pantheon-opencode 1.6.3-beta.11 β 1.6.3-beta.13
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/CHANGELOG.md +23 -6
- package/bin/pantheon-init.mjs +24 -1
- package/package.json +1 -1
- package/plugin.json +1 -1
- package/pyproject.toml +1 -1
- package/scripts/install/cli-ui.mjs +2 -1
- package/scripts/install/opencode.mjs +26 -10
- package/src/pantheon/jev-decision-tool.ts +60 -11
- package/src/pantheon/jev-decisions.ts +68 -5
- package/src/pantheon/jev-quality-score.ts +1 -1
- package/src/pantheon/jev-rank-verify.ts +22 -7
- package/src/pantheon/local-decision-router.ts +10 -0
- package/src/pantheon/pantheon-decision-catalog.ts +1 -1
- package/src/plugins/tui/dist/tui-v2.js +125 -22
- package/src/plugins/tui/dist/tui-v2.js.map +1 -1
- package/src/plugins/tui/package-lock.json +2 -2
- package/src/plugins/tui/package.json +1 -1
- package/src/plugins/tui/src/whats-new-highlights.ts +59 -5
- package/src/plugins/tui/src/whats-new-open.ts +41 -1
- package/src/plugins/tui/src/whats-new.tsx +61 -8
package/CHANGELOG.md
CHANGED
|
@@ -10,14 +10,31 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
|
|
|
10
10
|
<!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
|
|
11
11
|
move this section to a versioned entry and reset the template below. -->
|
|
12
12
|
|
|
13
|
-
## [v1.6.3-beta.
|
|
13
|
+
## [v1.6.3-beta.13] - 2026-11-05
|
|
14
|
+
|
|
15
|
+
## π What's New
|
|
16
|
+
- One coherent setup summary: a single closing block instead of two competing "Next steps" lists.
|
|
17
|
+
- Warnings restate their text inline β no more hunting for what scrolled past.
|
|
18
|
+
- The worktree guidance box is project-only (a global install from $HOME showed a nonsense path).
|
|
19
|
+
- The banner now prints the exact `source` line for the decision-layer flags file β the silent-off gap fixed.
|
|
20
|
+
|
|
21
|
+
## [v1.6.3-beta.12] - 2026-11-05
|
|
14
22
|
|
|
15
23
|
## π What's New
|
|
16
|
-
-
|
|
17
|
-
- The
|
|
18
|
-
-
|
|
19
|
-
-
|
|
20
|
-
-
|
|
24
|
+
- Delegation answers are domain-aware again: agent descriptions ride with every choice β live probe 5/5 exact (FastAPIβhermes, WCAGβaphrodite, Alembicβdemeter, securityβthemis, poemβnone).
|
|
25
|
+
- The input envelope is enforced on EVERY path now: presets, review, rank and verify all refuse missing/unknown keys with the exact cause.
|
|
26
|
+
- Rank accepts up to 20 candidates (typesafe_rank contract) and names the accepted range on refusal.
|
|
27
|
+
- Skills results report where the candidates came from (`candidateSource`), ending silent prefilter misses.
|
|
28
|
+
- The local router respects open gates: review/approval required β abstain (`GATE_OPEN`), never route around.
|
|
29
|
+
|
|
30
|
+
## [v1.6.3-beta.11] - 2026-11-05
|
|
31
|
+
|
|
32
|
+
### π What's New
|
|
33
|
+
- Measured MCP usage policy across the roster β governance ladder byte-identical in all 13 specialists.
|
|
34
|
+
- Skill-discovery pipeline with the trust cascade: Anthropic catalog β skills.sh β GitHub; Jev ranks, you install.
|
|
35
|
+
- Dead-route holds release (ledger schema v4) β the fix for budgets frozen by removed routes.
|
|
36
|
+
- Roster reform: nyx re-scoped to chaos & adversarial, talos aligned to hotfix, domain docs per agent.
|
|
37
|
+
- Jev validated end to end in the field: N4 dead, verify field-verified, sandbox 28/28.
|
|
21
38
|
|
|
22
39
|
### Added
|
|
23
40
|
|
package/bin/pantheon-init.mjs
CHANGED
|
@@ -1064,10 +1064,33 @@ async function main() {
|
|
|
1064
1064
|
console.log(` ${S.nextVerify}`)
|
|
1065
1065
|
console.log(` ${S.nextLaunch}`)
|
|
1066
1066
|
console.log(` ${S.nextAgents}`)
|
|
1067
|
-
|
|
1067
|
+
if (!isProject) {
|
|
1068
|
+
// The project-local hint only matters when this install was not one.
