orizu 0.5.17 → 0.5.19

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.
Files changed (41) hide show
  1. package/README.md +15 -0
  2. package/dist/archive-artifacts-cli.js +1 -3
  3. package/dist/artifact-pull.js +1 -3
  4. package/dist/diff-comments-cli.js +120 -32
  5. package/dist/error-response.js +17 -17
  6. package/dist/github-setup.js +3 -11
  7. package/dist/help.js +22 -2
  8. package/dist/index.js +43 -107
  9. package/dist/json-response.js +21 -0
  10. package/dist/markdown-report-input.js +67 -0
  11. package/dist/optimization-export-cli.js +30 -0
  12. package/dist/optimizations-list-cli.js +1 -3
  13. package/dist/prompt-draft-push.js +7 -5
  14. package/dist/prompt-length-wire.js +30 -0
  15. package/dist/prompt-report-cli.js +86 -0
  16. package/dist/prompt-summary-table.js +37 -0
  17. package/dist/scorer-draft-push.js +1 -22
  18. package/dist/scores-accept-cli.js +1 -3
  19. package/dist/zip-draft-push.js +1 -24
  20. package/package.json +1 -1
  21. package/src/archive-artifacts-cli.ts +1 -7
  22. package/src/artifact-pull.ts +1 -4
  23. package/src/diff-comments-cli.ts +177 -37
  24. package/src/error-response.ts +23 -21
  25. package/src/github-setup.ts +3 -12
  26. package/src/help.ts +22 -2
  27. package/src/index.ts +54 -132
  28. package/src/json-response.ts +37 -0
  29. package/src/markdown-report-input.ts +80 -0
  30. package/src/optimization-export-cli.ts +72 -0
  31. package/src/optimizations-list-cli.ts +1 -4
  32. package/src/prompt-draft-push.ts +8 -6
  33. package/src/prompt-length-wire.ts +65 -0
  34. package/src/prompt-report-cli.ts +135 -0
  35. package/src/prompt-summary-table.ts +57 -0
  36. package/src/scorer-draft-push.ts +1 -6
  37. package/src/scores-accept-cli.ts +1 -4
  38. package/src/zip-draft-push.ts +1 -6
  39. package/vendor/skills/orizu-cli/SKILL.md +1 -0
  40. package/vendor/skills/orizu-cli/references/cli-reference.md +15 -1
  41. package/vendor/skills/orizu-cli/references/prompt-control-plane.md +19 -0
package/README.md CHANGED
@@ -173,6 +173,14 @@ orizu logout
173
173
  <td>Archive or restore a prompt</td>
174
174
  <td><code>orizu prompts archive &lt;promptIdOrName&gt; --project ops-eval/support-qa</code><br><code>orizu prompts restore &lt;promptIdOrName&gt; --project ops-eval/support-qa</code></td>
175
175
  </tr>
176
+ <tr>
177
+ <td>Push a prompt with a report</td>
178
+ <td><code>orizu prompts push ./prompts/support --project ops-eval/support-qa --report-file ./prompt-report.md</code></td>
179
+ </tr>
180
+ <tr>
181
+ <td>Upload a report for an existing prompt version</td>
182
+ <td><code>orizu prompts report set --prompt &lt;promptIdOrName&gt; --project ops-eval/support-qa --version 3 --report-file ./prompt-report.md</code><br><code>prompts report upload</code> is an exact alias; select exactly one of <code>--version</code> or <code>--version-id</code>.</td>
183
+ </tr>
176
184
  <tr>
177
185
  <td colspan="2"><strong>Report comments</strong></td>
178
186
  </tr>
@@ -184,6 +192,13 @@ orizu logout
184
192
  <td>Add or update comments</td>
185
193
  <td><code>orizu comments add --run &lt;runId&gt; --body @comment.md --anchor "Summary" --lines 4:6</code><br><code>orizu comments reply &lt;commentId&gt; --body "Fixed"</code><br><code>orizu comments resolve &lt;commentId&gt;</code></td>
186
194
  </tr>
195
+ <tr>
196
+ <td colspan="2"><strong>Diff comments</strong></td>
197
+ </tr>
198
+ <tr>
199
+ <td>Export diff comments</td>
200
+ <td><code>orizu comments diff --run &lt;optimizationRunId&gt; [--from &lt;candidateId&gt; --to &lt;candidateId&gt;] [--detail hunk|diff|full] [--json]</code><br><code>orizu comments diff --prompt &lt;promptId&gt; [--from &lt;version&gt; --to &lt;version&gt;] [--detail hunk|diff|full] [--json]</code></td>
201
+ </tr>
187
202
  <tr>
188
203
  <td colspan="2"><strong>Datasets</strong></td>
189
204
  </tr>
@@ -7,6 +7,7 @@
7
7
  */
8
8
  import { authedFetch } from './http.js';
9
9
  import { extractErrorMessage } from './error-response.js';
10
+ import { sanitizeTerminalText } from './json-response.js';
10
11
  const UUID_PATTERN = /^[0-9a-f]{8}-[0-9a-f]{4}-[1-5][0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$/i;
11
12
  const ARTIFACT_TYPES = {
12
13
  app: 'app',
@@ -31,9 +32,6 @@ function argValue(args, flag) {
31
32
  }
32
33
  return args[index + 1];
33
34
  }
34
- function sanitizeTerminalText(value) {
35
- return String(value).replace(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '');
36
- }
37
35
  function columnWidths(headers, rows) {
38
36
  return headers.map((header, index) => Math.max(header.length, ...rows.map(row => row[index]?.length || 0)));
39
37
  }
@@ -16,11 +16,9 @@ import { mkdirSync, mkdtempSync, readdirSync, rmSync, writeFileSync } from 'fs';
16
16
  import { tmpdir } from 'os';
17
17
  import { dirname, join } from 'path';
18
18
  import { authedFetch } from './http.js';
19
+ import { sanitizeTerminalText } from './json-response.js';
19
20
  // Shared with the runners exec download path in index.ts.
