atris 3.68.3 → 3.68.5
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/commands/member.js +13 -3
- package/commands/mission.js +20 -0
- package/lib/engine-registry.js +15 -96
- package/lib/member-progress.js +82 -0
- package/lib/model-names.js +116 -0
- package/lib/runner-command.js +11 -10
- package/package.json +1 -1
package/commands/member.js
CHANGED
|
@@ -9925,6 +9925,11 @@ function memberDispatch(...args) {
|
|
|
9925
9925
|
if (!asJson) {
|
|
9926
9926
|
console.log(`${picked.member} (${picked.role}) is taking this because it ${picked.reason.replace(/^[A-Z]/, (c) => c.toLowerCase())}.`);
|
|
9927
9927
|
}
|
|
9928
|
+
// --ignite means do the work now, the same as the previewed member run;
|
|
9929
|
+
// without it dispatch only opens the mission.
|
|
9930
|
+
if (hasFlag(args, '--ignite') || hasFlag(args, '--wait') || hasFlag(args, '--sync')) {
|
|
9931
|
+
return memberRun(picked.member, ...forwardedArgs);
|
|
9932
|
+
}
|
|
9928
9933
|
return startMemberRunMission(picked.member, ideaText, forwardedArgs);
|
|
9929
9934
|
}
|
|
9930
9935
|
|
|
@@ -10494,16 +10499,21 @@ async function memberGo(...rawArgs) {
|
|
|
10494
10499
|
const root = process.cwd();
|
|
10495
10500
|
const teamDir = path.join(root, 'atris', 'team');
|
|
10496
10501
|
const coreOrder = ['auto-improver', 'executor', 'navigator', 'validator', 'architect', 'customer-lead', 'researcher'];
|
|
10497
|
-
let chosen = '
|
|
10502
|
+
let chosen = '';
|
|
10498
10503
|
for (const slug of coreOrder) {
|
|
10499
10504
|
const mDir = path.join(teamDir, slug);
|
|
10500
10505
|
if (!fs.existsSync(path.join(mDir, 'MEMBER.md'))) continue;
|
|
10501
|
-
|
|
10506
|
+
if (!chosen) chosen = slug;
|
|
10507
|
+
const gp = memberGoalPlaneStatus(slug, { memberDir: mDir, goalsJson: path.join(mDir, 'goals.json') });
|
|
10502
10508
|
if (gp?.goal?.next_push || gp?.goal?.title) {
|
|
10503
10509
|
chosen = slug;
|
|
10504
10510
|
break;
|
|
10505
10511
|
}
|
|
10506
10512
|
}
|
|
10513
|
+
if (!chosen) {
|
|
10514
|
+
const fallback = findAllMembers(teamDir).filter(isFleetMember)[0];
|
|
10515
|
+
if (fallback) chosen = memberSlugFromRecord(fallback);
|
|
10516
|
+
}
|
|
10507
10517
|
return memberSteerOrDispatch(chosen, ...rawArgs);
|
|
10508
10518
|
}
|
|
10509
10519
|
|
|
@@ -10846,7 +10856,7 @@ async function memberRoom(...args) {
|
|
|
10846
10856
|
: [' No work recorded yet. Run: atris member go'];
|
|
10847
10857
|
|
|
10848
10858
|
const lines = [
|
|
10849
|
-
'
|
|
10859
|
+
'Your team · workday briefing',
|
|
10850
10860
|
`Engines ready: ${engineSummary}`,
|
|
10851
10861
|
`Loop: ${room.utopia.loop}`,
|
|
10852
10862
|
'',
|
package/commands/mission.js
CHANGED
|
@@ -8,6 +8,7 @@ const { spawn, spawnSync } = require('child_process');
|
|
|
8
8
|
const { hasFlag, readFlag, readIntFlag } = require('../lib/arg-parser');
|
|
9
9
|
const { memberProcessPrompt } = require('../lib/member-context');
|
|
10
10
|
const { memberIdentityLine, missionObjectiveLine } = require('../lib/member-identity');
|
|
11
|
+
const { formatToolStep, appendProgress, reportProgress } = require('../lib/member-progress');
|
|
11
12
|
const {
|
|
12
13
|
compactErrorPayload,
|
|
13
14
|
compactSuccessPayload,
|
|
@@ -2208,6 +2209,10 @@ function missionVerifierCheckedText(verifierResult, mission, guard = null) {
|
|
|
2208
2209
|
? 'Engine verify passed after running the changed surface on this computer.'