|
|
1069
|
+
console.log(` ${S.nextProject}`)
|
|
1070
|
+
}
|
|
1068
1071
|
if (!decisionTarget) {
|
|
1069
1072
|
console.log(` ${S.nextJev}`)
|
|
1070
1073
|
}
|
|
1074
|
+
// The decision-layer extras (J1/J3/specialist callers) write an env file;
|
|
1075
|
+
// without sourcing it the gates stay off silently β the exact gap the
|
|
1076
|
+
// field hit on the beta.12 setup pass.
|
|
1077
|
+
try {
|
|
1078
|
+
const { existsSync } = await import('node:fs')
|
|
1079
|
+
const flagsPath = path.join(
|
|
1080
|
+
process.env.XDG_CONFIG_HOME ?? path.join(process.env.HOME ?? '', '.config'),
|
|
1081
|
+
'pantheon',
|
|
1082
|
+
'decision-flags.env',
|
|
1083
|
+
)
|
|
1084
|
+
if (existsSync(flagsPath)) {
|
|
1085
|
+
console.log('')
|
|
1086
|
+
console.log(' Decision-layer flags (J1/J3/specialist callers):')
|
|
1087
|
+
console.log(
|
|
1088
|
+
` source ${flagsPath} # add to your shell profile; applies to NEW opencode sessions`,
|
|
1089
|
+
)
|
|
1090
|
+
}
|
|
1091
|
+
} catch {
|
|
1092
|
+
// Hint is best-effort; never fail the install over it.
|
|
1093
|
+
}
|
|
1071
1094
|
console.log('')
|
|
1072
1095
|
|
|
1073
1096
|
return
|
package/package.json
CHANGED
package/plugin.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pantheon",
|
|
3
3
|
"publisher": "ils15",
|
|
4
|
-
"version": "1.6.3-beta.
|
|
4
|
+
"version": "1.6.3-beta.13",
|
|
5
5
|
"displayName": "Pantheon",
|
|
6
6
|
"description": "Conductor-Delegate orchestration platform with 14 specialized agents, 20 skills, and OpenCode support.",
|
|
7
7
|
"agents": "./src/agents",
|
package/pyproject.toml
CHANGED
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.backends._legacy:_Backend"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "pantheon"
|
|
7
|
-
version = "1.6.3-beta.
|
|
7
|
+
version = "1.6.3-beta.13"
|
|
8
8
|
description = "Multi-agent orchestration framework"
|
|
9
9
|
requires-python = ">=3.11"
|
|
10
10
|
# Python sources under src/mcp/ import fastmcp/pydantic at runtime, but those
|
|
@@ -386,8 +386,9 @@ export function printWorktreeGuidance({
|
|
|
386
386
|
projectRoot = process.cwd(),
|
|
387
387
|
interactive = false,
|
|
388
388
|
dryRun = false,
|
|
389
|
+
enabled = true,
|
|
389
390
|
} = {}) {
|
|
390
|
-
if (!interactive || dryRun || !tty()) return
|
|
391
|
+
if (!interactive || dryRun || !tty() || enabled === false) return
|
|
391
392
|
|
|
392
393
|
const recommendation = getWorktreeRecommendation(projectRoot)
|
|
393
394
|
const rows = [
|
|
@@ -1427,6 +1427,8 @@ export function setupRuntimePhase(
|
|
|
1427
1427
|
error(S.runtimeFailed(err.message))
|
|
1428
1428
|
warning(S.runtimeSkippedMcp)
|
|
1429
1429
|
stats.warnings += 1
|
|
1430
|
+
if (!Array.isArray(stats.warningMessages)) stats.warningMessages = []
|
|
1431
|
+
stats.warningMessages.push(S.runtimeSkippedMcp)
|
|
1430
1432
|
stats.errors += 1
|
|
1431
1433
|
return false
|
|
1432
1434
|
}
|
|
@@ -2406,6 +2408,10 @@ async function installPass(
|
|
|
2406
2408
|
// payload and the host config are already in place.