20
21
  export const ARTIFACT_MAX_BYTES = 25 * 1024 * 1024;
21
- function sanitizeTerminalText(value) {
22
- return String(value).replace(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '');
23
- }
24
22
  function expandHomePath(path) {
25
23
  if (path.startsWith('~/')) {
26
24
  const home = process.env.HOME || '';
@@ -1,4 +1,8 @@
1
+ import { extractErrorMessage, isJsonErrorResponsePayload, } from './error-response.js';
2
+ import { sanitizeHumanInlineText } from './json-response.js';
3
+ import { formatFiniteLengthCount, } from './prompt-length-wire.js';
1
4
  const DETAIL_LEVELS = ['hunk', 'diff', 'full'];
5
+ const CONTEXT_UNAVAILABLE_WITHOUT_REASON = ' Context unavailable (no reason supplied)';
2
6
  function isCandidatePair(pair) {
3
7
  return 'candidateId' in pair.from;
4
8
  }
@@ -8,11 +12,91 @@ function usage() {
8
12
  function sanitizeHumanText(ctx, value) {
9
13
  return ctx.sanitizeTerminalText(value).replaceAll('\r', '');
10
14
  }
11
- function sanitizeHumanInlineText(ctx, value) {
12
- return sanitizeHumanText(ctx, value).replaceAll('\n', '\\n');
15
+ // Quote untrusted multiline blocks so none of their rows can impersonate a
16
+ // genuine hunk row, whose marker begins immediately after the four-space indent.
17
+ function printHumanTextBlock(ctx, value) {
18
+ for (const line of sanitizeHumanText(ctx, value).split('\n')) {
19
+ ctx.printLine(` │ ${line}`);
20
+ }
21
+ }
22
+ function hasExplicitlyEmptyJsonError(response, rawBody) {
23
+ try {
24
+ const payload = JSON.parse(rawBody);
25
+ return isJsonErrorResponsePayload(response, payload) && payload.error.length === 0;
26
+ }
27
+ catch {
28
+ return false;
29
+ }
30
+ }
31
+ async function extractDiffCommentsErrorMessage(response) {
32
+ const statusFallback = `HTTP ${response.status}`;
33
+ try {
34
+ const rawBody = await response.text();
35
+ if (hasExplicitlyEmptyJsonError(response, rawBody)) {
36
+ return statusFallback;
37
+ }
38
+ const copiedResponse = new Response(rawBody, { headers: response.headers });
39
+ const extractedMessage = await extractErrorMessage(copiedResponse);
40
+ return extractedMessage.trim() ? extractedMessage : statusFallback;
41
+ }
42
+ catch {
43
+ return statusFallback;
44
+ }
13
45
  }
14
46
  function humanLineNumber(ctx, line) {
15
- return line === null ? '-' : sanitizeHumanText(ctx, String(line + 1));
47
+ return line === null
48
+ ? '-'
49
+ : sanitizeHumanInlineText(ctx.sanitizeTerminalText, String(line + 1));
50
+ }
51
+ function formatStats(stats) {
52
+ const tokens = formatFiniteLengthCount(stats.tokens);
53
+ return `${tokens === null ? '—' : `~${tokens}`} tokens · ` +
54
+ `${formatFiniteLengthCount(stats.lines) ?? '—'} lines · ` +
55
+ `${formatFiniteLengthCount(stats.chars) ?? '—'} chars · ` +
56
+ `${formatFiniteLengthCount(stats.words) ?? '—'} words`;
57
+ }
58
+ function formatDeltaValue(value, metric, signed = false) {
59
+ const formatted = formatFiniteLengthCount(value, signed);
60
+ if (formatted === null)
61
+ return '—';
62
+ return metric === 'tokens' ? `~${formatted}` : formatted;
63
+ }
64
+ function printLengthMeasurements(ctx, pair) {
65
+ function unavailableSuffix(reason) {
66
+ if (typeof reason !== 'string' || reason.length === 0)
67
+ return '';
68
+ return ` (${sanitizeHumanInlineText(ctx.sanitizeTerminalText, reason)})`;
69
+ }
70
+ if (!pair.lengthStats && !pair.lengthDelta) {
71
+ ctx.printLine(` Length: unavailable${unavailableSuffix(pair.lengthDeltaUnavailableReason ?? pair.degraded?.reason)}`);
72
+ return;
73
+ }
74
+ if (pair.lengthStats) {
75
+ ctx.printLine(pair.lengthStats.from
76
+ ? ` From: ${formatStats(pair.lengthStats.from)}`
77
+ : ` From: unavailable${unavailableSuffix(pair.lengthStats.fromUnavailableReason)}`);
78
+ ctx.printLine(pair.lengthStats.to
79
+ ? ` To: ${formatStats(pair.lengthStats.to)}`
80
+ : ` To: unavailable${unavailableSuffix(pair.lengthStats.toUnavailableReason)}`);
81
+ }
82
+ if (!pair.lengthDelta) {
83
+ ctx.printLine(` Length delta: unavailable${unavailableSuffix(pair.lengthDeltaUnavailableReason ?? pair.degraded?.reason)}`);
84
+ return;
85
+ }
86