|
|
2209
2210
|
: 'VERIFY FAILED: the engine verify pass did not end with VERDICT: PASS.';
|
|
2210
2211
|
}
|
|
2212
|
+
// A passing command plus a failing referee must name the referee, not the command.
|
|
2213
|
+
if (verifierResult.referee?.passed === false && verifierResult.status === 0) {
|
|
2214
|
+
return missionVerifierCheckedText(verifierResult.referee, mission, guard);
|
|
2215
|
+
}
|
|
2211
2216
|
const command = verifierResult.command || mission.verifier || 'configured verifier';
|
|
2212
2217
|
if (!verifierResult.passed && verifierResult.timed_out) {
|
|
2213
2218
|
return `VERIFY TIMED OUT: the check ran past its ${verifierTimeoutSeconds(verifierResult)}s window (took ${verifierElapsedSeconds(verifierResult)}s); the code is not failing, the suite is slow. Rerun with more room.`;
|
|
@@ -2242,6 +2247,9 @@ function missionVerifierHighLevelTestText(verifierResult, mission, guard = null)
|
|
|
2242
2247
|
? 'The verify engine inspected the tick diff and reported real changed-surface command output.'
|
|
2243
2248
|
: 'The verify engine did not provide a passing changed-surface verdict.';
|
|
2244
2249
|
}
|
|
2250
|
+
if (verifierResult.referee?.passed === false && verifierResult.status === 0) {
|
|
2251
|
+
return missionVerifierHighLevelTestText(verifierResult.referee, mission, guard);
|
|
2252
|
+
}
|
|
2245
2253
|
const command = verifierResult.command || mission.verifier || 'configured verifier';
|
|
2246
2254
|
if (!verifierResult.passed && verifierResult.timed_out) {
|
|
2247
2255
|
return `Verifier command timed out after ${verifierElapsedSeconds(verifierResult)}s against a ${verifierTimeoutSeconds(verifierResult)}s window: ${command}. No pass/fail verdict was reached; the suite outran the window rather than the code failing.`;
|
|
@@ -9370,6 +9378,9 @@ function spawnClaudeTick(mission, opts) {
|
|
|
9370
9378
|
env: { ...gitGuard.env, ATRIS_DISPATCH_DEPTH: String(dispatchDepth.depth + 1) },
|
|
9371
9379
|
});
|
|
9372
9380
|
updateMissionLockOwner(missionLock, proc.pid, { phase: 'tick' });
|
|
9381
|
+
const progressMember = mission.owner || 'mission';
|
|
9382
|
+
const seenToolIds = new Set();
|
|
9383
|
+
appendProgress(progressMember, `started: ${mission.objective || ''}`, mission.id);
|
|
9373
9384
|
|
|
9374
9385
|
let stdoutBuf = '';
|
|
9375
9386
|
let observedSessionIds = new Set();
|
|
@@ -9410,6 +9421,14 @@ function spawnClaudeTick(mission, opts) {
|
|
|
9410
9421
|
try {
|
|
9411
9422
|
const ev = JSON.parse(line);
|
|
9412
9423
|
if (ev.session_id) observedSessionIds.add(ev.session_id);
|
|
9424
|
+
if (ev.type === 'assistant' && Array.isArray(ev.message?.content)) {
|
|
9425
|
+
for (const block of ev.message.content) {
|
|
9426
|
+
const step = formatToolStep(block, cwd);
|
|
9427
|
+
if (!step || seenToolIds.has(block.id)) continue;
|
|
9428
|
+
if (block.id) seenToolIds.add(block.id);
|
|
9429
|
+
reportProgress(progressMember, step, mission.id);
|
|
9430
|
+
}
|
|
9431
|
+
}
|
|
9413
9432
|
if (ev.type === 'rate_limit_event' && ev.rate_limit_info) {
|
|
9414
9433
|
rateLimitInfo = ev.rate_limit_info;
|
|
9415
9434
|
}
|
|
@@ -9436,6 +9455,7 @@ function spawnClaudeTick(mission, opts) {
|
|
|
9436
9455
|
if (signal) signal.removeEventListener?.('abort', onAbort);
|
|
9437
9456
|
updateMissionLockOwner(missionLock, missionLock?.driverPid);
|
|
9438
9457
|
const ok = code === 0 && !isError && !timedOut && !aborted;
|
|
9458
|
+
appendProgress(progressMember, `finished: ${ok ? 'ok' : 'failed'}`, mission.id);
|
|
9439
9459
|
const errStr = stderr.slice(-2000);
|
|
9440
9460
|
const authExpired = !ok && missionRunnerAuthExpired(errStr, finalText, stdoutBuf);
|
|
9441
9461
|
resolve({
|
package/lib/engine-registry.js
CHANGED
|
@@ -15,6 +15,21 @@ const {
|
|
|
15
15
|
engineTakesModel,
|
|
16
16
|
engineEffortLevels,
|
|
17
17
|
} = require('./runner-command');
|
|
18
|
+
// The one model-name table, shared with runner-command: friendly spellings
|
|
19
|
+
// to the ids each cli accepts, plus retired-model upgrades.