|
|
2407
2409
|
warning(`Could not disable the builtin Context sidebar block: ${cause.message}`)
|
|
2408
2410
|
stats.warnings++
|
|
2411
|
+
if (!Array.isArray(stats.warningMessages)) stats.warningMessages = []
|
|
2412
|
+
stats.warningMessages.push(
|
|
2413
|
+
`Could not disable the builtin Context sidebar block: ${cause.message}`,
|
|
2414
|
+
)
|
|
2409
2415
|
}
|
|
2410
2416
|
}
|
|
2411
2417
|
if (componentSet.has('plugins')) advancePhase()
|
|
@@ -3091,6 +3097,8 @@ async function installPass(
|
|
|
3091
3097
|
for (const message of mcp.warnings ?? []) {
|
|
3092
3098
|
warning(message)
|
|
3093
3099
|
stats.warnings += 1
|
|
3100
|
+
if (!Array.isArray(stats.warningMessages)) stats.warningMessages = []
|
|
3101
|
+
stats.warningMessages.push(message)
|
|
3094
3102
|
}
|
|
3095
3103
|
}
|
|
3096
3104
|
|
|
@@ -3421,19 +3429,27 @@ async function installPass(
|
|
|
3421
3429
|
process.stdout.write(` [OK] ${created} component(s) installed\n`)
|
|
3422
3430
|
if (skipped > 0)
|
|
3423
3431
|
process.stdout.write(` ${colors.dim('\u2014')} ${skipped} already up-to-date\n`)
|
|
3424
|
-
|
|
3432
|
+
// The warning TEXT restates here β the count alone sent users to doctor to
|
|
3433
|
+
// learn what had already scrolled past.
|
|
3434
|
+
for (const message of stats.warningMessages ?? []) {
|
|
3435
|
+
process.stdout.write(` [WARN] ${message}\n`)
|
|
3436
|
+
}
|
|
3437
|
+
if (warnings > 0 && (stats.warningMessages ?? []).length === 0) {
|
|
3438
|
+
process.stdout.write(` [WARN] ${warnings} warning(s)\n`)
|
|
3439
|
+
}
|
|
3440
|
+
if (warnings > 0) {
|
|
3425
3441
|
process.stdout.write(
|
|
3426
|
-
`
|
|
3442
|
+
` ${colors.dim("Run 'pantheon-opencode doctor' for the full health check")}\n`,
|
|
3427
3443
|
)
|
|
3444
|
+
}
|
|
3428
3445
|
process.stdout.write('\n')
|
|
3429
|
-
|
|
3430
|
-
|
|
3431
|
-
|
|
3432
|
-
|
|
3433
|
-
|
|
3434
|
-
|
|
3435
|
-
|
|
3436
|
-
printWorktreeGuidance({ projectRoot: process.cwd(), interactive: true })
|
|
3446
|
+
// Worktree guidance is a project-install concept: a global install run
|
|
3447
|
+
// from $HOME derived a nonsense `<home>/.pantheon-worktrees/<user>` path.
|
|
3448
|
+
printWorktreeGuidance({
|
|
3449
|
+
projectRoot: process.cwd(),
|
|
3450
|
+
interactive: true,
|
|
3451
|
+
enabled: !(opts.global ?? isGlobalConfigDir(target)),
|
|
3452
|
+
})
|
|
3437
3453
|
}
|
|
3438
3454
|
|
|
3439
3455
|
// Beta 11 Setup phase 3: report whether this call APPLIED, so the caller can
|
|
@@ -24,8 +24,11 @@
|
|
|
24
24
|
*/
|
|
25
25
|
import { readUserDecisionRouterConfig } from '../../scripts/install/decision-router-config.mjs'
|
|
26
26
|
import {
|
|
27
|
+
checkJevEnvelope,
|
|
27
28
|
consumeJevDecisionResult,
|
|
28
29
|
isJevDecisionPreset,
|
|
30
|
+
JEV_EXPLICIT_ENVELOPE_KEYS,
|
|
31
|
+
JEV_PRESET_ENVELOPE_KEYS,
|
|
29
32
|
type JevDecisionPreparation,
|
|
30
33
|
type JevHostSkill,
|
|
31
34
|
type NativeJevRouteFailure,
|
|
@@ -41,12 +44,15 @@ import {
|
|
|
41
44
|
J4_QUALITY_SCORE_POINT,
|
|
42
45
|
type QualityScoreEvaluation,
|
|
43
46
|
qualityVerdict,
|
|
47
|
+
REVIEW_ENVELOPE_KEYS,
|
|
44
48
|
} from './jev-quality-score.ts'
|
|
45
49
|
import {
|
|
46
50
|
buildJevRankRequest,
|
|
47
51
|
buildJevVerifyRequest,
|
|
48
52
|
consumeJevRankResult,
|
|
49
53
|
consumeJevVerifyResult,
|
|
54
|
+
RANK_ENVELOPE_KEYS,
|
|
55
|
+
VERIFY_ENVELOPE_KEYS,
|
|
50
56
|
} from './jev-rank-verify.ts'
|
|
51
57
|
import { mapGateFailureToSkipReason } from './jev-risk-gate.ts'
|
|
52
58
|
import {
|
|
@@ -152,8 +158,12 @@ interface JevToolContext {
|
|
|
152
158
|
agent?: unknown
|
|
153
159
|
}
|
|
154
160
|
|
|
155
|
-
function refusal(reasonCode: JevDecisionRunFailure): string {
|
|
156
|
-
return JSON.stringify(
|
|
161
|
+
function refusal(reasonCode: JevDecisionRunFailure, detail?: string): string {
|
|
162
|
+
return JSON.stringify(
|
|
163
|
+
detail === undefined
|
|
164
|
+
? { advisory: false, effect: 'none', reasonCode }
|
|
165
|
+
: { advisory: false, effect: 'none', reasonCode, detail },
|
|
166
|
+
)
|
|
157
167
|
}
|
|
158
168
|
|
|
159
169
|
function isNonBlankString(value: unknown): value is string {
|
|
@@ -453,12 +463,46 @@ export async function runJevDecision(
|
|
|
453
463
|
if (caller !== 'zeus' && requested.mode !== undefined) {
|
|
454
464
|
return refusal('CALLER_NOT_ALLOWED')
|
|
455
465
|
}
|
|
466
|
+
// The ONE envelope gate every path passes through before its builder:
|
|
467
|
+
// schemaVersion required, unknown keys rejected β explicit questions,
|
|
468
|
+
// presets, review, rank and verify alike (beta.10 claimed this contract;
|
|
469
|
+
// beta.12 applies it uniformly). Builders keep their semantic checks.
