mandrel 2.84.0 → 2.86.0
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/.agents/docs/agentrc-reference.json +2 -1
- package/.agents/docs/configuration.md +4 -2
- package/.agents/schemas/agentrc.schema.json +41 -2
- package/.agents/schemas/story-deliver-terminal.schema.json +2 -2
- package/.agents/scripts/coverage-capture.js +1 -0
- package/.agents/scripts/evidence-gate.js +3 -2
- package/.agents/scripts/lib/close-lock-events.js +29 -9
- package/.agents/scripts/lib/close-validation/process.js +6 -0
- package/.agents/scripts/lib/close-validation/runner.js +22 -6
- package/.agents/scripts/lib/config-settings-schema-delivery.js +41 -2
- package/.agents/scripts/lib/full-suite-lock.js +100 -38
- package/.agents/scripts/lib/full-suite-queue.js +104 -37
- package/.agents/scripts/lib/generated/agentrc-validator.js +1 -1
- package/.agents/scripts/lib/orchestration/review-providers/claude-verdict.js +242 -0
- package/.agents/scripts/lib/orchestration/review-providers/code-review.js +151 -111
- package/.agents/scripts/lib/orchestration/single-story-close/runner.js +30 -4
- package/.agents/scripts/lib/orchestration/story-handoff.js +103 -11
- package/.agents/scripts/lib/suite-admission.js +410 -0
- package/.agents/workflows/helpers/deliver-story-reference.md +6 -0
- package/docs/CHANGELOG.md +15 -0
- package/package.json +1 -1
|
@@ -0,0 +1,242 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* review-providers/claude-verdict.js — reads `code-review`'s claude result
|
|
3
|
+
* envelope and auth-status payload into findings or a no-verdict reason.
|
|
4
|
+
*
|
|
5
|
+
* @typedef {import('./types.js').Finding} Finding
|
|
6
|
+
*/
|
|
7
|
+
|
|
8
|
+
import { buildFinding, unwrapEnvelope } from './parse-findings.js';
|
|
9
|
+
|
|
10
|
+
export const FINDINGS_SCHEMA_JSON = JSON.stringify({
|
|
11
|
+
type: 'object',
|
|
12
|
+
properties: {
|
|
13
|
+
findings: {
|
|
14
|
+
type: 'array',
|
|
15
|
+
items: {
|
|
16
|
+
type: 'object',
|
|
17
|
+
properties: {
|
|
18
|
+
title: { type: 'string' },
|
|
19
|
+
body: { type: 'string' },
|
|
20
|
+
file: { type: 'string' },
|
|
21
|
+
line: { type: 'integer' },
|
|
22
|
+
category: { type: 'string' },
|
|
23
|
+
},
|
|
24
|
+
required: ['title', 'body'],
|
|
25
|
+
},
|
|
26
|
+
},
|
|
27
|
+
},
|
|
28
|
+
required: ['findings'],
|
|
29
|
+
});
|
|
30
|
+
|
|
31
|
+
const MAX_DETAIL_CHARS = 300;
|
|
32
|
+
const USAGE_LIMIT_RE = /hit your .*limit|usage limit|rate limit/i;
|
|
33
|
+
|
|
34
|
+
export function clip(text) {
|
|
35
|
+
const flat = String(text ?? '')
|
|
36
|
+
.replace(/\s+/g, ' ')
|
|
37
|
+
.trim();
|
|
38
|
+
return flat.length > MAX_DETAIL_CHARS
|
|
39
|
+
? `${flat.slice(0, MAX_DETAIL_CHARS)}…`
|
|
40
|
+
: flat;
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
function tryParseJson(text) {
|
|
44
|
+
try {
|
|
45
|
+
return JSON.parse(text);
|
|
46
|
+
} catch {
|
|
47
|
+
return undefined;
|
|
48
|
+
}
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
/**
|
|
52
|
+
* `claude auth status --json`. Only `authMethod` and `subscriptionType`
|
|
53
|
+
* are ever named: the payload also carries the account email, org name
|
|
54
|
+
* and ids, which no reason or log line may repeat.
