pantheon-opencode 1.6.2 → 1.6.3-beta.10

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 (38) hide show
  1. package/AGENTS.global.md +1 -1
  2. package/AGENTS.md +1 -1
  3. package/CHANGELOG.md +69 -0
  4. package/README.md +14 -2
  5. package/bin/pantheon-init.mjs +1 -1
  6. package/package.json +1 -1
  7. package/plugin.json +1 -1
  8. package/pyproject.toml +1 -1
  9. package/scripts/decision-router-budget.mjs +4 -2
  10. package/scripts/decision-router-v2-egress.mjs +9 -2
  11. package/scripts/decision-router.mjs +58 -10
  12. package/scripts/install/decision-router-config.mjs +94 -8
  13. package/scripts/install/interactive.mjs +78 -10
  14. package/scripts/install/opencode.mjs +20 -0
  15. package/scripts/install/plugin.mjs +17 -0
  16. package/scripts/jev-live-probe.mjs +184 -0
  17. package/scripts/secret-scan.mjs +1 -1
  18. package/scripts/sync-tui.mjs +11 -0
  19. package/src/agents/zeus.md +24 -0
  20. package/src/pantheon/j-telemetry-file.ts +96 -0
  21. package/src/pantheon/jev-decision-tool.ts +240 -27
  22. package/src/pantheon/jev-decisions.ts +78 -14
  23. package/src/pantheon/jev-rank-verify.ts +435 -0
  24. package/src/pantheon/local-decision-budget.ts +11 -3
  25. package/src/pantheon/local-decision-provider-matrix.ts +3 -1
  26. package/src/pantheon/local-decision-usage-evidence.internal.ts +2 -2
  27. package/src/plugin-v2.ts +12 -2
  28. package/src/plugins/tui/dist/tui-v2.d.ts.map +1 -1
  29. package/src/plugins/tui/dist/tui-v2.js +457 -4
  30. package/src/plugins/tui/dist/tui-v2.js.map +1 -1
  31. package/src/plugins/tui/dist/tui.js +1 -1
  32. package/src/plugins/tui/package-lock.json +2 -2
  33. package/src/plugins/tui/package.json +1 -1
  34. package/src/plugins/tui/src/index-v2.tsx +6 -0
  35. package/src/plugins/tui/src/whats-new-highlights.ts +263 -0
  36. package/src/plugins/tui/src/whats-new-open.ts +80 -0
  37. package/src/plugins/tui/src/whats-new-state.ts +128 -0
  38. package/src/plugins/tui/src/whats-new.tsx +119 -0
package/AGENTS.global.md CHANGED
@@ -1,7 +1,7 @@
1
1
  <!-- pantheon:generated:begin — Pantheon-managed region; regenerated on install/upgrade, never edit inside -->
2
2
  <!-- Generated by scripts/build-agents-md.mjs — do not edit directly. -->
3
3
  <!-- Source of truth: src/instructions/*.instructions.md + the static sections in this script. -->
4
- <!-- pantheon:agents-md v2 artifact=global generated-at=2026-10-27T08:34:01-03:00 sha=0a6a8e2 -->
4
+ <!-- pantheon:agents-md v2 artifact=global generated-at=2026-11-04T19:23:07-03:00 sha=ad9f855 -->
5
5
 
6
6
  # Pantheon Agent System — OpenCode
7
7
 
package/AGENTS.md CHANGED
@@ -1,7 +1,7 @@
1
1
  <!-- pantheon:generated:begin — Pantheon-managed region; regenerated on install/upgrade, never edit inside -->
2
2
  <!-- Generated by scripts/build-agents-md.mjs — do not edit directly. -->
3
3
  <!-- Source of truth: project sections in scripts/build-agents-md.mjs. -->
4
- <!-- pantheon:agents-md v2 artifact=project generated-at=2026-10-27T08:34:01-03:00 sha=0a6a8e2 -->
4
+ <!-- pantheon:agents-md v2 artifact=project generated-at=2026-11-04T19:23:07-03:00 sha=ad9f855 -->
5
5
 
6
6
  # Pantheon — Repository Setup
7
7
 
package/CHANGELOG.md CHANGED
@@ -10,6 +10,75 @@ 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.10] - 2026-11-04
14
+
15
+ ### Fixed
16
+
17
+ - **routing** — The Jev input envelope is enforced: schemaVersion is required (SCHEMA_VERSION_INVALID) and unknown keys are rejected (UNKNOWN_INPUT_KEYS) — the declared contract is now the enforced contract (field report N2/N3).