|
|
470
|
+
const envelopeKeys =
|
|
471
|
+
requested.mode === undefined
|
|
472
|
+
? JEV_EXPLICIT_ENVELOPE_KEYS
|
|
473
|
+
: requested.mode === 'review'
|
|
474
|
+
? REVIEW_ENVELOPE_KEYS
|
|
475
|
+
: requested.mode === 'rank'
|
|
476
|
+
? RANK_ENVELOPE_KEYS
|
|
477
|
+
: requested.mode === 'verify'
|
|
478
|
+
? VERIFY_ENVELOPE_KEYS
|
|
479
|
+
: isJevDecisionPreset(requested.mode)
|
|
480
|
+
? JEV_PRESET_ENVELOPE_KEYS
|
|
481
|
+
: undefined
|
|
482
|
+
if (envelopeKeys !== undefined) {
|
|
483
|
+
const envelope = checkJevEnvelope(input, envelopeKeys)
|
|
484
|
+
if (!envelope.ok) {
|
|
485
|
+
const reasonCode: JevDecisionRunFailure =
|
|
486
|
+
requested.mode === undefined
|
|
487
|
+
? 'QUESTIONS_INVALID'
|
|
488
|
+
: requested.mode === 'review'
|
|
489
|
+
? 'REVIEW_INPUT_INVALID'
|
|
490
|
+
: requested.mode === 'rank'
|
|
491
|
+
? 'RANK_INPUT_INVALID'
|
|
492
|
+
: requested.mode === 'verify'
|
|
493
|
+
? 'VERIFY_INPUT_INVALID'
|
|
494
|
+
: 'PRESET_INVALID'
|
|
495
|
+
return refusal(reasonCode, envelope.detail)
|
|
496
|
+
}
|
|
497
|
+
}
|
|
498
|
+
|
|
456
499
|
// Review mode (J4): the bounded request is built before any gate work, and
|
|
457
500
|
// an invalid review input is a tool-surface refusal β same class as the
|
|
458
501
|
// caller check, never a gate failure. The v3.1 rank and verify modes follow
|
|
459
502
|
// the same prebuilt-request pattern (no gate work for a malformed input);
|
|
460
503
|
// neither has a telemetry point β they are tool modes, not shadow gates.
|
|
461
504
|
let prebuilt: Omit<NativeDecisionRequest, 'model'> | undefined
|
|
505
|
+
let candidateSource: JevDecisionPreparation['candidateSource'] | undefined
|
|
462
506
|
if (requested.mode === 'review') {
|
|
463
507
|
const prepared = buildJevReviewRequest(input ?? {})
|
|
464
508
|
if (!prepared.ok) {
|
|
@@ -476,7 +520,7 @@ export async function runJevDecision(
|
|
|
476
520
|
prebuilt = prepared.request
|
|
477
521
|
} else if (requested.mode === 'rank') {
|
|
478
522
|
const prepared = buildJevRankRequest(input ?? {})
|
|
479
|
-
if (!prepared.ok) return refusal(prepared.reasonCode)
|
|
523
|
+
if (!prepared.ok) return refusal(prepared.reasonCode, prepared.detail)
|
|
480
524
|
prebuilt = prepared.request
|
|
481
525
|
} else if (requested.mode === 'verify') {
|
|
482
526
|
const prepared = buildJevVerifyRequest(input ?? {})
|
|
@@ -501,12 +545,16 @@ export async function runJevDecision(
|
|
|
501
545
|
: requested.mode === undefined
|
|
502
546
|
? prepareJevDecisionRequest(route, input ?? {})
|
|
503
547
|
: isJevDecisionPreset(requested.mode)
|
|
504
|
-
?
