
+
config.timeoutResult(timeoutMs, ctx.getFailureDiagnostics(), ctx.getUsage()),
|
|
79
96
|
});
|
|
80
97
|
} catch {
|
|
81
98
|
return config.sessionFailedResult;
|
|
@@ -0,0 +1,73 @@
|
|
|
1
|
+
import type { Usage } from "@earendil-works/pi-ai";
|
|
2
|
+
import type { AgentSession } from "@earendil-works/pi-coding-agent";
|
|
3
|
+
|
|
4
|
+
function addOptional(first: number | undefined, second: number | undefined): number | undefined {
|
|
5
|
+
return first === undefined && second === undefined ? undefined : (first ?? 0) + (second ?? 0);
|
|
6
|
+
}
|
|
7
|
+
|
|
8
|
+
/** Add two provider Usage records without dropping optional reasoning/cache fields. */
|
|
9
|
+
export function addUsage(first: Usage, second: Usage): Usage {
|
|
10
|
+
const cacheWrite1h = addOptional(first.cacheWrite1h, second.cacheWrite1h);
|
|
11
|
+
const reasoning = addOptional(first.reasoning, second.reasoning);
|
|
12
|
+
return {
|
|
13
|
+
input: first.input + second.input,
|
|
14
|
+
output: first.output + second.output,
|
|
15
|
+
cacheRead: first.cacheRead + second.cacheRead,
|
|
16
|
+
cacheWrite: first.cacheWrite + second.cacheWrite,
|
|
17
|
+
...(cacheWrite1h === undefined ? {} : { cacheWrite1h }),
|
|
18
|
+
...(reasoning === undefined ? {} : { reasoning }),
|
|
19
|
+
totalTokens: first.totalTokens + second.totalTokens,
|
|
20
|
+
cost: {
|
|
21
|
+
input: first.cost.input + second.cost.input,
|
|
22
|
+
output: first.cost.output + second.cost.output,
|
|
23
|
+
cacheRead: first.cost.cacheRead + second.cost.cacheRead,
|
|
24
|
+
cacheWrite: first.cost.cacheWrite + second.cost.cacheWrite,
|
|
25
|
+
total: first.cost.total + second.cost.total,
|
|
26
|
+
},
|
|
27
|
+
};
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
/** Sum available Usage records, returning undefined when no model call reported usage. */
|
|
31
|
+
export function combineUsage(usages: Array<Usage | undefined>): Usage | undefined {
|
|
32
|
+
return usages.reduce<Usage | undefined>(
|
|
33
|
+
(total, usage) => (usage ? (total ? addUsage(total, usage) : usage) : total),
|
|
34
|
+
undefined,
|
|
35
|
+
);
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
/** Snapshot aggregate assistant usage from one isolated child transcript. */
|
|
39
|
+
export function collectChildUsage(session: AgentSession): Usage | undefined {
|
|
40
|
+
try {
|
|
41
|
+
const messages = session.messages as unknown;
|
|
42
|
+
if (!Array.isArray(messages)) return undefined;
|
|
43
|
+
return combineUsage(
|
|
44
|
+
messages.map((message) => {
|
|
45
|
+
if (!message || typeof message !== "object") return undefined;
|
|
46
|
+
const candidate = message as { role?: unknown; usage?: unknown };
|
|
47
|
+
return candidate.role === "assistant" && isUsage(candidate.usage)
|
|
48
|
+
? candidate.usage
|
|
49
|
+
: undefined;
|
|
50
|
+
}),
|
|
51
|
+
);
|
|
52
|
+
} catch {
|
|
53
|
+
return undefined;
|
|
54
|
+
}
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
function isUsage(value: unknown): value is Usage {
|
|
58
|
+
if (!value || typeof value !== "object") return false;
|
|
59
|
+
const usage = value as Partial<Usage>;
|
|
60
|
+
return (
|
|
61
|
+
typeof usage.input === "number" &&
|
|
62
|
+
typeof usage.output === "number" &&
|
|
63
|
+
typeof usage.cacheRead === "number" &&
|
|
64
|
+
typeof usage.cacheWrite === "number" &&
|
|
65
|
+
typeof usage.totalTokens === "number" &&
|
|
66
|
+
!!usage.cost &&
|
|
67
|
+
typeof usage.cost.input === "number" &&
|
|
68
|
+
typeof usage.cost.output === "number" &&
|
|
69
|
+
typeof usage.cost.cacheRead === "number" &&
|
|
70
|
+
typeof usage.cost.cacheWrite === "number" &&
|