18
+
19
+ ## [v1.6.3-beta.9] - 2026-11-04
20
+
21
+ ### Changed
22
+
23
+ - **agents** — The live scorecard now runs the full 13-specialist catalog (was a 5-agent sample); first full-catalog run: 6 exact hits, 2 domain-adjacent judgment calls, measured per agent.
24
+ - **agents** — Zeus gains consolidated MCP call-timing rules: when to recall, when to store, when to checkpoint, when vision fires, and the escalation order (local selector -> Jev -> MCP last resort).
25
+
26
+ ## [v1.6.3-beta.7] - 2026-11-04
27
+
28
+ ### Added
29
+
30
+ - **routing** — `jev-live-probe.mjs`: the operator-run live measurement tool. `--ping` is the calibration dry-run after `setup-model` (one call, exit 0 when reachable); `--scorecard` runs the live scenarios against the configured route and writes the dated scorecard into `bench/results/`. First live run: 7/8 correct specialist picks via OpenRouter.
31
+
32
+ ## [v1.6.3-beta.8] - 2026-11-04
33
+
34
+ ### Added
35
+
36
+ - **routing** — The v3.1 **rank** and **verify** primitives on `pantheon_jev_decide`: `mode: "rank"` picks the winner among up to 8 candidates with a fit score (the host answers one scalar per question, so full ordering is impossible — documented); `mode: "verify"` checks 1–4 claims as predicate questions with **BLOCK fail-closed**. Hostile inputs rejected. Zeus-only, same gate chain.
37
+ - **agents** — The live scorecard runs the full 13-specialist catalog (first full run: 6/8 exact, 2 domain-adjacent); Zeus gains consolidated MCP call-timing rules.
38
+
39
+ ## [v1.6.3-beta.7] - 2026-11-04
40
+
41
+ ### Added
42
+
43
+ - **install** — The setup wizard gains a **decision-layer step**: the shadow gates (`PANTHEON_J1_RISK_SCORE`, `PANTHEON_J3_ESCALATION`) and the specialist-caller allowlist (`athena,themis`) become first-class setup choices, written to `~/.config/pantheon/decision-flags.env` (source it from your shell profile). All advisory/shadow; none change behavior by default. Part of the ratified setup modernization.
44
+
45
+ ## [v1.6.3-beta.5] - 2026-11-04
46
+
47
+ ### Fixed
48
+
49
+ - **routing** — The ledger model-key slash fix (vendor ids like `typesafe/jev-1.13`) re-landed after being silently lost in a branch reset: the shipped beta.4 refused the OpenRouter model key with `invalid-model`. Field-verified end to end on the user's host: `advisory:true`, real Jev answer via OpenRouter at $0.00003/decision.
50
+ - **install** — The Jev wizard shows the full live route matrix with credential-state hints (checked hint when `OPENCODE_API_KEY`/`TYPESAFE_API_KEY`/`CLOUDFLARE_API_TOKEN` is in the shell; the exact export when not).
51
+
52
+ ## [v1.6.3-beta.4] - 2026-11-04
53
+
54
+ ### Added
55
+
56
+ - **routing** — `decision-router setup-model cloudflare-workers-jev/clef --account-id=<id>` provisions the Cloudflare Workers AI custom provider entry (accountId + `models.clef`, type `"decision"`); the token rides `CLOUDFLARE_API_TOKEN` in the host environment. Docs record the OpenAI Decisions protocol and the what's-new TUI window as deliberately unwired candidates.
57
+
58
+ ## [v1.6.3-beta.3] - 2026-11-04
59
+
60
+ ### Fixed
61
+
62
+ - **routing** — The OpenRouter-hosted Jev route settles end to end: the usage-evidence validator and the budget ledger now admit the `openrouter` accounting provider and vendor-slash model ids (`typesafe/jev-1.13`). Field-verified on the user's host: Zeus → `pantheon_jev_decide` (mode skills) → real Jev answer via the stored OpenRouter credential, $0.00003 per decision, usage reconciled.
63
+
64
+ ## [v1.6.3-beta.2] - 2026-11-04
65
+
66
+ ### Removed
67
+
68
+ - **routing** — The Console Jev **free variant** route (`opencode-console-jev-free` / `jev-1.13-free`) is removed from the matrix, the wizard, and the ledger: the free tier is rate-limited hard enough to 429 mid-work (field-measured), which is exactly the failure mode a decision route must not have. Existing configs selecting it degrade to skip with a fix hint pointing at the Zen-credited route (`OPENCODE_API_KEY`) or the TypeSafe direct route.