|
|
505
|
-
|
|
506
|
-
|
|
507
|
-
|
|
508
|
-
|
|
509
|
-
|
|
548
|
+
? (async () => {
|
|
549
|
+
const prepared = prepareJevPresetRequest(
|
|
550
|
+
route,
|
|
551
|
+
requested.mode,
|
|
552
|
+
input ?? {},
|
|
553
|
+
requested.mode === 'skills' ? await readHostSkills(runtime) : undefined,
|
|
554
|
+
)
|
|
555
|
+
if (prepared.ok) candidateSource = prepared.candidateSource
|
|
556
|
+
return prepared
|
|
557
|
+
})()
|
|
510
558
|
: { ok: false as const, reasonCode: 'PRESET_INVALID' as const },
|
|
511
559
|
})
|
|
512
560
|
if (!result.ok) {
|
|
@@ -639,6 +687,7 @@ export async function runJevDecision(
|
|
|
639
687
|
billing: result.call.selection.route.billing,
|
|
640
688
|
},
|
|
641
689
|
...(requested.mode === undefined ? {} : { preset: requested.mode }),
|
|
690
|
+
...(candidateSource === undefined ? {} : { candidateSource }),
|
|
642
691
|
answers: advice.answers,
|
|
643
692
|
missing: advice.missing,
|
|
644
693
|
usage: { costUsd: advice.costUsd, tokens: advice.tokens },
|
|
@@ -654,7 +703,7 @@ const TOOL_DESCRIPTION =
|
|
|
654
703
|
'host), `mode: "delegation"` to pick ' +
|
|
655
704
|
'a specialist, `mode: "review"` to score a finished delegated result against named criteria ' +
|
|
656
705
|
'(advisory quality score + needs-review flag; the result content itself is never sent β ' +
|
|
657
|
-
'shape metrics and criteria only), `mode: "rank"` to pick the best of 2-
|
|
706
|
+
'shape metrics and criteria only), `mode: "rank"` to pick the best of 2-20 named candidates ' +
|
|
658
707
|
'(returns the winner and its fit score; no full ordering), `mode: "verify"` to check 1-4 ' +
|
|
659
708
|
'named claims (PASS/FAIL per claim; any refusal, timeout or malformed answer is BLOCK, ' +
|
|
660
709
|
'fail-closed), or pass explicit typed `questions` (choice, score, ' +
|
|
@@ -784,7 +833,7 @@ const JEV_INPUT_SCHEMA: Record<string, unknown> = {
|
|
|
784
833
|
candidates: {
|
|
785
834
|
type: 'array',
|
|
786
835
|
minItems: 2,
|
|
787
|
-
maxItems:
|
|
836
|
+
maxItems: 20,
|
|
788
837
|
items: RANK_CANDIDATE_SCHEMA,
|
|
789
838
|
},
|
|
790
839
|
claims: {
|
|
@@ -224,14 +224,59 @@ export type JevDecisionPreparationDetail =
|
|
|
224
224
|
| 'SCHEMA_VERSION_INVALID'
|
|
225
225
|
| 'UNKNOWN_INPUT_KEYS'
|
|
226
226
|
|
|
227
|
+
export type JevSkillCandidateSource = 'prefilter' | 'full-catalog' | 'host-skills'
|
|
228
|
+
|
|
227
229
|
export type JevDecisionPreparation =
|
|
228
|
-
| {
|
|
230
|
+
| {
|
|
231
|
+
readonly ok: true
|
|
232
|
+
readonly request: NativeDecisionRequest
|
|
233
|
+
/** Skills mode only: which path produced the candidate pool. */
|
|
234
|
+
readonly candidateSource?: JevSkillCandidateSource
|
|
235
|
+
}
|
|
229
236
|
| {
|
|
230
237
|
readonly ok: false
|
|
231
238
|
readonly reasonCode: JevDecisionPreparationFailure
|
|
232
239
|
readonly detail?: JevDecisionPreparationDetail
|
|
233
240
|
}
|
|
234
241
|
|
|
242
|
+
/**
|
|
243
|
+
* The ONE envelope check every tool path passes through before its builder:
|
|
244
|
+
* `schemaVersion` is required (and must be 1) and unknown keys are rejected β
|
|
245
|
+
* the declared contract, now enforced uniformly (beta.10 claimed it; beta.12
|
|
246
|
+
* applies it to presets, review, rank and verify too).