|
|
20
|
+
const {
|
|
21
|
+
CLAUDE_FAMILY_ENGINES,
|
|
22
|
+
CLAUDE_MODEL_ALIASES,
|
|
23
|
+
RETIRED_CLAUDE_MODELS,
|
|
24
|
+
CLAUDE_MODEL_SPELLING,
|
|
25
|
+
CLAUDE_LONG_CONTEXT,
|
|
26
|
+
GROK_MODEL_SPELLING,
|
|
27
|
+
currentClaudeModel,
|
|
28
|
+
claudeModelId,
|
|
29
|
+
retiredModelsNamed,
|
|
30
|
+
normalizeRosterModel,
|
|
31
|
+
savedRosterModel,
|
|
32
|
+
} = require('./model-names');
|
|
18
33
|
// Through the module object so tests can see when ranking runs.
|
|
19
34
|
const routerBrain = require('./router-brain');
|
|
20
35
|
const { routerPickExplanation } = routerBrain;
|
|
@@ -584,102 +599,6 @@ function rosterPickExpired(pick, now = new Date()) {
|
|
|
584
599
|
return localDay(rosterDate(now)).getTime() > until.getTime();
|
|
585
600
|
}
|
|
586
601
|
|
|
587
|
-
// Claude-family engines run through the claude CLI, which rejects loose
|
|
588
|
-
// spellings like "opus-5.5". Assign saves the id the CLI accepts, and every
|
|
589
|
-
// read normalizes again, so older saved picks and hand edits work too.
|
|
590
|
-
const CLAUDE_FAMILY_ENGINES = Object.freeze(['claude', 'fable', 'haiku']);
|
|
591
|
-
const CLAUDE_MODEL_ALIASES = Object.freeze(['opus', 'sonnet', 'haiku', 'fable', 'opusplan', 'default']);
|
|
592
|
-
// A retired model runs as its replacement, so a saved pick keeps working
|
|
593
|
-
// after the upgrade. Retiring the next one is one line here.
|
|
594
|
-
const RETIRED_CLAUDE_MODELS = Object.freeze({
|
|
595
|
-
'claude-fable-5': 'claude-fable-5-1',
|
|
596
|
-
'claude-haiku-4-5': 'claude-haiku-5-5',
|
|
597
|
-
'claude-haiku-4-5-20251001': 'claude-haiku-5-5',
|
|
598
|
-
});
|
|
599
|
-
const CLAUDE_MODEL_SPELLING = /^(opus|sonnet|haiku|fable)[\s_-]*(\d+)(?:[\s._-]+(\d+))?$/;
|
|
600
|
-
const CLAUDE_LONG_CONTEXT = /\s*\[1m\]$/;
|
|
601
|
-
|
|
602
|
-
// Grok takes friendly names too: "grok 4.7 fast" is grok-4.7-build-fast and
|
|
603
|
-
// "grok 4.7" is grok-4.7. Any full grok- id passes through as typed.