69
+
70
+ ## [v1.6.3-beta.1] - 2026-11-04
71
+
72
+ ### Added
73
+
74
+ - **routing** — The decision-provider matrix expands to the host-embedded decision modules confirmed in the 2.2.5 binary: `openrouter-jev` (OpenRouter-hosted `typesafe/jev-1.13`), `vercel-gateway-jev` (Vercel AI Gateway `typesafe-ai/jev`), and `cloudflare-workers-jev` (Workers AI `clef`; needs a custom entry with your accountId). Route resolution stays config-only; `doctor` gains route-aware rows per provider.
75
+ - **routing** — Specialist caller allowlist: `PANTHEON_JEV_SPECIALIST_CALLERS` (CSV, default empty = Zeus only) lets named agents call `pantheon_jev_decide` with explicit typed questions; preset modes stay orchestrator-owned. The permission assertion and the success output carry the real caller; caps and the `decision.egress` kill-switch are enforced per call regardless of caller.
76
+ - **routing** — `DECISION_CALL_FAILED` surfaces the provider reason class, integer status and retry window (rate limits now self-explain); `QUESTIONS_INVALID` carries a per-cause `detail` token (ten causes, one each); the tool description states the required-fields contract on its surface.
77
+
78
+ ### Fixed
79
+
80
+ - **routing** — The TypeSafe provisioner writes a **complete** custom provider entry: the `@ai-sdk/openai-compatible` adapter, the `api.typesafe.ai/v1` baseURL, and an `apiKey` env reference (`{env:TYPESAFE_API_KEY}` — the key itself never lands in the config file). Field evidence: a models-only entry is inert because models.dev has no typesafe-ai entry, so `/connect` and `opencode auth login` cannot supply it and the decisions call 404s. `doctor` now warns when the entry lacks the apiKey; the activation path is `export TYPESAFE_API_KEY=...` + restart.
81
+
13
82
  ## [v1.6.2] - 2026-11-03
14
83
 
15
84
  ### Added
package/README.md CHANGED
@@ -112,7 +112,11 @@ see [Quick Start](docs/QUICKSTART.md).
112
112
  preset/source, git state, and a `Context` block (tokens, a percent gauge,
113
113
  `$<spent> spent`, and a muted `cache <n>% · <n.n> tok/s` line); the native
114
114
  prompt footer keeps its own line, and Pantheon only appends
115
- `▸<preset> · ✦<n> ativos` while delegations are running. The installer
115
+ `▸<preset> · ✦<n> ativos` while delegations are running. When the Pantheon
116
+ version changes since the last TUI start, a one-time dismissable
117
+ "What's new" dialog lists that version's highlights from the shipped
118
+ changelog — once per version, `esc`/`q` dismisses; see
119
+ [What's new window](docs/reference/whats-new.md). The installer
116
120
  selects the `plugins` component by default; in interactive setup you can
117
121
  change the component selection. A component-scoped install includes the TUI
118
122
  only when `plugins` is selected. The TUI ships no quota panel: for provider
@@ -232,7 +236,15 @@ asks, after a delegation completes — one typed round-trip returns an advisory
232
236
  quality score against named criteria plus a needs-review flag. The result
233
237
  content never leaves the machine (shape metrics and criteria only, hostile
234
238
  criteria rejected outright), and the verdict is advisory telemetry plus a
235
- flag: it never blocks, retries, or alters the delegated result.
239
+ flag: it never blocks, retries, or alters the delegated result. Two more
240
+ orchestrator-owned tool modes round out the surface: `mode: "rank"` picks the
241
+ best of 2–8 named candidates for a state in one typed round-trip (it returns
242
+ the winner and that candidate's fit score — the host API answers one scalar
243
+ per question, so no full ordering is returned), and `mode: "verify"` checks
244
+ 1–4 named claims with one predicate question per claim, mapping each answer to
245
+ `PASS` / `FAIL` / `BLOCK` where every refusal, timeout or malformed answer is
246
+ `BLOCK` (fail-closed); claims are bounded caller-provided strings, hostile
247
+ shapes rejected outright, and neither mode adds a telemetry point.
236
248
 
237
249
  The installer may retain a legacy target-specific preference; it does not control
238
250
  runtime routing. `npx pantheon-opencode decision-router ...` manages stored
@@ -1036,7 +1036,7 @@ async function main() {
1036
1036
  const provisioned = provisionTypesafeProvider()
1037
1037
  console.log(
1038
1038
  provisioned === true
1039
- ? ' TypeSafe provider entry merged into opencode.json (models.typesafe-ai, type "decision"). Sign in with `opencode auth login` (TypeSafe) or /connect.'
1039
+ ? ' TypeSafe provider entry merged into opencode.json (adapter + baseURL + apiKey via {env:TYPESAFE_API_KEY}, type "decision"). Activate: export TYPESAFE_API_KEY=<your TypeSafe key> and restart OpenCode. /connect does not carry custom providers.'