|
|
55
|
+
*
|
|
56
|
+
* @param {{ status: number, stdout: string }} result
|
|
57
|
+
* @returns {{ ok: true, detail: string } | { ok: false, reason: string }}
|
|
58
|
+
*/
|
|
59
|
+
export function readAuthStatus(result) {
|
|
60
|
+
if (result.status !== 0) {
|
|
61
|
+
return {
|
|
62
|
+
ok: false,
|
|
63
|
+
reason: `claude auth status exited ${result.status}: the CLI is not logged in or unusable`,
|
|
64
|
+
};
|
|
65
|
+
}
|
|
66
|
+
const parsed = tryParseJson(result.stdout);
|
|
67
|
+
if (!parsed || typeof parsed !== 'object') {
|
|
68
|
+
return {
|
|
69
|
+
ok: false,
|
|
70
|
+
reason: 'claude auth status returned unparseable output',
|
|
71
|
+
};
|
|
72
|
+
}
|
|
73
|
+
const method =
|
|
74
|
+
typeof parsed.authMethod === 'string' ? parsed.authMethod : null;
|
|
75
|
+
if (parsed.loggedIn !== true) {
|
|
76
|
+
return {
|
|
77
|
+
ok: false,
|
|
78
|
+
reason: `claude CLI is not logged in${method ? ` (authMethod: ${method})` : ''}`,
|
|
79
|
+
};
|
|
80
|
+
}
|
|
81
|
+
const plan =
|
|
82
|
+
typeof parsed.subscriptionType === 'string'
|
|
83
|
+
? `, subscriptionType: ${parsed.subscriptionType}`
|
|
84
|
+
: '';
|
|
85
|
+
return { ok: true, detail: `authMethod: ${method ?? 'unknown'}${plan}` };
|
|
86
|
+
}
|
|
87
|
+
|
|
88
|
+
/**
|
|
89
|
+
* The end index of the balanced JSON array opening at `start`, string-aware.
|
|
90
|
+
*
|
|
91
|
+
* @param {string} text
|
|
92
|
+
* @param {number} start - index of a `[`.
|
|
93
|
+
* @returns {number} index of the closing `]`, or -1.
|
|
94
|
+
*/
|
|
95
|
+
function matchArrayEnd(text, start) {
|
|
96
|
+
let depth = 0;
|
|
97
|
+
let inString = false;
|
|
98
|
+
for (let i = start; i < text.length; i += 1) {
|
|
99
|
+
const ch = text[i];
|
|
100
|
+
if (inString) {
|
|
101
|
+
if (ch === '\\') i += 1;
|
|
102
|
+
else if (ch === '"') inString = false;
|
|
103
|
+
} else if (ch === '"') inString = true;
|
|
104
|
+
else if (ch === '[') depth += 1;
|
|
105
|
+
else if (ch === ']') {
|
|
106
|
+
depth -= 1;
|
|
107
|
+
if (depth === 0) return i;
|
|
108
|
+
}
|
|
109
|
+
}
|
|
110
|
+
return -1;
|
|
111
|
+
}
|
|
112
|
+
|
|
113
|
+
/**
|
|
114
|
+
* The last top-level JSON array in free text.
|
|
115
|
+
*
|
|
116
|
+
* @param {string} text
|
|
117
|
+
* @returns {unknown[] | undefined}
|
|
118
|
+
*/
|
|
119
|
+
function lastBareArray(text) {
|
|
120
|
+
let found;
|
|
121
|
+
let i = text.indexOf('[');
|
|
122
|
+
while (i !== -1) {
|
|
123
|
+
const end = matchArrayEnd(text, i);
|
|
124
|
+
const parsed =
|
|
125
|
+
end === -1 ? undefined : tryParseJson(text.slice(i, end + 1));
|
|
126
|
+
if (Array.isArray(parsed)) {
|
|
127
|
+
found = parsed;
|
|
128
|
+
i = text.indexOf('[', end + 1);
|
|
129
|
+
} else {
|
|
130
|
+
i = text.indexOf('[', i + 1);
|
|
131
|
+
}
|
|
132
|
+
}
|
|
133
|
+
return found;
|
|
134
|
+
}
|
|
135
|
+
|
|
136
|
+
/**
|
|
137
|
+
* The findings list from a JSON value: a bare array or `{ findings }`.
|
|
138
|
+
*
|
|
139
|
+
* @param {unknown} value
|
|
140
|
+
* @returns {unknown[] | undefined}
|
|
141
|
+
*/
|
|
142
|
+
function asFindingList(value) {
|
|
143
|
+
if (value === undefined) return undefined;
|
|
144
|
+
const unwrapped = unwrapEnvelope(value);
|
|
145
|
+
return Array.isArray(unwrapped) ? unwrapped : undefined;
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
/**
|
|
149
|
+
* Fallback over the envelope's `result` text: the last fenced block, else
|
|
150
|
+
* the last top-level JSON array — so a prose preamble is still a verdict.