|
|
604
|
-
const GROK_MODEL_SPELLING = /^grok[\s_-]*(\d+(?:\.\d+)?)(?:[\s_-]+(?:build[\s_-]+)?(fast))?$/;
|
|
605
|
-
|
|
606
|
-
function normalizeGrokModel(raw) {
|
|
607
|
-
const match = GROK_MODEL_SPELLING.exec(raw.toLowerCase());
|
|
608
|
-
if (match) return `grok-${match[1]}${match[2] ? '-build-fast' : ''}`;
|
|
609
|
-
if (/^grok-[a-z0-9][a-z0-9._-]*$/i.test(raw)) return raw;
|
|
610
|
-
throw new Error(`grok does not know the model "${raw}". use grok 4.7 fast, grok 4.7, or a full grok- id`);
|
|
611
|
-
}
|
|
612
|
-
|
|
613
|
-
// The model to run for a claude id: its replacement when it is retired,
|
|
614
|
-
// else the id as given. Keeps a "[1m]" tail.
|
|
615
|
-
function currentClaudeModel(value) {
|
|
616
|
-
const raw = String(value || '').trim();
|
|
617
|
-
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
618
|
-
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
619
|
-
const next = RETIRED_CLAUDE_MODELS[base.toLowerCase()];
|
|
620
|
-
return next ? `${next}${suffix}` : raw;
|
|
621
|
-
}
|
|
622
|
-
|
|
623
|
-
// The id the claude cli accepts for a display name typed with a space:
|
|
624
|
-
// "opus 5.5" is claude-opus-5-5, "fable 5.1" is claude-fable-5-1. A full id,
|
|
625
|
-
// a bare alias ("opus"), a "[1m]" tail and anything else pass through as given
|
|
626
|
-
// (a retired id still becomes its replacement).
|
|
627
|
-
const CLAUDE_DISPLAY_NAME = /^(opus|sonnet|haiku|fable)\s+(\d+)(?:\.(\d+))?$/i;
|
|
628
|
-
|
|
629
|
-
function claudeModelId(value) {
|
|
630
|
-
const raw = String(value || '').trim();
|
|
631
|
-
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
632
|
-
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
633
|
-
const match = CLAUDE_DISPLAY_NAME.exec(base);
|
|
634
|
-
if (!match) return currentClaudeModel(raw);
|
|
635
|
-
return currentClaudeModel(`claude-${match[1].toLowerCase()}-${match[2]}${match[3] ? `-${match[3]}` : ''}${suffix}`);
|
|
636
|
-
}
|
|
637
|
-
|
|
638
|
-
// The retired models a roster file's text still names, by id or spelled
|
|
639
|
-
// out, each once, so the roster can say they were upgraded.
|
|
640
|
-
function retiredModelsNamed(text) {
|
|
641
|
-
const found = [];
|
|
642
|
-
for (const id of Object.keys(RETIRED_CLAUDE_MODELS)) {
|
|
643
|
-
const [, family, major, minor] = /^claude-([a-z]+)-(\d+)-(\d+)/.exec(id) || [];
|
|
644
|
-
if (!family) continue;
|
|
645
|
-
const spelled = new RegExp(`\\b(?:claude[\\s_-]*)?${family}[\\s_-]*${major}[\\s._-]+${minor}(?![\\d.])`, 'i');
|
|
646
|
-
const base = `claude-${family}-${major}-${minor}`;
|
|
647
|
-
if (spelled.test(String(text || '')) && !found.includes(base)) found.push(base);
|
|
648
|
-
}
|
|
649
|
-
return found;
|
|
650
|
-
}
|
|
651
|
-
|
|
652
|
-
function normalizeRosterModel(engineId, value) {
|
|
653
|
-
const raw = String(value || '').trim();
|
|
654
|
-
if (!raw) return '';
|
|
655
|
-
if (engineId === 'grok') return normalizeGrokModel(raw);
|
|
656
|
-
if (!CLAUDE_FAMILY_ENGINES.includes(engineId)) return raw;
|
|
657
|
-
// "[1m]" asks the claude cli for the long context window; keep it on
|
|
658
|
-
// whatever name it follows.