1040
1040
  : ` TypeSafe provider entry NOT merged (${provisioned}); merge it manually — see docs/reference/native-decisions.md.`,
1041
1041
  )
1042
1042
  }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "pantheon-opencode",
3
- "version": "1.6.2",
3
+ "version": "1.6.3-beta.10",
4
4
  "description": "Multi-agent orchestration for OpenCode \u2014 14 specialized agents, TDD, Themis review gates.",
5
5
  "type": "module",
6
6
  "workspaces": [
package/plugin.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "pantheon",
3
3
  "publisher": "ils15",
4
- "version": "1.6.2",
4
+ "version": "1.6.3-beta.10",
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 = "1.6.2"
7
+ version = "1.6.3-beta.10"
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
@@ -26,10 +26,12 @@ var LOCAL_DECISION_BUDGET_PROVIDERS = Object.freeze([
26
26
  "openai-decisions",
27
27
  "typesafe-jev",
28
28
  "opencode-console",
29
- "jev-free"
29
+ "openrouter",
30
+ "vercel-ai-gateway",
31
+ "cloudflare-workers-ai"
30
32
  ]);
31
33
  var ALLOWED_PROVIDERS = new Set(LOCAL_DECISION_BUDGET_PROVIDERS);
32
- var MODEL_KEY = /^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/;
34
+ var MODEL_KEY = /^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$/;
33
35
  var RESERVATION_ID = /^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/;
34
36
  var CLOCK_ROLLBACK_TOLERANCE_MS = 6e4;
35
37
  function resolveLocalDecisionBudgetDbPath(env = process.env) {
@@ -14,13 +14,20 @@ var DECISION_ROUTER_MODELS = Object.freeze({
14
14
  "openai-decisions": Object.freeze(["gpt-6-luna"]),
15
15
  "typesafe-jev": Object.freeze(["jev-1.14.0"]),
16
16
  "opencode-console-jev": Object.freeze(["jev-1.13"]),
17
- "opencode-console-jev-free": Object.freeze(["jev-1.13-free"])
17
+ // Host-embedded decision modules confirmed in the 2.2.5 binary; the model
18
+ // ids are the provider-side decision models (see
19
+ // docs/reference/native-decisions.md for credential acquisition).
20
+ "openrouter-jev": Object.freeze(["typesafe/jev-1.13"]),
21
+ "vercel-gateway-jev": Object.freeze(["typesafe-ai/jev"]),
22
+ "cloudflare-workers-jev": Object.freeze(["clef"])
18
23
  });
19
24
  var DECISION_PROVIDER_BUDGET_MAPPING = Object.freeze({
20
25
  "openai-decisions": "openai-decisions",
21
26
  "typesafe-jev": "typesafe-jev",
22
27
  "opencode-console-jev": "opencode-console",
23
- "opencode-console-jev-free": "jev-free"
28
+ "openrouter-jev": "openrouter",
29
+ "vercel-gateway-jev": "vercel-ai-gateway",
30
+ "cloudflare-workers-jev": "cloudflare-workers-ai"
24
31
  });
25
32
  var REMOTELY_ELIGIBLE_DECISION_PROVIDERS = Object.freeze([
26
33
  "openai-decisions",
@@ -32,6 +32,7 @@ import {
32
32
  DECISION_ROUTER_MODELS,
33
33
  diffDecisionRouterSurfaces,
34
34
  inspectTypesafeHostProvider,
35
+ provisionCloudflareHostProvider,
35
36
  provisionTypesafeHostProvider,
36
37
  readUserDecisionRouterConfig,
37
38
  resolveUserDecisionRouterConfigPath,
@@ -314,7 +315,9 @@ export async function runDecisionRouterCommand(
314
315
  )
315
316
  }
316
317
  const noConsent = flagArgs.includes('--no-consent')
317
- const capArgs = flagArgs.filter((arg) => arg !== '--no-consent')
318
+ const capArgs = flagArgs.filter(
319
+ (arg) => arg !== '--no-consent' && !arg.startsWith('--account-id='),
320
+ )
318
321
  // Exactly the two caps flags with values; parseCaps validates them and
319
322
  // leaves `rest` empty because setup-model has no positional caps targets.
320
323
  if (capArgs.length !== 4) throw new DecisionRouterUsageError(USAGE)
@@ -337,6 +340,28 @@ export async function runDecisionRouterCommand(
337
340
  noConsent ? 'Egress consent NOT recorded (--no-consent).' : 'Egress consent recorded: yes.',
338
341
  )
339
342
 
343
+ // The Cloudflare route needs a custom entry with the accountId (the
344
+ // host's embedded cloudflare-workers-ai module reads the token from
345
+ // CLOUDFLARE_API_TOKEN and the account from the entry options).