|
|
247
|
+
*/
|
|
248
|
+
export type JevEnvelopeCheck = { ok: true } | { ok: false; detail: JevDecisionPreparationDetail }
|
|
249
|
+
|
|
250
|
+
export function checkJevEnvelope(
|
|
251
|
+
input: unknown,
|
|
252
|
+
allowedKeys: ReadonlySet<string>,
|
|
253
|
+
): JevEnvelopeCheck {
|
|
254
|
+
if (
|
|
255
|
+
input === null ||
|
|
256
|
+
typeof input !== 'object' ||
|
|
257
|
+
Array.isArray(input) ||
|
|
258
|
+
(input as Record<string, unknown>).schemaVersion !== 1
|
|
259
|
+
) {
|
|
260
|
+
return { ok: false, detail: 'SCHEMA_VERSION_INVALID' }
|
|
261
|
+
}
|
|
262
|
+
const unknown = Object.keys(input as Record<string, unknown>).filter(
|
|
263
|
+
(key) => !allowedKeys.has(key),
|
|
264
|
+
)
|
|
265
|
+
if (unknown.length > 0) return { ok: false, detail: 'UNKNOWN_INPUT_KEYS' }
|
|
266
|
+
return { ok: true }
|
|
267
|
+
}
|
|
268
|
+
|
|
269
|
+
export const JEV_EXPLICIT_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
270
|
+
'schemaVersion',
|
|
271
|
+
'state',
|
|
272
|
+
'questions',
|
|
273
|
+
])
|
|
274
|
+
export const JEV_PRESET_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
275
|
+
'schemaVersion',
|
|
276
|
+
'mode',
|
|
277
|
+
'state',
|
|
278
|
+
])
|
|
279
|
+
|
|
235
280
|
function isBoundedNonBlankString(value: unknown, max: number): value is string {
|
|
236
281
|
return typeof value === 'string' && value.trim().length > 0 && value.length <= max
|
|
237
282
|
}
|
|
@@ -420,12 +465,21 @@ export function isJevDecisionPreset(value: unknown): value is JevDecisionPreset
|
|
|
420
465
|
}
|
|
421
466
|
|
|
422
467
|
/** Every canonical specialist except the orchestrator itself. */
|
|
423
|
-
const SPECIALIST_CHOICES: ReadonlyArray<{ value: string }> = Object.freeze(
|
|
468
|
+
const SPECIALIST_CHOICES: ReadonlyArray<{ value: string; description?: string }> = Object.freeze(
|
|
424
469
|
PANTHEON_AGENT_CATALOG.filter((agent) => agent.id !== 'zeus').map((agent) => ({
|
|
425
470
|
value: agent.id,
|
|
471
|
+
description: boundedCatalogDescription(agent.description),
|
|
426
472
|
})),
|
|
427
473
|
)
|
|
428
474
|
|
|
475
|
+
/** Catalog descriptions are short; truncate defensively at the documented bound. */
|
|
476
|
+
function boundedCatalogDescription(value: unknown): string | undefined {
|
|
477
|
+
if (typeof value !== 'string' || value.trim() === '') return undefined
|
|
478
|
+
return value.length > JEV_DECISION_MAX_DESCRIPTION_LENGTH
|
|
479
|
+
? `${value.slice(0, JEV_DECISION_MAX_DESCRIPTION_LENGTH - 1)}β¦`
|
|
480
|
+
: value
|
|
481
|
+
}
|
|
482
|
+
|
|
429
483
|
const SKILL_DESCRIPTIONS = new Map(
|
|
430
484
|
PANTHEON_SKILL_CATALOG.map((entry) => [entry.id, entry.description]),
|
|
431
485
|
)
|
|
@@ -521,7 +575,10 @@ export function prepareJevPresetRequest(
|
|
|
521
575
|
instructions:
|
|
522
576
|
'Choose one listed specialist only when the described task clearly needs it; ' +
|
|
523
577
|
'otherwise choose none. This is advisory only; nothing is dispatched.',
|
|
524
|
-
choices: [
|
|
578
|
+
choices: [
|
|
579
|
+
...SPECIALIST_CHOICES,
|
|
580
|
+
{ value: 'none', description: 'No specialist fits the described task.' },
|
|
581
|
+
],
|
|
525
582
|
},
|
|
526
583
|
],
|
|
527
584
|
} satisfies NativeDecisionRequest
|
|
@@ -529,6 +586,9 @@ export function prepareJevPresetRequest(
|
|
|
529
586
|
}
|
|
530
587
|
|
|
531
588
|
const shortlist = selectPantheonSkills(state).skillIds
|
|
589
|
+
const hostHasSkills = Array.isArray(hostSkills) && hostSkills.length > 0
|
|
590
|
+