|
|
659
|
-
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
660
|
-
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
661
|
-
const lower = base.toLowerCase();
|
|
662
|
-
if (CLAUDE_MODEL_ALIASES.includes(lower)) return `${lower}${suffix}`;
|
|
663
|
-
const match = CLAUDE_MODEL_SPELLING.exec(lower.replace(/^claude[\s_-]*/, ''));
|
|
664
|
-
if (match) {
|
|
665
|
-
const id = `claude-${match[1]}-${match[2]}${match[3] ? `-${match[3]}` : ''}`;
|
|
666
|
-
return currentClaudeModel(`${id}${suffix}`);
|
|
667
|
-
}
|
|
668
|
-
if (lower.startsWith('claude-')) return currentClaudeModel(`${base}${suffix}`);
|
|
669
|
-
throw new Error(`${engineId} does not know the model "${raw}". use opus, sonnet, haiku, fable, opusplan, default, a version like opus 5.5, sonnet 5, haiku 5.5, or fable 5.1, or a full claude- id; add [1m] for the long context window`);
|
|
670
|
-
}
|
|
671
|
-
|
|
672
|
-
// A saved pick's model, normalized the same way assign does. null means the
|
|
673
|
-
// saved name is one the claude cli would reject, so the pick cannot run.
|
|
674
|
-
function savedRosterModel(engineId, value) {
|
|
675
|
-
try {
|
|
676
|
-
return normalizeRosterModel(engineId, value);
|
|
677
|
-
} catch {
|
|
678
|
-
return null;
|
|
679
|
-
}
|
|
680
|
-
}
|
|
681
|
-
|
|
682
|
-
|
|
683
602
|
// A registry saved before an engine learned a job keeps its old role list;
|
|
684
603
|
// the roster still honors the job the engine can do today, including a job
|
|
685
604
|
// it only takes by roster pick.
|
|
@@ -0,0 +1,82 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
// Live progress for a working member: one plain line per tool step, printed to
|
|
4
|
+
// stderr and appended to <dir>/<member>.log so another shell can tail it.
|
|
5
|
+
|
|
6
|
+
const fs = require('fs');
|
|
7
|
+
const os = require('os');
|
|
8
|
+
const path = require('path');
|
|
9
|
+
|
|
10
|
+
function realish(p) {
|
|
11
|
+
try { return fs.realpathSync(p); } catch {}
|
|
12
|
+
try { return path.join(fs.realpathSync(path.dirname(p)), path.basename(p)); } catch {}
|
|
13
|
+
return p;
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
function insideRel(cwd, p) {
|
|
17
|
+
const rel = path.relative(cwd, p);
|
|
18
|
+
return rel && !rel.startsWith('..') && !path.isAbsolute(rel) ? rel : null;
|
|
19
|
+
}
|
|
20
|
+
|
|
21
|
+
// Relative to cwd when inside it; symlinked roots (macOS /var vs /private/var)
|
|
22
|
+
// still count as inside.
|
|
23
|
+
function relativeTo(cwd, file) {
|
|
24
|
+
const p = String(file || '');
|
|
25
|
+
if (!p || !cwd || !path.isAbsolute(p)) return p;
|
|
26
|
+
return insideRel(cwd, p) || insideRel(realish(cwd), realish(p)) || p;
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
function formatToolStep(block, cwd) {
|
|
30
|
+
if (!block || block.type !== 'tool_use') return null;
|
|
31
|
+
const name = String(block.name || 'tool');
|
|
32
|
+
const input = block.input || {};
|
|
33
|
+
switch (name) {
|
|
34
|
+
case 'Read':
|
|
35
|
+
return `reading ${relativeTo(cwd, input.file_path)}`;
|
|
36
|
+
case 'Edit':
|
|
37
|
+
case 'MultiEdit':
|
|
38
|
+
case 'Write':
|
|
39
|
+
return `editing ${relativeTo(cwd, input.file_path)}`;
|
|
40
|
+
case 'Bash': {
|
|
41
|
+
const first = String(input.command || '').split('\n')[0].trim();
|
|
42
|
+
return `running: ${first.slice(0, 70)}`;
|
|
43
|
+
}
|
|
44
|
+
case 'Grep':
|
|
45
|
+
return `searching for ${input.pattern || ''}`;
|
|
46
|
+
case 'Glob':
|
|
47
|
+
return `finding ${input.pattern || ''}`;
|
|
48
|
+
default:
|
|
49
|
+
return name.toLowerCase();
|
|
50
|
+
}
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
function progressDir() {
|
|
54
|
+
return process.env.ATRIS_PROGRESS_DIR || path.join(os.homedir(), '.atris', 'progress');
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
function safeMember(member) {
|
|
58
|
+
return String(member || 'mission').replace(/[^A-Za-z0-9._-]/g, '_') || 'mission';
|
|
59
|
+
}
|
|
60
|
+
|
|
61
|
+
function hhmm(date = new Date()) {
|
|
62
|
+
return `${String(date.getHours()).padStart(2, '0')}:${String(date.getMinutes()).padStart(2, '0')}`;
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
function appendProgress(member, line, missionId) {
|
|
66
|
+
// Test runs never write into the real home folder unless they name a dir.