346
+ if (provider === 'cloudflare-workers-jev') {
347
+ const accountId = flagArgs.find((arg) => arg.startsWith('--account-id='))
348
+ const provision = provisionCloudflareHostProvider({
349
+ accountId: accountId?.split('=')[1],
350
+ env,
351
+ })
352
+ if (provision.ok) {
353
+ stdout(
354
+ `Cloudflare provider entry merged into ${provision.path} (accountId + models.clef, type "decision"). ` +
355
+ 'The token comes from CLOUDFLARE_API_TOKEN (Workers AI-scoped token) in the host environment.',
356
+ )
357
+ } else {
358
+ stdout(
359
+ `⚠ Cloudflare provider entry NOT merged: ${provision.reason}` +
360
+ ' — pass --account-id=<32-hex Cloudflare account id>.',
361
+ )
362
+ }
363
+ }
364
+
340
365
  // The TypeSafe route needs a host-side provider entry the decisions API
341
366
  // can resolve; the Console routes are host-native and need none. The
342
367
  // merge is additive and idempotent; a failure is a printed warning (the
@@ -346,8 +371,8 @@ export async function runDecisionRouterCommand(
346
371
  const provision = provisionTypesafeHostProvider({ env })
347
372
  if (provision.ok) {
348
373
  stdout(
349
- `TypeSafe provider entry merged into ${provision.path} (models.typesafe-ai, type "decision"). ` +
350
- 'Sign in with `opencode auth login` (TypeSafe) or /connect.',
374
+ `TypeSafe provider entry merged into ${provision.path} (adapter + baseURL + apiKey via {env:TYPESAFE_API_KEY}, type "decision"). ` +
375
+ 'Activate: export TYPESAFE_API_KEY=<your TypeSafe key> and restart OpenCode. /connect does not carry custom providers.',
351
376
  )
352
377
  } else {
353
378
  stdout(`⚠ TypeSafe provider entry NOT merged: ${provision.reason}`)
@@ -489,18 +514,41 @@ export async function runDecisionRouterCommand(
489
514
  // Route-aware checks, not a catalog probe: decision models are not chat
490
515
  // models and never appear in `opencode models` (field evidence 2.2.5:
491
516
  // 561 entries, 0 jev, while the decisions API answers the same ref).
492
- if (target.provider === 'typesafe-jev') {
517
+ if (target.provider === 'openrouter-jev' || target.provider === 'vercel-gateway-jev') {
518
+ const envName =
519
+ target.provider === 'openrouter-jev' ? 'OPENROUTER_API_KEY' : 'AI_GATEWAY_API_KEY'
520
+ pass(
521
+ 'native-route',
522
+ `host-native ${target.provider === 'openrouter-jev' ? 'openrouter' : 'vercel-ai-gateway'} route; credential ${envName} via /connect or the environment`,
523
+ )
524
+ } else if (target.provider === 'cloudflare-workers-jev') {
525
+ warn(
526
+ 'native-route',
527
+ 'host module exists but the provider needs a custom entry with your Cloudflare accountId (models.dev does not carry it)',
528
+ 'add providers["cloudflare-workers-ai"] with accountId + CLOUDFLARE_API_TOKEN (scoped Workers AI token); model clef',
529
+ )
530
+ } else if (target.provider === 'typesafe-jev') {
493
531
  const hostProvider = inspectTypesafeHostProvider({ env })
494
- if (hostProvider.present) {
495
- pass(
532
+ if (!hostProvider.present) {
533
+ fail(
496
534
  'native-route',
497
- `typesafe-ai provider entry present (${hostProvider.path}); credential via \`opencode auth login\` (TypeSafe) or /connect`,
535
+ `opencode.json has no providers.typesafe-ai entry (${hostProvider.path})`,
536
+ 'decision-router setup-model typesafe-jev/jev-1.14.0 merges the full provider entry (adapter, baseURL, apiKey ref)',
537
+ )
538
+ } else if (!hostProvider.hasCredential) {
539
+ warn(
540
+ 'native-route',
541
+ 'provider entry has no apiKey',
542
+ 'set the TypeSafe key: export TYPESAFE_API_KEY=... and re-run setup-model (writes {env:TYPESAFE_API_KEY}), or add options.apiKey to the entry',
498
543
  )
499
544
  } else {
500
- fail(
545
+ pass(
501
546
  'native-route',
502
- `opencode.json has no providers.typesafe-ai entry (${hostProvider.path})`,
503
- 'decision-router setup-model typesafe-jev/jev-1.14.0 merges it (type: "decision")',
547
+ `typesafe-ai provider entry present (${hostProvider.path}); credential ${
548
+ hostProvider.viaEnvRef
549
+ ? `via env reference (${hostProvider.apiKeyRef}) — resolved by the host at call time`
550
+ : 'inline in the entry'
551
+ }`,
504
552
  )
505
553
  }
506
554
  } else {
@@ -29,7 +29,12 @@ export const DECISION_ROUTER_MODELS = Object.freeze({
29
29
  'openai-decisions': Object.freeze(['gpt-6-luna']),
30
30
  'typesafe-jev': Object.freeze(['jev-1.14.0']),
31
31
  'opencode-console-jev': Object.freeze(['jev-1.13']),
32
- 'opencode-console-jev-free': Object.freeze(['jev-1.13-free']),
32
+ // Host-embedded decision modules confirmed in the 2.2.5 binary; the model
33
+ // ids are the provider-side decision models (see
34
+ // docs/reference/native-decisions.md for credential acquisition).