|
|
151
|
+
*
|
|
152
|
+
* @param {string} text
|
|
153
|
+
* @returns {unknown[] | undefined}
|
|
154
|
+
*/
|
|
155
|
+
function parseVerdictText(text) {
|
|
156
|
+
const fences = [...text.matchAll(/```[a-z]*[^\S\n]*\n([\s\S]*?)```/gi)];
|
|
157
|
+
const lastFence = fences.at(-1);
|
|
158
|
+
const fenced = lastFence
|
|
159
|
+
? asFindingList(tryParseJson(lastFence[1].trim()))
|
|
160
|
+
: undefined;
|
|
161
|
+
return fenced ?? lastBareArray(text);
|
|
162
|
+
}
|
|
163
|
+
|
|
164
|
+
function toFindings(entries) {
|
|
165
|
+
return entries
|
|
166
|
+
.map((entry) =>
|
|
167
|
+
buildFinding(entry, {
|
|
168
|
+
mapSeverity: () => 'critical',
|
|
169
|
+
defaultCategory: 'bug',
|
|
170
|
+
}),
|
|
171
|
+
)
|
|
172
|
+
.filter(Boolean);
|
|
173
|
+
}
|
|
174
|
+
|
|
175
|
+
function failureReason(result, envelope) {
|
|
176
|
+
const resultText =
|
|
177
|
+
typeof envelope?.result === 'string' ? envelope.result : '';
|
|
178
|
+
const text = [result.stderr, resultText, result.stdout].join('\n');
|
|
179
|
+
if (USAGE_LIMIT_RE.test(text)) {
|
|
180
|
+
return 'the claude CLI account is out of quota (usage limit reached)';
|
|
181
|
+
}
|
|
182
|
+
if (result.status !== 0) {
|
|
183
|
+
return `claude --print exited with status ${result.status}: ${
|
|
184
|
+
clip(result.stderr || resultText || result.stdout) || '<no output>'
|
|
185
|
+
}`;
|
|
186
|
+
}
|
|
187
|
+
if (envelope?.is_error === true) {
|
|
188
|
+
return `the reviewer returned an error result: ${clip(resultText) || '<no output>'}`;
|
|
189
|
+
}
|
|
190
|
+
return `the reviewer run ended with subtype ${clip(envelope?.subtype)}`;
|
|
191
|
+
}
|
|
192
|
+
|
|
193
|
+
function parseEnvelope(stdout) {
|
|
194
|
+
const parsed = tryParseJson(stdout);
|
|
195
|
+
return parsed && typeof parsed === 'object' && !Array.isArray(parsed)
|
|
196
|
+
? parsed
|
|
197
|
+
: null;
|
|
198
|
+
}
|
|
199
|
+
|
|
200
|
+
function isSuccess(result, envelope) {
|
|
201
|
+
return (
|
|
202
|
+
result.status === 0 &&
|
|
203
|
+
envelope !== null &&
|
|
204
|
+
envelope.is_error !== true &&
|
|
205
|
+
envelope.subtype === 'success'
|
|
206
|
+
);
|
|
207
|
+
}
|
|
208
|
+
|
|
209
|
+
function noVerdictReason(result, envelope) {
|
|
210
|
+
if (result.status === 0 && envelope === null) {
|
|
211
|
+
return 'the reviewer output was not a JSON result envelope';
|
|
212
|
+
}
|
|
213
|
+
return failureReason(result, envelope);
|
|
214
|
+
}
|
|
215
|
+
|
|
216
|
+
function structuredFindings(structured) {
|
|
217
|
+
return Array.isArray(structured?.findings) ? structured.findings : undefined;
|
|
218
|
+
}
|
|
219
|
+
|
|
220
|
+
function resultTextFindings(text) {
|
|
221
|
+
return typeof text === 'string' ? parseVerdictText(text) : undefined;
|
|
222
|
+
}
|
|
223
|
+
|
|
224
|
+
/**
|
|
225
|
+
* Verdict resolution: (a) a successful envelope's `structured_output`,
|
|
226
|
+
* (b) the fallback parser over its `result` text, (c) no verdict.