|
|
67
|
+
if (!process.env.ATRIS_PROGRESS_DIR && process.env.NODE_TEST_CONTEXT) return;
|
|
68
|
+
try {
|
|
69
|
+
const dir = progressDir();
|
|
70
|
+
fs.mkdirSync(dir, { recursive: true });
|
|
71
|
+
const entry = `${new Date().toISOString()} ${missionId || '-'} ${line}\n`;
|
|
72
|
+
fs.appendFileSync(path.join(dir, `${safeMember(member)}.log`), entry, 'utf8');
|
|
73
|
+
} catch {}
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
// Print one step for the operator and keep it in the member's progress log.
|
|
77
|
+
function reportProgress(member, line, missionId) {
|
|
78
|
+
try { process.stderr.write(` ${member || 'mission'} ${hhmm()} ${line}\n`); } catch {}
|
|
79
|
+
appendProgress(member, line, missionId);
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
module.exports = { formatToolStep, appendProgress, reportProgress };
|
|
@@ -0,0 +1,116 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
// Friendly model names to the ids each engine's cli accepts. This is the ONE
|
|
4
|
+
// table of spellings: engine-registry checks roster picks with it and
|
|
5
|
+
// runner-command normalizes launch models with it. "opus 5.5" is
|
|
6
|
+
// claude-opus-5-5, "grok 4.7 fast" is grok-4.7-build-fast. Kept free of
|
|
7
|
+
// requires so both files can load it without a circular dependency.
|
|
8
|
+
|
|
9
|
+
// Claude-family engines run through the claude CLI, which rejects loose
|
|
10
|
+
// spellings like "opus-5.5". Assign saves the id the CLI accepts, and every
|
|
11
|
+
// read normalizes again, so older saved picks and hand edits work too.
|
|
12
|
+
const CLAUDE_FAMILY_ENGINES = Object.freeze(['claude', 'fable', 'haiku']);
|
|
13
|
+
const CLAUDE_MODEL_ALIASES = Object.freeze(['opus', 'sonnet', 'haiku', 'fable', 'opusplan', 'default']);
|
|
14
|
+
// A retired model runs as its replacement, so a saved pick keeps working
|
|
15
|
+
// after the upgrade. Retiring the next one is one line here.
|
|
16
|
+
const RETIRED_CLAUDE_MODELS = Object.freeze({
|
|
17
|
+
'claude-fable-5': 'claude-fable-5-1',
|
|
18
|
+
'claude-haiku-4-5': 'claude-haiku-5-5',
|
|
19
|
+
'claude-haiku-4-5-20251001': 'claude-haiku-5-5',
|
|
20
|
+
});
|
|
21
|
+
const CLAUDE_MODEL_SPELLING = /^(opus|sonnet|haiku|fable)[\s_-]*(\d+)(?:[\s._-]+(\d+))?$/;
|
|
22
|
+
const CLAUDE_LONG_CONTEXT = /\s*\[1m\]$/;
|
|
23
|
+
|
|
24
|
+
// Grok takes friendly names too: "grok 4.7 fast" is grok-4.7-build-fast and
|
|
25
|
+
// "grok 4.7" is grok-4.7. Any full grok- id passes through as typed.