35
+ 'openrouter-jev': Object.freeze(['typesafe/jev-1.13']),
36
+ 'vercel-gateway-jev': Object.freeze(['typesafe-ai/jev']),
37
+ 'cloudflare-workers-jev': Object.freeze(['clef']),
33
38
  })
34
39
 
35
40
  /** Single mapping from a user-level provider ID to the local budget provider. */
@@ -37,7 +42,9 @@ export const DECISION_PROVIDER_BUDGET_MAPPING = Object.freeze({
37
42
  'openai-decisions': 'openai-decisions',
38
43
  'typesafe-jev': 'typesafe-jev',
39
44
  'opencode-console-jev': 'opencode-console',
40
- 'opencode-console-jev-free': 'jev-free',
45
+ 'openrouter-jev': 'openrouter',
46
+ 'vercel-gateway-jev': 'vercel-ai-gateway',
47
+ 'cloudflare-workers-jev': 'cloudflare-workers-ai',
41
48
  })
42
49
 
43
50
  /** Providers whose usage and price are documented well enough to reconcile. */
@@ -591,9 +598,22 @@ export function provisionTypesafeHostProvider({ env = process.env } = {}) {
591
598
  }
592
599
  data.providers ??= {}
593
600
  data.providers['typesafe-ai'] ??= {}
594
- data.providers['typesafe-ai'].models ??= {}
595
- if (data.providers['typesafe-ai'].models['jev-1.14.0'] === undefined) {
596
- data.providers['typesafe-ai'].models['jev-1.14.0'] = {
601
+ // A complete custom provider entry: without the npm adapter and the
602
+ // transport options (baseURL + apiKey) the host cannot resolve the
603
+ // provider at all (field evidence: a models-only entry 404s on the
604
+ // decisions API; models.dev does not carry typesafe-ai, so /connect and
605
+ // `opencode auth login` cannot supply it either). The credential rides
606
+ // the `{env:...}` reference — the key itself stays out of the config
607
+ // file; an entry the user already completed is preserved as-is.
608
+ const entry = data.providers['typesafe-ai']
609
+ entry.npm ??= '@ai-sdk/openai-compatible'
610
+ entry.name ??= 'TypeSafe AI'
611
+ entry.options ??= {}
612
+ entry.options.baseURL ??= 'https://api.typesafe.ai/v1'
613
+ entry.options.apiKey ??= '{env:TYPESAFE_API_KEY}'
614
+ entry.models ??= {}
615
+ if (entry.models['jev-1.14.0'] === undefined) {
616
+ entry.models['jev-1.14.0'] = {
597
617
  name: 'Jev (TypeSafe direct)',
598
618
  modelID: 'jev-latest',
599
619
  type: 'decision',
@@ -607,6 +627,51 @@ export function provisionTypesafeHostProvider({ env = process.env } = {}) {
607
627
  }
608
628
  }
609
629
 
630
+ /**
631
+ * Additively merge the Cloudflare Workers AI decision provider entry into the
632
+ * user's host `opencode.json`. The host's embedded `cloudflare-workers-ai`
633
+ * decisions module needs the accountId in the entry options and reads the
634
+ * token from `CLOUDFLARE_API_TOKEN` (a Workers AI-scoped token — not the
635
+ * global Cloudflare key). Existing entries are preserved.