+ if (pair.lengthDelta.splitUnavailableReason) {
87
+ ctx.printLine(` Split unavailable: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, pair.lengthDelta.splitUnavailableReason)}`);
88
+ }
89
+ for (const [metric, label] of [
90
+ ['tokens', 'Tokens'],
91
+ ['lines', 'Lines'],
92
+ ['chars', 'Chars'],
93
+ ['words', 'Words'],
94
+ ]) {
95
+ const delta = pair.lengthDelta[metric];
96
+ ctx.printLine(` ${label}: removed ${formatDeltaValue(delta.removed, metric)} · ` +
97
+ `added ${formatDeltaValue(delta.added, metric)} · ` +
98
+ `net ${formatDeltaValue(delta.net, metric, true)}`);
99
+ }
16
100
  }
17
101
  function buildParams(ctx) {
18
102
  const runId = ctx.getArg('--run');
@@ -29,9 +113,15 @@ function buildParams(ctx) {
29
113
  ]) {
30
114
  ctx.rejectDashPrefixedOptionValue(name, value);
31
115
  }
116
+ if (runId === '' || promptId === '') {
117
+ throw new Error(usage());
118
+ }
32
119
  if (Number(Boolean(runId)) + Number(Boolean(promptId)) !== 1) {
33
120
  throw new Error(usage());
34
121
  }
122
+ if (from === '' || to === '') {
123
+ throw new Error('--from and --to must not be empty');
124
+ }
35
125
  if (Boolean(from) !== Boolean(to)) {
36
126
  throw new Error('--from and --to must be provided together');
37
127
  }
@@ -65,15 +155,15 @@ function printDiffComments(ctx, payload) {
65
155
  const to = hasCandidateIds ? pair.to.candidateId : String(pair.to.version);
66
156
  ctx.printLine('');
67
157
  ctx.printLine(`${hasCandidateIds ? 'Candidates' : 'Versions'} ` +
68
- `${sanitizeHumanInlineText(ctx, from)} → ${sanitizeHumanInlineText(ctx, to)}`);
158
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, from)} → ` +
159
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, to)}`);
160
+ printLengthMeasurements(ctx, pair);
69
161
  if (pair.degraded) {
70
- ctx.printLine(` Context unavailable: ${sanitizeHumanText(ctx, pair.degraded.reason)}`);
162
+ ctx.printLine(` Context unavailable: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, pair.degraded.reason)}`);
71
163
  }
72
164
  if (pair.diff !== null) {
73
165
  ctx.printLine(' Diff:');
74
- for (const line of sanitizeHumanText(ctx, pair.diff).split('\n')) {
75
- ctx.printLine(` ${line}`);
76
- }
166
+ printHumanTextBlock(ctx, pair.diff);
77
167
  }
78
168
  if (pair.bodies !== null) {
79
169
  for (const [label, body] of [
@@ -81,24 +171,31 @@ function printDiffComments(ctx, payload) {
81
171
  ['To body', pair.bodies.to],
82
172
  ]) {
83
173
  ctx.printLine(` ${label}:`);
84
- for (const line of sanitizeHumanText(ctx, body).split('\n')) {
85
- ctx.printLine(` ${line}`);
86
- }
174
+ printHumanTextBlock(ctx, body);
87
175
  }
88
176
  }
89
177
  for (const comment of pair.comments) {
90
178
  const authorName = comment.author?.person?.name || comment.author?.name || 'Unknown';
91
- ctx.printLine(` ${sanitizeHumanInlineText(ctx, comment.id)} · ` +
92
- `${sanitizeHumanInlineText(ctx, authorName)} · ` +
93
- `${sanitizeHumanInlineText(ctx, comment.createdAt)} · ` +
94
- `${sanitizeHumanInlineText(ctx, comment.anchor.side)} line ` +
95
- `${sanitizeHumanInlineText(ctx, String(comment.anchor.line + 1))}`);
96
- for (const line of sanitizeHumanText(ctx, comment.body).split('\n')) {
97
- ctx.printLine(` ${line}`);
179
+ ctx.printLine(` ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.id)} · ` +
180
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, authorName)} · ` +
181
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.createdAt)} · ` +
182
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.anchor.side)} line ` +
183
+ `${sanitizeHumanInlineText(ctx.sanitizeTerminalText, String(comment.anchor.line + 1))}`);
184
+ printHumanTextBlock(ctx, comment.body);
185
+ if (comment.contextUnavailableReason) {
186
+ ctx.printLine(` Context unavailable: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.contextUnavailableReason)}`);
187
+ if (comment.context) {
188
+ ctx.printLine(` Anchored line: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.context.lineText)}`);
189
+ }
190
+ continue;
98
191
  }
99
192
  if (!comment.context) {
100
- ctx.printLine(` Context unavailable: ` +