|
|
26
|
+
const GROK_MODEL_SPELLING = /^grok[\s_-]*(\d+(?:\.\d+)?)(?:[\s_-]+(?:build[\s_-]+)?(fast))?$/;
|
|
27
|
+
|
|
28
|
+
function normalizeGrokModel(raw) {
|
|
29
|
+
const match = GROK_MODEL_SPELLING.exec(raw.toLowerCase());
|
|
30
|
+
if (match) return `grok-${match[1]}${match[2] ? '-build-fast' : ''}`;
|
|
31
|
+
if (/^grok-[a-z0-9][a-z0-9._-]*$/i.test(raw)) return raw;
|
|
32
|
+
throw new Error(`grok does not know the model "${raw}". use grok 4.7 fast, grok 4.7, or a full grok- id`);
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
// The model to run for a claude id: its replacement when it is retired,
|
|
36
|
+
// else the id as given. Keeps a "[1m]" tail.
|
|
37
|
+
function currentClaudeModel(value) {
|
|
38
|
+
const raw = String(value || '').trim();
|
|
39
|
+
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
40
|
+
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
41
|
+
const next = RETIRED_CLAUDE_MODELS[base.toLowerCase()];
|
|
42
|
+
return next ? `${next}${suffix}` : raw;
|
|
43
|
+
}
|
|
44
|
+
|
|
45
|
+
// The id the claude cli accepts for a display name typed with a space:
|
|
46
|
+
// "opus 5.5" is claude-opus-5-5, "fable 5.1" is claude-fable-5-1. A full id,
|
|
47
|
+
// a bare alias ("opus"), a "[1m]" tail and anything else pass through as given
|
|
48
|
+
// (a retired id still becomes its replacement).
|
|
49
|
+
const CLAUDE_DISPLAY_NAME = /^(opus|sonnet|haiku|fable)\s+(\d+)(?:\.(\d+))?$/i;
|
|
50
|
+
|
|
51
|
+
function claudeModelId(value) {
|
|
52
|
+
const raw = String(value || '').trim();
|
|
53
|
+
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
54
|
+
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
55
|
+
const match = CLAUDE_DISPLAY_NAME.exec(base);
|
|
56
|
+
if (!match) return currentClaudeModel(raw);
|
|
57
|
+
return currentClaudeModel(`claude-${match[1].toLowerCase()}-${match[2]}${match[3] ? `-${match[3]}` : ''}${suffix}`);
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
// The retired models a roster file's text still names, by id or spelled
|
|
61
|
+
// out, each once, so the roster can say they were upgraded.
|
|
62
|
+
function retiredModelsNamed(text) {
|
|
63
|
+
const found = [];
|
|
64
|
+
for (const id of Object.keys(RETIRED_CLAUDE_MODELS)) {
|
|
65
|
+
const [, family, major, minor] = /^claude-([a-z]+)-(\d+)-(\d+)/.exec(id) || [];
|
|
66
|
+
if (!family) continue;
|
|
67
|
+
const spelled = new RegExp(`\\b(?:claude[\\s_-]*)?${family}[\\s_-]*${major}[\\s._-]+${minor}(?![\\d.])`, 'i');
|
|
68
|
+
const base = `claude-${family}-${major}-${minor}`;
|
|
69
|
+
if (spelled.test(String(text || '')) && !found.includes(base)) found.push(base);
|
|
70
|
+
}
|
|
71
|
+
return found;
|
|
72
|
+
}
|
|
73
|
+
|
|
74
|
+
function normalizeRosterModel(engineId, value) {
|
|
75
|
+
const raw = String(value || '').trim();
|
|
76
|
+
if (!raw) return '';
|
|
77
|
+
if (engineId === 'grok') return normalizeGrokModel(raw);
|
|
78
|
+
if (!CLAUDE_FAMILY_ENGINES.includes(engineId)) return raw;
|
|
79
|
+
// "[1m]" asks the claude cli for the long context window; keep it on
|
|
80
|
+
// whatever name it follows.