636
+ * @param {{ accountId: string, env?: Record<string, string | undefined> }} options
637
+ * @returns {{ ok: true, path: string } | { ok: false, reason: string }}
638
+ */
639
+ export function provisionCloudflareHostProvider({ accountId, env = process.env } = {}) {
640
+ const trimmed = String(accountId ?? '').trim()
641
+ if (!/^[a-f0-9]{32}$/i.test(trimmed)) {
642
+ return { ok: false, reason: 'accountId must be the 32-hex Cloudflare account id' }
643
+ }
644
+ const configPath = resolveHostOpenCodeConfigPath(env)
645
+ if (!existsSync(configPath)) return { ok: false, reason: 'opencode.json not found' }
646
+ let data
647
+ try {
648
+ const parsed = JSON.parse(readFileSync(configPath, 'utf8'))
649
+ if (parsed === null || typeof parsed !== 'object' || Array.isArray(parsed)) {
650
+ return { ok: false, reason: 'opencode.json is not a JSON object' }
651
+ }
652
+ data = parsed
653
+ } catch {
654
+ return { ok: false, reason: 'opencode.json is not plain JSON (comments?)' }
655
+ }
656
+ data.providers ??= {}
657
+ data.providers['cloudflare-workers-ai'] ??= {}
658
+ data.providers['cloudflare-workers-ai'].options ??= {}
659
+ data.providers['cloudflare-workers-ai'].options.accountId ??= trimmed
660
+ data.providers['cloudflare-workers-ai'].models ??= {}
661
+ if (data.providers['cloudflare-workers-ai'].models['clef'] === undefined) {
662
+ data.providers['cloudflare-workers-ai'].models['clef'] = {
663
+ name: 'Clef (Cloudflare Workers AI)',
664
+ type: 'decision',
665
+ }
666
+ }
667
+ try {
668
+ writeFileSync(configPath, `${JSON.stringify(data, null, 2)}\n`)
669
+ return { ok: true, path: configPath }
670
+ } catch (error) {
671
+ return { ok: false, reason: String(error instanceof Error ? error.message : error) }
672
+ }
673
+ }
674
+
610
675
  /**
611
676
  * Read-only check that the TypeSafe provider entry exists in the host config.
612
677
  * Never writes; an unreadable config reports present:false with the path for
@@ -618,9 +683,30 @@ export function inspectTypesafeHostProvider({ env = process.env } = {}) {
618
683
  const configPath = resolveHostOpenCodeConfigPath(env)
619
684
  try {
620
685
  const parsed = JSON.parse(readFileSync(configPath, 'utf8'))
621
- const model = parsed?.providers?.['typesafe-ai']?.models?.['jev-1.14.0']
622
- return { present: model !== undefined && model !== null, path: configPath }
686
+ const entry = parsed?.providers?.['typesafe-ai']
687
+ const model = entry?.models?.['jev-1.14.0']
688
+ const present = model !== undefined && model !== null
689
+ const apiKey = entry?.options?.apiKey
690
+ // Credential states: inline key, `{env:...}` reference, or missing. The
691
+ // referenced env var is resolved by the HOST at call time — the doctor
692
+ // cannot see the host's environment, so a reference counts as present
693
+ // with a note instead of a verdict.
694
+ const hasCredential = typeof apiKey === 'string' && apiKey.length > 0
695
+ const viaEnvRef = hasCredential && apiKey.startsWith('{env:')
696
+ return {
697
+ present,
698
+ hasCredential,
699
+ viaEnvRef,
700
+ apiKeyRef: hasCredential ? apiKey : null,
701
+ path: configPath,
702
+ }
623
703
  } catch {
624
- return { present: false, path: configPath }
704
+ return {
705
+ present: false,
706
+ hasCredential: false,
707
+ viaEnvRef: false,
708
+ apiKeyRef: null,
709
+ path: configPath,
710
+ }
625
711
  }
626
712
  }
@@ -1,5 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
 
3
+ import { mkdirSync, writeFileSync } from 'node:fs'
4
+ import { join } from 'node:path'
3
5
  import * as clack from '@clack/prompts'
4
6
  import {
5
7
  createDecisionRouterConfig,
@@ -146,17 +148,22 @@ export async function runInteractiveInstall({
146
148
  {
147
149
  label: 'OpenCode Console Jev (stock)',
148
150
  value: 'opencode-console-jev/jev-1.13',
149
- hint: 'Billed to your Console; uses host-owned auth',
151
+ hint: consoleJevHint(process.env),
150
152
  },
151
153
  {
152
- label: 'Console Jev free variant',
153
- value: 'opencode-console-jev-free/jev-1.13-free',
154
- hint: 'No paid budget consumed; host-owned auth',
154
+ label: 'OpenRouter-hosted Jev',
155
+ value: 'openrouter-jev/typesafe/jev-1.13',
156
+ hint: openrouterJevHint(),
155
157
  },
156
158
  {
157
159
  label: 'TypeSafe Jev (direct)',
158
160
  value: 'typesafe-jev/jev-1.14.0',
159
- hint: 'Installer merges the provider entry; you sign in with your TypeSafe key (/connect)',
161
+ hint: typesafeJevHint(process.env),
162
+ },
163
+ {
164
+ label: 'Cloudflare Clef',
165
+ value: 'cloudflare-workers-jev/clef',
166
+ hint: cloudflareJevHint(process.env),
160
167
  },
161
168
  ],
162
169
  initialValue: 'skip',
@@ -164,13 +171,51 @@ export async function runInteractiveInstall({
164
171
  if (cancelPrompt(prompts, jevActivation, 'decisionRouter')) {
165
172
  return { canceled: true, cancelPoint: 'decisionRouter' }
166
173
  }
174
+
175
+ // Decision-layer options: the shadow gates and the specialist-caller
176
+ // allowlist are first-class setup choices. The installer writes a flags
177
+ // file the user sources (the host reads these at plugin registration from
178
+ // its own environment — the installer never edits shell profiles silently).