const candidateSource: JevSkillCandidateSource =
|
|
591
|
+
shortlist.length > 0 ? 'prefilter' : hostHasSkills ? 'host-skills' : 'full-catalog'
|
|
532
592
|
const pool = shortlist.length > 0 ? shortlist : CATALOG_SKILL_IDS
|
|
533
593
|
const descriptions = new Map(PANTHEON_SKILL_CATALOG.map((entry) => [entry.id, entry.description]))
|
|
534
594
|
const request = {
|
|
@@ -541,11 +601,14 @@ export function prepareJevPresetRequest(
|
|
|
541
601
|
instructions:
|
|
542
602
|
'Choose the skills worth stacking for the described task, or none when none apply. ' +
|
|
543
603
|
'This is advisory only; nothing is loaded or executed by this decision.',
|
|
544
|
-
choices: [
|
|
604
|
+
choices: [
|
|
605
|
+
...mergeHostSkills(pool, hostSkills, descriptions),
|
|
606
|
+
{ value: 'none', description: 'No skill stack fits the described task.' },
|
|
607
|
+
],
|
|
545
608
|
},
|
|
546
609
|
],
|
|
547
610
|
} satisfies NativeDecisionRequest
|
|
548
|
-
return { ok: true, request: Object.freeze(request) }
|
|
611
|
+
return { ok: true, request: Object.freeze(request), candidateSource }
|
|
549
612
|
}
|
|
550
613
|
// βββ Response consumption βββββββββββββββββββββββββββββββββββββββββββββββ
|
|
551
614
|
|
|
@@ -356,7 +356,7 @@ export function encodeQualityScoreReason(input: {
|
|
|
356
356
|
* ANY other key β `state`, `questions`, or a smuggled content field β refuses
|
|
357
357
|
* the preparation, so no content slot exists even at the input boundary.
|
|
358
358
|
*/
|
|
359
|
-
const REVIEW_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
359
|
+
export const REVIEW_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
360
360
|
'schemaVersion',
|
|
361
361
|
'mode',
|
|
362
362
|
'delegateId',
|
|
@@ -54,8 +54,8 @@ import { NATIVE_DECISION_MAX_PROMPT_LENGTH } from './v2-native-decisions.ts'
|
|
|
54
54
|
|
|
55
55
|
/** A ranking needs at least two candidates to choose between. */
|
|
56
56
|
export const RANK_MIN_CANDIDATES = 2
|
|
57
|
-
/**
|
|
58
|
-
export const RANK_MAX_CANDIDATES =
|
|
57
|
+
/** Mirrors the typesafe_rank reference contract (1-20 accepted upstream). */
|
|
58
|
+
export const RANK_MAX_CANDIDATES = 20
|
|
59
59
|
|
|
60
60
|
/** Question name of the rank's choice question. */
|
|
61
61
|
export const RANK_WINNER_QUESTION_NAME = 'winner'
|
|
@@ -164,7 +164,7 @@ function isRecord(value: unknown): value is Record<string, unknown> {
|
|
|
164
164
|
|
|
165
165
|
// βββ Rank: request building ββββββββββββββββββββββββββββββββββββββββββββββ
|
|
166
166
|
|
|
167
|
-
const RANK_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
167
|
+
export const RANK_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
168
168
|
'schemaVersion',
|
|
169
169
|
'mode',
|
|
170
170
|
'state',
|
|
@@ -173,13 +173,24 @@ const RANK_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
|
173
173
|
|
|
174
174
|
export type JevRankBuildFailure = 'RANK_INPUT_INVALID' | 'CANDIDATE_INVALID' | 'STATE_INVALID'
|
|
175
175
|
|
|
176
|
+
/** Per-cause token attached to a rank build refusal (e.g. the accepted count range). */
|
|
177
|
+
export type JevRankBuildDetail = 'RANK_CANDIDATE_COUNT_RANGE'
|
|
178
|
+
|
|
176
179
|
export type JevRankBuild =
|
|
177
180
|
| { readonly ok: true; readonly request: Omit<NativeDecisionRequest, 'model'> }
|
|
178
|
-
| {
|
|
181
|
+
| {
|
|
182
|
+
readonly ok: false
|
|
183
|
+
readonly reasonCode: JevRankBuildFailure
|
|
184
|
+
readonly detail?: JevRankBuildDetail
|
|
185
|
+
}
|
|
179
186
|
|
|
180
187
|