101
- `${sanitizeHumanText(ctx, comment.contextUnavailableReason ?? 'body_unresolvable')}`);
193
+ ctx.printLine(CONTEXT_UNAVAILABLE_WITHOUT_REASON);
194
+ continue;
195
+ }
196
+ if (!comment.context.hunk) {
197
+ ctx.printLine(CONTEXT_UNAVAILABLE_WITHOUT_REASON);
198
+ ctx.printLine(` Anchored line: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, comment.context.lineText)}`);
102
199
  continue;
103
200
  }
104
201
  for (const line of comment.context.hunk.lines) {
@@ -109,7 +206,7 @@ function printDiffComments(ctx, payload) {
109
206
  const marker = isAnchored ? '>' : ' ';
110
207
  ctx.printLine(` ${marker} old:${humanLineNumber(ctx, line.oldLine)} ` +
111
208
  `new:${humanLineNumber(ctx, line.newLine)} | ` +
112
- `${prefix}${sanitizeHumanText(ctx, line.text)}`);
209
+ `${prefix}${sanitizeHumanInlineText(ctx.sanitizeTerminalText, line.text)}`);
113
210
  }
114
211
  }
115
212
  }
@@ -118,17 +215,8 @@ export async function diffCommentsCommand(ctx) {
118
215
  const params = buildParams(ctx);
119
216
  const response = await ctx.authedFetch(`/api/cli/diff-comments?${params.toString()}`);
120
217
  if (!response.ok) {
121
- let message = `HTTP ${response.status}`;
122
- try {
123
- const payload = await response.clone().json();
124
- if (typeof payload.error === 'string' && payload.error.trim()) {
125
- message = sanitizeHumanInlineText(ctx, payload.error);
126
- }
127
- }
128
- catch {
129
- // Keep the status-only error for non-JSON responses.
130
- }
131
- throw new Error(`Failed to fetch diff comments: ${message}`);
218
+ const message = await extractDiffCommentsErrorMessage(response);
219
+ throw new Error(`Failed to fetch diff comments: ${sanitizeHumanInlineText(ctx.sanitizeTerminalText, message)}`);
132
220
  }
133
221
  const payload = await ctx.parseJsonResponse(response, 'Diff comments');
134
222
  if (ctx.hasJsonFlag()) {
@@ -1,26 +1,26 @@
1
+ import { sanitizeTerminalText } from './json-response.js';
1
2
  const ERROR_BODY_PREVIEW_LIMIT = 180;
3
+ export function isJsonErrorResponsePayload(response, payload) {
4
+ const contentType = response.headers.get('content-type') || '';
5
+ return (contentType.includes('application/json') &&
6
+ payload !== null &&
7
+ typeof payload === 'object' &&
8
+ !Array.isArray(payload) &&
9
+ typeof payload.error === 'string');
10
+ }
2
11
  function boundedTerminalPreview(value) {
3
- return String(value)
4
- .replace(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '')
5
- .slice(0, ERROR_BODY_PREVIEW_LIMIT);
12
+ return sanitizeTerminalText(value).slice(0, ERROR_BODY_PREVIEW_LIMIT);
6
13
  }
7
14
  export async function extractErrorMessage(response) {
8
- const contentType = response.headers.get('content-type') || '';
9
15
  const rawBody = await response.text();
10
- if (contentType.includes('application/json')) {
11
- try {
12
- const payload = JSON.parse(rawBody);
13
- if (payload &&
14
- typeof payload === 'object' &&
15
- !Array.isArray(payload) &&
16
- typeof payload.error === 'string' &&
17
- payload.error.length > 0) {
18
- return boundedTerminalPreview(payload.error);
19
- }
20
- }
21
- catch {
22
- // Fall through to raw body
16
+ try {
17
+ const payload = JSON.parse(rawBody);
18
+ if (isJsonErrorResponsePayload(response, payload) && payload.error.length > 0) {
19
+ return boundedTerminalPreview(payload.error);
23
20
  }
24
21
  }
22
+ catch {
23
+ // Fall through to raw body
24
+ }
25
25
  return boundedTerminalPreview(rawBody);
26
26
  }
@@ -17,16 +17,8 @@ import { spawnSync } from 'child_process';
17
17
  import { existsSync, mkdirSync } from 'fs';
18
18
  import { join } from 'path';
19
19
  import { authedFetch } from './http.js';
20
+ import { sanitizeTerminalText } from './json-response.js';
20
21
  import { assertWorkspaceDirUsable, workspaceExists } from './workspace.js';
21
- /**
22
- * Strip terminal control characters from server-supplied strings before they
23
- * are printed (mirrors index.ts `sanitizeTerminalText`), so a hostile or
24
- * mangled provisioning `warnings[]`/`defaultBranch` value can't inject escape
25
- * sequences into the user's terminal.