|
|
227
|
+
*
|
|
228
|
+
* @param {{ status: number, stdout: string, stderr: string }} result
|
|
229
|
+
* @returns {{ findings: Finding[] } | { reason: string }}
|
|
230
|
+
*/
|
|
231
|
+
export function resolveVerdict(result) {
|
|
232
|
+
const envelope = parseEnvelope(result.stdout);
|
|
233
|
+
if (!isSuccess(result, envelope)) {
|
|
234
|
+
return { reason: noVerdictReason(result, envelope) };
|
|
235
|
+
}
|
|
236
|
+
const entries =
|
|
237
|
+
structuredFindings(envelope.structured_output) ??
|
|
238
|
+
resultTextFindings(envelope.result);
|
|
239
|
+
return entries
|
|
240
|
+
? { findings: toFindings(entries) }
|
|
241
|
+
: { reason: 'the reviewer output did not parse as a findings verdict' };
|
|
242
|
+
}
|
|
@@ -6,26 +6,58 @@
|
|
|
6
6
|
* id and the diff text — never `acceptance[]` or the self-eval verdict: this
|
|
7
7
|
* is a bug review, not a second acceptance scoring.
|
|
8
8
|
*
|
|
9
|
-
*
|
|
10
|
-
*
|
|
11
|
-
*
|
|
9
|
+
* Cost is bounded by stripping context, not by pinning a model: an empty
|
|
10
|
+
* OS temp cwd (no CLAUDE.md discovery), user-only settings, no MCP, no tools.
|
|
11
|
+
*
|
|
12
|
+
* A missing CLI throws at construction (the chain entry is optional). A
|
|
13
|
+
* review with no verdict records one degraded gate via `getDegradations()`
|
|
14
|
+
* and returns no findings, so "never ran" cannot read as "found nothing".
|
|
12
15
|
*
|
|
13
16
|
* @typedef {import('./types.js').Finding} Finding
|
|
14
17
|
* @typedef {import('./types.js').ReviewInput} ReviewInput
|
|
15
18
|
* @typedef {import('./types.js').ReviewProvider} ReviewProvider
|
|
19
|
+
* @typedef {import('./degraded-gates.js').GateDegradation} GateDegradation
|
|
16
20
|
*/
|
|
17
21
|
|
|
22
|
+
import fs from 'node:fs';
|
|
23
|
+
import os from 'node:os';
|
|
24
|
+
import path from 'node:path';
|
|
18
25
|
import { spawnCaptureAsync } from '../../child-exec.js';
|
|
19
26
|
import { gitSpawn } from '../../git-utils.js';
|
|
20
27
|
import { PROJECT_ROOT } from '../../project-root.js';
|
|
21
|
-
import {
|
|
28
|
+
import {
|
|
29
|
+
clip,
|
|
30
|
+
FINDINGS_SCHEMA_JSON,
|
|
31
|
+
readAuthStatus,
|
|
32
|
+
resolveVerdict,
|
|
33
|
+
} from './claude-verdict.js';
|
|
22
34
|
import { renderDepthDirective } from './review-depth.js';
|
|
23
35
|
import { probeClaudeCli } from './security-review.js';
|
|
24
36
|
|
|
25
|
-
const CLAUDE_ARGS = Object.freeze([
|
|
37
|
+
const CLAUDE_ARGS = Object.freeze([
|
|
38
|
+
'--print',
|
|
39
|
+
'--effort',
|
|
40
|
+
'low',
|
|
41
|
+
'--output-format',
|
|
42
|
+
'json',
|
|
43
|
+
'--json-schema',
|
|
44
|
+
FINDINGS_SCHEMA_JSON,
|
|
45
|
+
'--setting-sources',
|
|
46
|
+
'user',
|
|
47
|
+
'--strict-mcp-config',
|
|
48
|
+
'--tools',
|
|
49
|
+
'',
|
|
50
|
+
]);
|
|
51
|
+
const AUTH_ARGS = Object.freeze(['auth', 'status', '--json']);
|
|
26
52
|
const INVOKE_TIMEOUT_MS = 10 * 60 * 1000;
|
|
53
|
+
const AUTH_TIMEOUT_MS = 15 * 1000;
|
|
27
54
|
/** Larger diffs are truncated; the reviewer is told so. */
|
|
28
55
|