export type JevRankPreparation =
|
|
181
188
|
| { readonly ok: true; readonly request: NativeDecisionRequest }
|
|
182
|
-
| {
|
|
189
|
+
| {
|
|
190
|
+
readonly ok: false
|
|
191
|
+
readonly reasonCode: JevRankBuildFailure
|
|
192
|
+
readonly detail?: JevRankBuildDetail
|
|
193
|
+
}
|
|
183
194
|
|
|
184
195
|
/**
|
|
185
196
|
* Validate the rank tool input and build the model-less request. The state is
|
|
@@ -205,7 +216,7 @@ export function buildJevRankRequest(input: unknown): JevRankBuild {
|
|
|
205
216
|
if (rawCandidates === undefined) return { ok: false, reasonCode: 'RANK_INPUT_INVALID' }
|
|
206
217
|
if (!Array.isArray(rawCandidates)) return { ok: false, reasonCode: 'CANDIDATE_INVALID' }
|
|
207
218
|
if (rawCandidates.length < RANK_MIN_CANDIDATES || rawCandidates.length > RANK_MAX_CANDIDATES) {
|
|
208
|
-
return { ok: false, reasonCode: 'CANDIDATE_INVALID' }
|
|
219
|
+
return { ok: false, reasonCode: 'CANDIDATE_INVALID', detail: 'RANK_CANDIDATE_COUNT_RANGE' }
|
|
209
220
|
}
|
|
210
221
|
const seen = new Set<string>()
|
|
211
222
|
const choices: Array<{ value: string; description?: string }> = []
|
|
@@ -274,7 +285,11 @@ export function prepareJevRankRequest(
|
|
|
274
285
|
|
|
275
286
|
// βββ Verify: request building ββββββββββββββββββββββββββββββββββββββββββββ
|
|
276
287
|
|
|
277
|
-
const VERIFY_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
288
|
+
export const VERIFY_ENVELOPE_KEYS: ReadonlySet<string> = new Set([
|
|
289
|
+
'schemaVersion',
|
|
290
|
+
'mode',
|
|
291
|
+
'claims',
|
|
292
|
+
])
|
|
278
293
|
|
|
279
294
|
export type JevVerifyBuildFailure = 'VERIFY_INPUT_INVALID' | 'CLAIM_INVALID'
|
|
280
295
|
|
|
@@ -32,6 +32,7 @@ export interface LocalDecisionInput {
|
|
|
32
32
|
|
|
33
33
|
export type LocalDecisionRouteReasonCode =
|
|
34
34
|
| 'DISABLED'
|
|
35
|
+
| 'GATE_OPEN'
|
|
35
36
|
| 'INVALID_INPUT'
|
|
36
37
|
| 'NO_DOMAIN'
|
|
37
38
|
| 'AMBIGUOUS_DOMAIN'
|
|
@@ -192,6 +193,15 @@ function evaluateEnabledInput(input: unknown): LocalDecisionRecommendation {
|
|
|
192
193
|
}
|
|
193
194
|
}
|
|
194
195
|
|
|
196
|
+
// D1 (field report beta.11β12): gateStatus must influence the output. An
|
|
197
|
+
// open review/approval gate abstains β the advisory never routes around a
|
|
198
|
+
// gate; resolving the gate is the caller's move, then this router answers.
|
|
199
|
+
if (gateStatus === 'review_required' || gateStatus === 'human_approval_required') {
|
|
200
|
+
return {
|
|
201
|
+
...createAbstention('GATE_OPEN', complexity.reasonCode, gateStatus),
|
|
202
|
+
complexity,
|
|
203
|
+
}
|
|
204
|
+
}
|
|
195
205
|
return {
|
|
196
206
|
schemaVersion: 1,
|
|
197
207
|
policyVersion: LOCAL_DECISION_ROUTER_POLICY_VERSION,
|
|
@@ -53,7 +53,7 @@ export const PANTHEON_AGENT_CATALOG: readonly DecisionCatalogEntry[] = Object.fr
|
|
|
53
53
|
{ id: 'themis', description: 'Quality and security review gate.' },
|
|
54
54
|
{ id: 'prometheus', description: 'Infrastructure, Docker, CI, and model providers.' },
|
|
55
55
|
{ id: 'hephaestus', description: 'AI pipelines: LangChain, RAG, vector stores.' },
|
|
56
|
-
{ id: 'nyx', description: '
|
|
56
|
+
{ id: 'nyx', description: 'Chaos and adversarial testing: fuzzing, hostile inputs, boundaries.' },
|
|
57
57
|
{ id: 'gaia', description: 'Remote sensing and geospatial analysis.' },
|
|
58
58
|
{ id: 'iris', description: 'GitHub branches, pull requests, issues, releases.' },
|
|
59
59
|
{ id: 'mnemosyne', description: 'Memory bank and ADR authoring on request.' },
|