26
- */
27
- function sanitizeServerText(value) {
28
- return String(value).replace(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '');
29
- }
30
22
  const DEFAULT_POLL_INTERVAL_MS = 2000;
31
23
  // ALI-1141: long enough for a slow org-picker + repo-selection walk, but the
32
24
  // wait MUST be finite — when the App is already installed in the target org,
@@ -231,10 +223,10 @@ export async function runHostedAttach(options, io) {
231
223
  repoFullName = data.repo;
232
224
  provisioned = true;
233
225
  if (data.rulesetApplied === true) {
234
- io.print(` Default-branch protection applied (ruleset "orizu-default-branch-protection"${data.defaultBranch ? ` on ${sanitizeServerText(data.defaultBranch)}` : ''}).`);
226
+ io.print(` Default-branch protection applied (ruleset "orizu-default-branch-protection"${data.defaultBranch ? ` on ${sanitizeTerminalText(data.defaultBranch)}` : ''}).`);
235
227
  }
236
228
  for (const warning of data.warnings ?? []) {
237
- io.print(` ⚠ ${sanitizeServerText(warning)}`);
229
+ io.print(` ⚠ ${sanitizeTerminalText(warning)}`);
238
230
  }
239
231
  }
240
232
  const cleanUrl = `https://github.com/${repoFullName}.git`;
package/dist/help.js CHANGED
@@ -337,10 +337,30 @@ export const COMMAND_DOCS = [
337
337
  },
338
338
  {
339
339
  path: ['prompts', 'push'],
340
- usage: 'orizu prompts push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--session <session-id>] [--json]',
340
+ usage: 'orizu prompts push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--session <session-id>] [--report-file <path>] [--json]',
341
341
  summary: 'Push a local prompt artifact directory as a new version (with --session: a commit-first git draft).',
342
342
  group: 'Prompts and judges',
343
343
  },
344
+ {
345
+ path: ['prompts', 'report', 'set'],
346
+ usage: 'orizu prompts report set --prompt <id-or-name> --project <team/project> (--version <n> | --version-id <id>) (--report <markdown|@file> | --report-file <path>) [--json]',
347
+ summary: 'Upload or replace the Markdown report for one explicit prompt version.',
348
+ group: 'Prompts and judges',
349
+ aliases: [['prompts', 'report', 'upload']],
350
+ options: [
351
+ { name: '--prompt <id-or-name>', help: 'Prompt ID or name.', required: true },
352
+ { name: '--project <team/project>', help: 'Project slug containing the prompt.', required: true },
353
+ { name: '--version <n>', help: 'Positive prompt version number.' },
354
+ { name: '--version-id <id>', help: 'Explicit prompt version ID.' },
355
+ { name: '--report <markdown|@file>', help: 'Inline Markdown or an @file reference.' },
356
+ { name: '--report-file <path>', help: 'Markdown file to upload.' },
357
+ { name: '--json', help: 'Emit the updated prompt report payload.' },
358
+ ],
359
+ examples: [
360
+ 'orizu prompts report set --prompt Generator --project core/evals --version 3 --report-file ./report.md',
361
+ 'orizu prompts report upload --prompt Generator --project core/evals --version-id <id> --report @./report.md',
362
+ ],
363
+ },
344
364
  {
345
365
  path: ['prompts', 'labels', 'set'],
346
366
  usage: 'orizu prompts labels set <prompt-name> <label> --version <version-id> [--project <team/project>] [--json]',
@@ -373,7 +393,7 @@ export const COMMAND_DOCS = [
373
393
  },
374
394
  {
375
395
  path: ['judges', 'push'],
376
- usage: 'orizu judges push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--session <session-id>] [--json]',
396
+ usage: 'orizu judges push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--session <session-id>] [--report-file <path>] [--json]',
377
397
  summary: 'Push a local judge artifact directory as a new version (with --session: a commit-first git draft).',
378
398
  group: 'Prompts and judges',
379
399
  },
package/dist/index.js CHANGED
@@ -41,12 +41,18 @@ import { hostedCommand } from './hosted-session-cli.js';
41
41
  import { workspaceSyncCommand } from './workspace-sync.js';
42
42
  import { runGitCredentialInvocation } from './git-credential.js';
43
43