|
71
|
+
typeof usage.cost.total === "number"
|
|
72
|
+
);
|
|
73
|
+
}
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
import { redactDebugData } from "@mrclrchtr/supi-core/debug";
|
|
2
|
+
|
|
3
|
+
/** Maximum retained characters for one provider-owned diagnostic summary. */
|
|
4
|
+
export const MAX_CHILD_ERROR_CHARACTERS = 500;
|
|
5
|
+
|
|
6
|
+
const ANSI_ESCAPE_RE = new RegExp(`${String.fromCharCode(27)}\\[[0-?]*[ -/]*[@-~]`, "g");
|
|
7
|
+
const JSON_SECRET_RE =
|
|
8
|
+
/(["']?(?:token|password|passwd|secret|api[_-]?key|authorization|credential)["']?\s*:\s*)(["'])[^"']*\2/gi;
|
|
9
|
+
const UNQUOTED_SECRET_RE =
|
|
10
|
+
/(["']?(?:token|password|passwd|secret|api[_-]?key|authorization|credential)["']?\s*:\s*)[^\s,}\]]+/gi;
|
|
11
|
+
|
|
12
|
+
function stripControlCharacters(value: string): string {
|
|
13
|
+
return Array.from(value, (character) => {
|
|
14
|
+
const code = character.codePointAt(0) ?? 0;
|
|
15
|
+
return code < 32 || (code >= 127 && code <= 159) ? " " : character;
|
|
16
|
+
}).join("");
|
|
17
|
+
}
|
|
18
|
+
|
|
19
|
+
/** Redact, normalize, and bound provider-owned error text before session persistence. */
|
|
20
|
+
export function sanitizeChildErrorText(value: unknown): string | undefined {
|
|
21
|
+
if (typeof value !== "string" || !value.trim()) return undefined;
|
|
22
|
+
const redacted = redactDebugData(value)
|
|
23
|
+
.replace(JSON_SECRET_RE, (_match, prefix: string, quote: string) => {
|
|
24
|
+
return `${prefix}${quote}[REDACTED]${quote}`;
|
|
25
|
+
})
|
|
26
|
+
.replace(UNQUOTED_SECRET_RE, (_match, prefix: string) => `${prefix}[REDACTED]`);
|
|
27
|
+
const normalized = stripControlCharacters(redacted.replace(ANSI_ESCAPE_RE, " "))
|
|
28
|
+
.replace(/\s+/g, " ")
|
|
29
|
+
.trim();
|
|
30
|
+
if (!normalized) return undefined;
|
|
31
|
+
if (normalized.length <= MAX_CHILD_ERROR_CHARACTERS) return normalized;
|
|
32
|
+
let end = MAX_CHILD_ERROR_CHARACTERS - 1;
|
|
33
|
+
if (/[\uD800-\uDBFF]/.test(normalized[end - 1] ?? "")) end--;
|
|
34
|
+
return `${normalized.slice(0, end)}…`;
|
|
35
|
+
}
|
package/src/tool/output-page.ts
CHANGED
|
@@ -13,6 +13,32 @@ export interface TextPage {
|
|
|
13
13
|
totalCharacters: number;
|
|
14
14
|
}
|
|
15
15
|
|
|
16
|
+
/** One selected 1-based line range before character paging is applied. */
|
|
17
|
+
export interface LineSelection {
|
|
18
|
+
text: string;
|
|
19
|
+
startLine: number;
|
|
20
|
+
endLine: number;
|
|
21
|
+
totalLines: number;
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
/** Select a bounded 1-based line range from file text. */
|
|
25
|
+
export function selectLineRange(text: string, startLine: number, lineCount = 200): LineSelection {
|
|
26
|
+
const lines = text.split("\n");
|
|
27
|
+
if (!Number.isSafeInteger(startLine) || startLine < 1 || startLine > lines.length) {
|
|
28
|
+
throw new Error(`Start line must be an integer between 1 and ${lines.length}.`);
|
|
29
|
+
}
|
|
30
|
+
if (!Number.isSafeInteger(lineCount) || lineCount < 1 || lineCount > MAX_PAGE_LINES) {
|
|
31
|
+
throw new Error(`Line count must be an integer between 1 and ${MAX_PAGE_LINES}.`);
|
|
32
|
+
}
|
|
33
|
+
const endLine = Math.min(lines.length, startLine + lineCount - 1);
|
|
34
|
+
return {
|
|
35
|
+
text: lines.slice(startLine - 1, endLine).join("\n"),
|
|
36
|
+
startLine,
|
|
37
|
+
endLine,
|
|
38
|
+
totalLines: lines.length,
|
|
39
|
+
};
|
|
40
|
+
}
|
|
41
|
+
|
|
16
42
|
function lineBoundedEnd(text: string, start: number, proposedEnd: number): number {
|
|
17
43
|
let lines = 1;
|
|
18
44
|
for (let index = start; index < proposedEnd; index++) {
|