const MAX_DIFF_CHARS = 200_000;
|
|
56
|
+
const DEGRADATION_BASE = Object.freeze({
|
|
57
|
+
tool: 'code-review',
|
|
58
|
+
gate: 'codeReview',
|
|
59
|
+
surface: 'model-bug-review',
|
|
60
|
+
});
|
|
29
61
|
|
|
30
62
|
const PROMPT_HEAD =
|
|
31
63
|
'You are reviewing the diff `{baseRef}...{headRef}` for Story #{ticketId} ' +
|
|
@@ -37,11 +69,10 @@ const PROMPT_HEAD =
|
|
|
37
69
|
'line in the new version, why it is wrong, and how to show it fails (an ' +
|
|
38
70
|
'input, command or test that exposes it).\n\n' +
|
|
39
71
|
'The diff below is authoritative — files on disk may not reflect it. ' +
|
|
40
|
-
'
|
|
41
|
-
'it
|
|
42
|
-
'
|
|
43
|
-
'"
|
|
44
|
-
'Emit [] if there is nothing you would block the merge for.\n\n';
|
|
72
|
+
'Answer with a JSON object of this exact shape, no prose around it:\n\n' +
|
|
73
|
+
'{"findings":[{"title":"...","body":"Why it is wrong: ... How to show it ' +
|
|
74
|
+
'fails: ...","file":"...","line":1,"category":"bug"}]}\n\n' +
|
|
75
|
+
'Emit {"findings":[]} if there is nothing you would block the merge for.\n\n';
|
|
45
76
|
|
|
46
77
|
/**
|
|
47
78
|
* @param {ReviewInput} input
|
|
@@ -62,64 +93,37 @@ function buildPrompt(input, diff) {
|
|
|
62
93
|
}
|
|
63
94
|
|
|
64
95
|
/**
|
|
65
|
-
*
|
|
66
|
-
*
|
|
96
|
+
* A Windows shell joins argv unquoted, which would drop `--tools ""` and
|
|
97
|
+
* split the schema; quote each arg for the MSVCRT argv parser.
|
|
67
98
|
*
|
|
68
|
-
* @param {string}
|
|
69
|
-
* @
|
|
70
|
-
* @returns {Promise<{ status: number, stdout: string, stderr: string }>}
|
|
71
|
-
*/
|
|
72
|
-
function invokeClaude(prompt, run) {
|
|
73
|
-
return spawnCaptureAsync('claude', [...CLAUDE_ARGS], {
|
|
74
|
-
cwd: PROJECT_ROOT,
|
|
75
|
-
input: prompt,
|
|
76
|
-
shell: process.platform === 'win32',
|
|
77
|
-
timeout: INVOKE_TIMEOUT_MS,
|
|
78
|
-
...(run ? { run } : {}),
|
|
79
|
-
});
|
|
80
|
-
}
|
|
81
|
-
|
|
82
|
-
/** A model often fences its JSON; strip one surrounding fence. */
|
|
83
|
-
function stripFence(text) {
|
|
84
|
-
const match = /^\s*```[a-z]*\s*\n([\s\S]*?)\n\s*```\s*$/i.exec(text ?? '');
|
|
85
|
-
return match ? match[1] : (text ?? '');
|
|
86
|
-
}
|
|
87
|
-
|
|
88
|
-
/**
|
|
89
|
-
* @param {string} title
|
|
90
|
-
* @param {string} detail
|
|
91
|
-
* @returns {Finding}
|
|
99
|
+
* @param {string} arg
|
|
100
|
+
* @returns {string}
|
|
92
101
|
*/
|
|
93
|
-
function
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
title,
|
|
97
|
-
body:
|
|
98
|
-
`${detail} The model bug review did not produce a verdict — inspect ` +
|
|
99
|
-
'the diff manually before merging. Advisory only; the chain did not halt.',
|
|
100
|
-
category: 'bug',
|
|
101
|
-
};
|
|
102
|
+
function quoteWindowsArg(arg) {
|
|
103
|
+
const escaped = arg.replace(/(\\*)"/g, '$1$1\\"').replace(/(\\+)$/, '$1$1');
|
|
104
|
+
return `"${escaped}"`;
|
|
102
105
|
}
|
|
103
106
|
|
|
104
107
|
/**
|
|
105
|
-
*
|
|
106
|
-
*
|
|
108
|
+
* Async, so close's concurrent gate children keep draining.