  import { pushPromptDraft } from './prompt-draft-push.js';
44
+ import { readMarkdownReportInput } from './markdown-report-input.js';
45
+ import { promptPushErrorMessage, promptReportCommand } from './prompt-report-cli.js';
46
+ import { printPromptSummaryTable } from './prompt-summary-table.js';
44
47
  import { assertSnapshotManifestConfined, verifyGepaRunnerDirsFromArgs, verifyRunnerDirRegistered } from './runner-dir-verify.js';
45
48
  import { runScorersRegister } from './scorer-draft-push.js';
46
49
  import { runZipArtifactPush } from './zip-draft-push.js';
47
50
  import { runGithubLink, runInteractiveHostedSetup } from './github-setup.js';
51
+ import { parseJsonResponse, sanitizeTerminalText } from './json-response.js';
48
52
  import { reportCommentsCommand, throwDeprecatedPromptCommentsCommand, } from './report-comments-cli.js';
49
53
  import { diffCommentsCommand } from './diff-comments-cli.js';
54
+ import { exportOptimizationRunCommand } from './optimization-export-cli.js';
55
+ export { parseJsonResponse, sanitizeTerminalText } from './json-response.js';
50
56
  function getErrorMessage(error) {
51
57
  return error instanceof Error ? error.message : String(error);
52
58
  }
@@ -176,9 +182,6 @@ function createCodeVerifier() {
176
182
  export function createCodeChallenge(verifier) {
177
183
  return createHash('sha256').update(verifier).digest('base64url');
178
184
  }
179
- export function sanitizeTerminalText(value) {
180
- return String(value).replace(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '');
181
- }
182
185
  function printError(message) {
183
186
  console.error(sanitizeTerminalText(message));
184
187
  }
@@ -241,45 +244,9 @@ export function formatTerminalLink(url) {
241
244
  }
242
245
  return `\u001B]8;;${safeUrl}\u0007${safeUrl}\u001B]8;;\u0007`;
243
246
  }
244
- export async function parseJsonResponse(response, context) {
245
- const contentType = response.headers.get('content-type') || '';
246
- const rawBody = await response.text();
247
- if (!contentType.includes('application/json')) {
248
- throw new Error(`${context} returned non-JSON response (status ${response.status}). ` +
249
- `Body preview: ${sanitizeTerminalText(rawBody.slice(0, 180))}`);
250
- }
251
- try {
252
- return JSON.parse(rawBody);
253
- }
254
- catch {
255
- throw new Error(`${context} returned invalid JSON (status ${response.status}). ` +
256
- `Body preview: ${sanitizeTerminalText(rawBody.slice(0, 180))}`);
257
- }
258
- }
259
247
  function shellQuote(value) {
260
248
  return `"${value.replace(/\\/g, '\\\\').replace(/"/g, '\\"')}"`;
261
249
  }
262
- function printPromptSummaries(items, emptyMessage) {
263
- if (items.length === 0) {
264
- printLine(emptyMessage);
265
- return;
266
- }
267
- const rows = items.map(item => ({
268
- id: sanitizeTerminalText(item.id),
269
- name: sanitizeTerminalText(item.name),
270
- role: sanitizeTerminalText(item.role),
271
- status: sanitizeTerminalText(item.status || 'active'),
272
- }));
273
- const idWidth = Math.max('ID'.length, ...rows.map(row => row.id.length));
274
- const nameWidth = Math.max('NAME'.length, ...rows.map(row => row.name.length));
275
- const roleWidth = Math.max('ROLE'.length, ...rows.map(row => row.role.length));
276
- const statusWidth = Math.max('STATUS'.length, ...rows.map(row => row.status.length));
277
- printLine(`${'ID'.padEnd(idWidth)} ${'NAME'.padEnd(nameWidth)} ${'ROLE'.padEnd(roleWidth)} ${'STATUS'.padEnd(statusWidth)}`);
278
- printLine(`${'-'.repeat(idWidth)} ${'-'.repeat(nameWidth)} ${'-'.repeat(roleWidth)} ${'-'.repeat(statusWidth)}`);
279
- rows.forEach(row => {
280
- printLine(`${row.id.padEnd(idWidth)} ${row.name.padEnd(nameWidth)} ${row.role.padEnd(roleWidth)} ${row.status.padEnd(statusWidth)}`);
281
- });
282
- }
283
250
  function readJsonObjectArg(valueArg, label) {
284
251
  if (!valueArg) {
285
252
  return {};
@@ -918,7 +885,7 @@ async function listPrompts() {
918
885
  printJson(data);
919
886
  return;
920
887
  }
921
- printPromptSummaries(data.prompts, 'No prompts found.');
888