|
|
81
|
+
const suffix = CLAUDE_LONG_CONTEXT.test(raw) ? '[1m]' : '';
|
|
82
|
+
const base = raw.replace(CLAUDE_LONG_CONTEXT, '').trim();
|
|
83
|
+
const lower = base.toLowerCase();
|
|
84
|
+
if (CLAUDE_MODEL_ALIASES.includes(lower)) return `${lower}${suffix}`;
|
|
85
|
+
const match = CLAUDE_MODEL_SPELLING.exec(lower.replace(/^claude[\s_-]*/, ''));
|
|
86
|
+
if (match) {
|
|
87
|
+
const id = `claude-${match[1]}-${match[2]}${match[3] ? `-${match[3]}` : ''}`;
|
|
88
|
+
return currentClaudeModel(`${id}${suffix}`);
|
|
89
|
+
}
|
|
90
|
+
if (lower.startsWith('claude-')) return currentClaudeModel(`${base}${suffix}`);
|
|
91
|
+
throw new Error(`${engineId} does not know the model "${raw}". use opus, sonnet, haiku, fable, opusplan, default, a version like opus 5.5, sonnet 5, haiku 5.5, or fable 5.1, or a full claude- id; add [1m] for the long context window`);
|
|
92
|
+
}
|
|
93
|
+
|
|
94
|
+
// A saved pick's model, normalized the same way assign does. null means the
|
|
95
|
+
// saved name is one the claude cli would reject, so the pick cannot run.
|
|
96
|
+
function savedRosterModel(engineId, value) {
|
|
97
|
+
try {
|
|
98
|
+
return normalizeRosterModel(engineId, value);
|
|
99
|
+
} catch {
|
|
100
|
+
return null;
|
|
101
|
+
}
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
module.exports = {
|
|
105
|
+
CLAUDE_FAMILY_ENGINES,
|
|
106
|
+
CLAUDE_MODEL_ALIASES,
|
|
107
|
+
RETIRED_CLAUDE_MODELS,
|
|
108
|
+
CLAUDE_MODEL_SPELLING,
|
|
109
|
+
CLAUDE_LONG_CONTEXT,
|
|
110
|
+
GROK_MODEL_SPELLING,
|
|
111
|
+
currentClaudeModel,
|
|
112
|
+
claudeModelId,
|
|
113
|
+
retiredModelsNamed,
|
|
114
|
+
normalizeRosterModel,
|
|
115
|
+
savedRosterModel,
|
|
116
|
+
};
|
package/lib/runner-command.js
CHANGED
|
@@ -16,19 +16,20 @@ const DEFAULT_CLAUDE_RUNNER_BIN = 'claude';
|
|
|
16
16
|
const EFFORT_WORDS = Object.freeze(['low', 'medium', 'high', 'xhigh', 'max']);
|
|
17
17
|
const CLAUDE_EFFORTS = EFFORT_WORDS;
|
|
18
18
|
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
'opus 4.8': 'claude-opus-4-8',
|
|
25
|
-
'opus-4.8': 'claude-opus-4-8',
|
|
26
|
-
});
|
|
19
|
+
// The one model-name table lives in model-names.js, shared with
|
|
20
|
+
// engine-registry. Friendly names become the ids the claude cli accepts;
|
|
21
|
+
// anything it does not know passes through unchanged (runners like glm and
|
|
22
|
+
// swe name their own models here, so this never throws).
|
|
23
|
+
const { claudeModelId, currentClaudeModel, normalizeRosterModel } = require('./model-names');
|
|
27
24
|
|
|
28
25
|
function normalizeClaudeModelName(value) {
|
|
29
26
|
const trimmed = String(value || '').trim();
|
|
30
27
|
if (!trimmed) return '';
|
|
31
|
-
|
|
28
|
+
try {
|
|
29
|
+
return normalizeRosterModel('claude', trimmed);
|
|
30
|
+
} catch {
|
|
31
|
+
return trimmed;
|
|
32
|
+
}
|
|
32
33
|
}
|
|
33
34
|
|
|
34
35
|
// Canonical runner profiles. Each entry is the config an operator would pick
|
|
@@ -246,7 +247,7 @@ function ownModelsOnlyEngineForBin(bin) {
|
|
|
246
247
|
// claude env pin) and rides its own default instead.
|
|
247
248
|
function runnerModelFor(engineId, model) {
|
|
248
249
|
// Required here, not at the top: engine-registry requires this file.
|
|
249
|
-
const { isClaudeFamilyModel
|
|
250
|
+
const { isClaudeFamilyModel } = require('./engine-registry');
|
|
250
251
|
const name = String(engineId || '').trim().toLowerCase();
|
|
251
252
|
// A display name like "opus 5.5" only becomes an id for the claude cli (or
|
|
252
253
|
// an unnamed engine). Any other engine's model name is handed over as is.
|