|
|
109
|
+
*
|
|
110
|
+
* @param {ReadonlyArray<string>} args
|
|
111
|
+
* @param {{ cwd: string, input?: string, timeout: number, run?: Function, platform: string }} opts
|
|
112
|
+
* @returns {Promise<{ status: number, stdout: string, stderr: string }>}
|
|
107
113
|
*/
|
|
108
|
-
function
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
117
|
-
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
];
|
|
122
|
-
}
|
|
114
|
+
function runClaude(args, { cwd, input, timeout, run, platform }) {
|
|
115
|
+
const shell = platform === 'win32';
|
|
116
|
+
return spawnCaptureAsync(
|
|
117
|
+
'claude',
|
|
118
|
+
shell ? args.map(quoteWindowsArg) : [...args],
|
|
119
|
+
{
|
|
120
|
+
cwd,
|
|
121
|
+
input: input ?? '',
|
|
122
|
+
shell,
|
|
123
|
+
timeout,
|
|
124
|
+
...(run ? { run } : {}),
|
|
125
|
+
},
|
|
126
|
+
);
|
|
123
127
|
}
|
|
124
128
|
|
|
125
129
|
/**
|
|
@@ -139,69 +143,105 @@ function assertInput(input) {
|
|
|
139
143
|
}
|
|
140
144
|
}
|
|
141
145
|
|
|
146
|
+
function runAuthStatus(runDir, deps) {
|
|
147
|
+
return runClaude(AUTH_ARGS, {
|
|
148
|
+
cwd: runDir,
|
|
149
|
+
timeout: AUTH_TIMEOUT_MS,
|
|
150
|
+
run: deps.authSpawnFn,
|
|
151
|
+
platform: deps.platform,
|
|
152
|
+
});
|
|
153
|
+
}
|
|
154
|
+
|
|
155
|
+
function runReviewer(runDir, prompt, deps) {
|
|
156
|
+
return runClaude(CLAUDE_ARGS, {
|
|
157
|
+
cwd: runDir,
|
|
158
|
+
input: prompt,
|
|
159
|
+
timeout: INVOKE_TIMEOUT_MS,
|
|
160
|
+
run: deps.spawnFn,
|
|
161
|
+
platform: deps.platform,
|
|
162
|
+
});
|
|
163
|
+
}
|
|
164
|
+
|
|
165
|
+
/**
|
|
166
|
+
* Preflight, then the review, both from a fresh OS temp dir removed after.
|
|
167
|
+
*
|
|
168
|
+
* @returns {Promise<{ findings: Finding[] } | { reason: string }>}
|
|
169
|
+
*/
|
|
170
|
+
async function reviewInTempDir(prompt, deps) {
|
|
171
|
+
const runDir = fs.mkdtempSync(path.join(os.tmpdir(), 'mandrel-code-review-'));
|
|
172
|
+
try {
|
|
173
|
+
const auth = readAuthStatus(await runAuthStatus(runDir, deps));
|
|
174
|
+
if (!auth.ok) return auth;
|
|
175
|
+
deps.logger?.info?.(
|
|
176
|
+
`[code-review] Invoking claude --print --effort low (${auth.detail})...`,
|
|
177
|
+
);
|
|
178
|
+
return resolveVerdict(await runReviewer(runDir, prompt, deps));
|
|
179
|
+
} finally {
|
|
180
|
+
fs.rmSync(runDir, { recursive: true, force: true });
|
|
181
|
+
}
|
|
182
|
+
}
|
|
183
|
+
|
|
184
|
+
/**
|
|
185
|
+
* @returns {{ reason: string } | { text: string }}
|
|
186
|
+
*/
|
|
187
|
+
function readDiff(gitSpawnFn, { baseRef, headRef }) {
|
|
188
|
+
const range = `${baseRef}...${headRef}`;
|
|
189
|
+
const diff = gitSpawnFn(PROJECT_ROOT, 'diff', '--no-color', range);
|
|
190
|
+
if (diff.status === 0) return { text: diff.stdout };
|
|
191
|
+
return {
|
|
192
|
+
reason: `git diff ${range} failed: ${clip(diff.stderr) || '<no output>'}`,
|
|
193
|
+
};
|
|
194
|
+
}
|
|
195
|
+
|
|
196
|
+
function assertCliPresent(probeFn) {
|
|
197
|
+
if (probeFn()) return;
|
|
198
|
+
throw new Error(
|
|
199
|
+
'[ReviewProviderFactory] codeReview provider "code-review" requires ' +
|
|
200
|
+
'the `claude` CLI on PATH but it was not detected. Install the ' +
|
|
201
|
+
'Claude Code CLI, or keep the entry `optional: true` so hosts ' +
|
|
202
|
+
'without it skip the model bug review.',
|
|
203
|
+
);
|
|
204
|
+
}
|
|
205
|
+
|
|
142
206
|
/**
|
|
143
207
|
* @param {{
|
|
144
208
|
* probeFn?: () => boolean,
|
|
145
209
|
* gitSpawnFn?: typeof gitSpawn,
|
|
146
210
|
* spawnFn?: Function,
|
|
211
|
+
* authSpawnFn?: Function,
|
|
212
|
+
* platform?: string,
|
|
147
213
|
* logger?: { info?: Function, warn?: Function },
|
|
148
|
-
* }} [deps] - `spawnFn`
|
|
149
|
-
*
|
|
214
|
+
* }} [deps] - `spawnFn` / `authSpawnFn` replace the async runner for the
|
|
215
|
+
* review and the `claude auth status` preflight.