+ printPromptSummaryTable(data.prompts, 'No prompts found.', printLine);
922
889
  }
923
890
  async function listJudges() {
924
891
  const project = getArg('--project') || await resolveProjectSlug(null);
@@ -936,7 +903,7 @@ async function listJudges() {
936
903
  printJson(data);
937
904
  return;
938
905
  }
939
- printPromptSummaries(data.judges, 'No judges found.');
906
+ printPromptSummaryTable(data.judges, 'No judges found.', printLine);
940
907
  }
941
908
  async function listScorers() {
942
909
  const project = getArg('--project') || await resolveProjectSlug(null);
@@ -1014,14 +981,17 @@ async function pushRunnerArtifact(kind) {
1014
981
  }
1015
982
  async function pushPromptArtifact(kind) {
1016
983
  const promptDir = getPositionalArg(2);
1017
- const project = getArg('--project') || await resolveProjectSlug(null);
984
+ const projectArg = getArg('--project');
1018
985
  const runnerVersionArg = getArg('--runner-version');
1019
986
  const parentVersionId = getArg('--parent') || undefined;
1020
987
  const sessionId = getArg('--session') || undefined;
1021
- const usage = `Usage: orizu ${kind === 'judge' ? 'judges' : 'prompts'} push <dir> --project <team/project> [--runner-version <id>] [--parent <version-id>] [--session <session-id>] [--json]`;
1022
- if (!promptDir) {
988
+ const usage = `Usage: orizu ${kind === 'judge' ? 'judges' : 'prompts'} push <dir> --project <team/project> [--runner-version <id>] [--parent <version-id>] [--session <session-id>] [--report-file <path>] [--json]`;
989
+ if (!promptDir)
1023
990
  throw new Error(usage);
1024
- }
991
+ if (cliArgs.some(arg => arg === '--report' || arg.startsWith('--report=')))
992
+ throw new Error(`${kind === 'judge' ? 'judges' : 'prompts'} push accepts report input only through --report-file`);
993
+ const report = readMarkdownReportInput(cliArgs, 'Prompt');
994
+ const project = projectArg || await resolveProjectSlug(null);
1025
995
  const promptRoot = expandHomePath(promptDir);
1026
996
  const manifest = readJsonFile(join(promptRoot, 'orizu.prompt.json'));
1027
997
  const primaryText = readPromptPrimaryText(manifest, promptRoot);
@@ -1040,6 +1010,7 @@ async function pushPromptArtifact(kind) {
1040
1010
  kind, project, sessionId, manifest, sidecars, runnerVersionId, parentVersionId,
1041
1011
  displayName: String(manifest.name || promptDir),
1042
1012
  bodyKind: primaryText.bodyKind, primaryBody: primaryText.body,
1013
+ report,
1043
1014
  });
1044
1015
  if (hasJsonFlag())
1045
1016
  printJson(data);
@@ -1074,10 +1045,16 @@ async function pushPromptArtifact(kind) {
1074
1045
  parentVersionId,
1075
1046
  versionLabel: manifest.version_label,
1076
1047
  createdBy: manifest.provenance || { kind: 'human-edit' },
1048
+ ...(report
1049
+ ? {
1050
+ reportMarkdown: report.markdown,
1051
+ reportSourceName: report.sourceName,
1052
+ }
1053
+ : {}),
1077
1054
  }),
1078
1055
  });
1079
1056
  if (!response.ok) {
1080
- throw new Error(`Failed to push ${kind}: ${await response.text()}`);
1057
+ throw new Error(await promptPushErrorMessage(response, kind, Boolean(report)));
1081
1058
  }
1082
1059
  const data = await parseJsonResponse(response, `${kind} push`);
1083
1060
  if (hasJsonFlag()) {
@@ -1658,7 +1635,6 @@ async function execScorer() {
1658
1635
  printLine(`Wrote scorer result to ${sanitizeTerminalText(outPath)}`);
1659
1636
  }
1660
1637
  }
1661
- const MARKDOWN_REPORT_MAX_BYTES = 2 * 1024 * 1024;
1662
1638
  function parseOptionalNumberFlag(name) {
1663
1639
  const value = getArg(name);
1664
1640
  if (!value) {
@@ -1673,47 +1649,11 @@ function parseOptionalNumberFlag(name) {
1673
1649
  function hasObjectKeys(value) {
1674
1650
  return Object.keys(value).length > 0;
1675
1651
  }
1676
- function readMarkdownReportInput(reportLabel) {
1677
- const report = getArg('--report');
1678
- const reportFile = getArg('--report-file');
1679
- rejectDashPrefixedOptionValue('--report', report);
1680
- rejectDashPrefixedOptionValue('--report-file', reportFile);
1681
- if (report && reportFile) {