|
|
216
|
+
* @returns {ReviewProvider & { getDegradations: () => GateDegradation[] }}
|
|
150
217
|
*/
|
|
151
218
|
export function createCodeReviewProviderForRegistry(deps = {}) {
|
|
152
|
-
|
|
153
|
-
if (!probeFn()) {
|
|
154
|
-
throw new Error(
|
|
155
|
-
'[ReviewProviderFactory] codeReview provider "code-review" requires ' +
|
|
156
|
-
'the `claude` CLI on PATH but it was not detected. Install the ' +
|
|
157
|
-
'Claude Code CLI, or keep the entry `optional: true` so hosts ' +
|
|
158
|
-
'without it skip the model bug review.',
|
|
159
|
-
);
|
|
160
|
-
}
|
|
219
|
+
assertCliPresent(deps.probeFn ?? probeClaudeCli);
|
|
161
220
|
const gitSpawnFn = deps.gitSpawnFn ?? gitSpawn;
|
|
162
|
-
const {
|
|
221
|
+
const runDeps = { ...deps, platform: deps.platform ?? process.platform };
|
|
222
|
+
/** @type {GateDegradation[]} */
|
|
223
|
+
let degradations = [];
|
|
224
|
+
|
|
225
|
+
function settle(verdict) {
|
|
226
|
+
if (!('reason' in verdict)) return verdict.findings;
|
|
227
|
+
degradations.push({ ...DEGRADATION_BASE, reason: verdict.reason });
|
|
228
|
+
deps.logger?.warn?.(`[code-review] No verdict: ${verdict.reason}.`);
|
|
229
|
+
return [];
|
|
230
|
+
}
|
|
163
231
|
|
|
164
232
|
return {
|
|
165
233
|
async runReview(input) {
|
|
234
|
+
degradations = [];
|
|
166
235
|
assertInput(input);
|
|
167
|
-
const
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
`${baseRef}...${headRef}`,
|
|
173
|
-
);
|
|
174
|
-
if (diff.status !== 0) {
|
|
175
|
-
return [
|
|
176
|
-
advisory(
|
|
177
|
-
'Code review could not read the diff',
|
|
178
|
-
`\`git diff ${baseRef}...${headRef}\` failed: ${diff.stderr || '<no output>'}.`,
|
|
179
|
-
),
|
|
180
|
-
];
|
|
181
|
-
}
|
|
182
|
-
if (diff.stdout.trim().length === 0) return [];
|
|
183
|
-
|
|
184
|
-
logger?.info?.(
|
|
185
|
-
`[code-review] Invoking claude --print --effort low for Story #${ticketId} (${baseRef}...${headRef})...`,
|
|
186
|
-
);
|
|
187
|
-
const result = await invokeClaude(
|
|
188
|
-
buildPrompt(input, diff.stdout),
|
|
189
|
-
spawnFn,
|
|
236
|
+
const diff = readDiff(gitSpawnFn, input);
|
|
237
|
+
if ('reason' in diff) return settle(diff);
|
|
238
|
+
if (diff.text.trim().length === 0) return [];
|
|
239
|
+
return settle(
|
|
240
|
+
await reviewInTempDir(buildPrompt(input, diff.text), runDeps),
|
|
190
241
|
);
|
|
191
|
-
|
|
192
|
-
|
|
193
|
-
|
|
194
|
-
);
|
|
195
|
-
return [
|
|
196
|
-
advisory(
|
|
197
|
-
'Code review did not complete',
|
|
198
|
-
`\`claude --print\` exited with status ${result.status}: ${
|
|
199
|
-
result.stderr || result.stdout || '<no output>'
|
|
200
|
-
}.`,
|
|
201
|
-
),
|
|
202
|
-
];
|
|
203
|
-
}
|
|
204
|
-
return parseFindings(result.stdout);
|
|
242
|
+
},
|
|
243
|
+
getDegradations() {
|
|
244
|
+
return degradations.map((d) => ({ ...d }));
|
|
205
245
|
},
|
|
206
246
|
};
|
|
207
247
|
}
|
|
@@ -257,19 +257,38 @@ function baselinesEnvelopeGates(validationGates) {
|
|
|
257
257
|
return out;
|
|
258
258
|
}
|
|
259
259
|
|
|
260
|
+
/**
|
|
261
|
+
* `overridden` > `degraded` > `passed`: an override records the human
|
|
262
|
+
* decision on a failed review; `degraded` says a review surface produced no
|
|
263
|
+
* verdict (reported, not blocking — a secondary review does not gate merge).
|
|
264
|
+
*
|
|
265
|
+
* @param {object|null} reviewOverride
|
|
266
|
+
* @param {boolean} [reviewDegraded]
|
|
267
|
+
* @returns {'overridden'|'degraded'|'passed'}
|
|
268
|
+
*/
|
|
269
|
+
function codeReviewGate(reviewOverride, reviewDegraded) {
|
|
270
|
+
if (reviewOverride) return 'overridden';
|
|
271
|
+
return reviewDegraded ? 'degraded' : 'passed';
|
|
272
|
+
}
|
|
273
|
+
|
|
260
274
|
/**
|
|
261
275
|
* @param {{ skipValidation?: boolean, skipSync?: boolean }} options
|
|
262
276
|
* @param {Record<string, string>|null} validationGates
|
|
263
277
|
* @param {object|null} reviewOverride
|
|
278
|
+
* @param {boolean} [reviewDegraded]
|
|
264
279
|
* @returns {Record<string, string>}
|
|
265
280
|
*/
|
|
266
|
-
function closeEnvelopeGates(
|
|
281
|
+
function closeEnvelopeGates(
|
|
282
|
+
options,
|
|
283
|
+
validationGates,
|
|
284
|
+
reviewOverride,
|
|
285
|
+
reviewDegraded,
|
|
286
|
+
) {
|
|
267
287
|
return {
|
|
268
288
|
validation: options.skipValidation ? 'skipped' : 'passed',
|
|
269
289
|
...baselinesEnvelopeGates(validationGates),
|
|
270
290
|
baseSync: options.skipSync ? 'skipped' : 'passed',
|
|
271
|
-
|
|
272
|
-
codeReview: reviewOverride ? 'overridden' : 'passed',
|
|
291
|
+
codeReview: codeReviewGate(reviewOverride, reviewDegraded),
|
|
273
292
|
};
|
|
274
293
|
}
|
|
275
294
|
|
|
@@ -466,7 +485,13 @@ async function openAndReviewPr(ctx, deps) {
|
|
|
466
485
|
const reviewOverride = reviewOutcome.halted
|
|
467
486
|
? await resolveReviewHalt(ctx, deps, { prUrl, prNumber, reviewOutcome })
|
|
468
487
|
: null;
|
|
469
|
-
return {
|
|
488
|
+
return {
|
|
489
|
+
prUrl,
|
|
490
|
+
prNumber,
|
|
491
|
+
alreadyMerged: false,
|
|
492
|
+
reviewOverride,
|
|
493
|
+
reviewDegraded: reviewOutcome.degraded === true,
|
|
494
|
+
};
|
|
470
495
|
}
|
|
471
496
|
|
|
472
497
|
async function releaseLease({ storyId }, deps) {
|
|
@@ -1181,6 +1206,7 @@ async function finishPhase(ctx, deps) {
|
|
|
1181
1206
|
options,
|
|
1182
1207
|
ctx.prePush.validationGates,
|
|
1183
1208
|
ctx.pr.reviewOverride,
|
|
1209
|
+
ctx.pr.reviewDegraded,
|
|
1184
1210
|
),
|
|
1185
1211
|
};
|
|
1186
1212
|
if (!waitForMerge) {
|