1682
- throw new Error('Use either --report or --report-file, not both');
1683
- }
1684
- if (!report && !reportFile) {
1685
- return null;
1686
- }
1687
- let markdown;
1688
- let sourceName;
1689
- if (reportFile) {
1690
- const expandedPath = expandHomePath(reportFile);
1691
- markdown = readSourceFile(reportFile);
1692
- sourceName = basename(expandedPath);
1693
- }
1694
- else if (report?.startsWith('@')) {
1695
- const path = report.slice(1);
1696
- const expandedPath = expandHomePath(path);
1697
- markdown = readSourceFile(path);
1698
- sourceName = basename(expandedPath);
1699
- }
1700
- else {
1701
- markdown = report || '';
1702
- sourceName = 'inline';
1703
- }
1704
- if (!markdown.trim()) {
1705
- throw new Error(`${reportLabel} report markdown must not be blank`);
1706
- }
1707
- if (Buffer.byteLength(markdown, 'utf8') > MARKDOWN_REPORT_MAX_BYTES) {
1708
- throw new Error(`${reportLabel} report exceeds ${MARKDOWN_REPORT_MAX_BYTES} bytes`);
1709
- }
1710
- return { markdown, sourceName };
1711
- }
1712
1652
  function readOptimizationReportInput() {
1713
- return readMarkdownReportInput('Optimization');
1653
+ return readMarkdownReportInput(cliArgs, 'Optimization');
1714
1654
  }
1715
1655
  function readTaskReportInput() {
1716
- return readMarkdownReportInput('Task');
1656
+ return readMarkdownReportInput(cliArgs, 'Task');
1717
1657
  }
1718
1658
  async function startOptimizationRun() {
1719
1659
  const project = getArg('--project') || await resolveProjectSlug(null);
@@ -1853,27 +1793,6 @@ async function updateOptimizationRunLifecycle(action) {
1853
1793
  printLine(`Cancelled optimization run ${id}`);
1854
1794
  }
1855
1795
  }
1856
- async function exportOptimizationRun() {
1857
- const runId = getPositionalArg(2) || getArg('--run-id');
1858
- const outPathArg = getArg('--out');
1859
- if (!runId) {
1860
- throw new Error('Usage: orizu optimizations export <run-id> [--out <path>] [--json]');
1861
- }
1862
- const response = await authedFetch(`/api/cli/optimization-runs/${encodeURIComponent(runId)}/export`);
1863
- if (!response.ok) {
1864
- throw new Error(`Failed to export optimization run: ${await response.text()}`);
1865
- }
1866
- const data = await parseJsonResponse(response, 'Optimization export');
1867
- if (hasJsonFlag() && !outPathArg) {
1868
- printJson(data);
1869
- return;
1870
- }
1871
- const filename = outPathArg
1872
- ? expandHomePath(outPathArg)
1873
- : `${runId}.optimization.json`;
1874
- writeTextFileEnsuringDir(filename, `${JSON.stringify(data, null, 2)}\n`);
1875
- printLine(`Saved optimization export to ${sanitizeTerminalText(filename)}`);
1876
- }
1877
1796
  // ALI-1073 list moved to optimizations-list-cli.ts (line ratchet, ALI-976);
1878
1797
  // ALI-1175 labels its best scores as agent-reported evidence there.
1879
1798
  function removeFlagWithValue(args, flag) {
@@ -4929,6 +4848,13 @@ export async function main(rawArgs = process.argv.slice(2)) {
4929
4848
  await pushPromptArtifact('prompt');
4930
4849
  return;
4931
4850
  }
4851
+ if (command === 'prompts' && subcommand === 'report' &&
4852
+ (cliArgs[2] === 'set' || cliArgs[2] === 'upload')) {
4853
+ await promptReportCommand(cliArgs, {
4854
+ resolveProjectSlug, json: hasJsonFlag(), printJson, printLine,
4855
+ });
4856
+ return;
4857
+ }
4932
4858
  if (command === 'prompts' && subcommand === 'archive') {
4933
4859
  await setPromptArchivedState(true);
4934
4860
  return;
@@ -5032,7 +4958,17 @@ export async function main(rawArgs = process.argv.slice(2)) {
5032
4958
  return;
5033
4959
  }
5034
4960
  if (command === 'optimizations' && subcommand === 'export') {
5035
- await exportOptimizationRun();
4961
+ await exportOptimizationRunCommand({
4962
+ getArg,
4963
+ getPositionalArg,
4964
+ authedFetch,
4965
+ hasJsonFlag,
4966
+ printJson,
4967
+ printLine,
4968
+ expandHomePath,
4969
+ writeTextFileEnsuringDir,
4970
+ sanitizeTerminalText,
4971
+ });
5036
4972
  return;
5037
4973
  }
5038
4974
  if (command === 'optimizations' && subcommand === 'list') {