179
+ const jevFlags = await prompts.multiselect({
180
+ message: 'Decision layer extras (all advisory/shadow; none change behavior by default):',
181
+ options: [
182
+ {
183
+ label: 'J1 risk-score shadow gate',
184
+ value: 'PANTHEON_J1_RISK_SCORE=on',
185
+ hint: 'Records a Jev risk-score per exfil-capable tool call; observe-only',
186
+ },
187
+ {
188
+ label: 'J3 escalation shadow classifier',
189
+ value: 'PANTHEON_J3_ESCALATION=on',
190
+ hint: 'Records cheap-vs-main per delegate dispatch; never applies overrides in shadow',
191
+ },
192
+ {
193
+ label: 'Specialist callers: athena + themis (explicit questions)',
194
+ value: 'CALLERS=athena,themis',
195
+ hint: 'Allowlisted specialists may ask bounded Jev questions mid-task; presets stay Zeus-only',
196
+ },
197
+ ],
198
+ required: false,
199
+ initialValues: [],
200
+ })
201
+ if (cancelPrompt(prompts, jevFlags, 'decisionFlags')) {
202
+ return { canceled: true, cancelPoint: 'decisionFlags' }
203
+ }
204
+ if (Array.isArray(jevFlags) && jevFlags.length > 0) {
205
+ const flagsDir = join(process.env.XDG_CONFIG_HOME || `${process.env.HOME}/.config`, 'pantheon')
206
+ mkdirSync(flagsDir, { recursive: true, mode: 0o700 })
207
+ const flagsPath = join(flagsDir, 'decision-flags.env')
208
+ const lines = jevFlags.map((flag) => `export ${flag}`).join('\n')
209
+ writeFileSync(
210
+ flagsPath,
211
+ `# Pantheon decision layer — source from your shell profile
212
+ ${lines}
213
+ `,
214
+ )
215
+ }
167
216
  // Only known targets activate; anything else (a stale stub, a renamed
168
217
  // option) degrades to skip instead of configuring a wrong route.
169
- const KNOWN_JEV_TARGETS = new Set([
170
- 'opencode-console-jev/jev-1.13',
171
- 'opencode-console-jev-free/jev-1.13-free',
172
- 'typesafe-jev/jev-1.14.0',
173
- ])
218
+ const KNOWN_JEV_TARGETS = new Set(['opencode-console-jev/jev-1.13', 'typesafe-jev/jev-1.14.0'])
174
219
  const activationChoice =
175
220
  typeof jevActivation === 'string' && KNOWN_JEV_TARGETS.has(jevActivation)
176
221
  ? jevActivation
@@ -209,3 +254,26 @@ export async function runInteractiveInstall({
209
254
  decisionModelActivation,
210
255
  }
211
256
  }
257
+
258
+ /** Credential-state hints for the Jev wizard routes (set at prompt time). */
259
+ function consoleJevHint(env = process.env) {
260
+ return env.OPENCODE_API_KEY
261
+ ? 'Zen credits via OPENCODE_API_KEY — ✓ detected in this shell'
262
+ : 'Zen credits via OPENCODE_API_KEY — export it first (opencode.ai/console)'
263
+ }
264
+
265
+ function openrouterJevHint() {
266
+ return 'Uses the host OpenRouter credential (opencode auth list); billed per decision'
267
+ }
268
+
269
+ function typesafeJevHint(env = process.env) {
270
+ return env.TYPESAFE_API_KEY
271
+ ? '✓ TYPESAFE_API_KEY detected — the installer merges the provider entry'
272
+ : 'export TYPESAFE_API_KEY first; the installer merges the provider entry'
273
+ }
274
+
275
+ function cloudflareJevHint(env = process.env) {
276
+ return env.CLOUDFLARE_API_TOKEN
277
+ ? '✓ CLOUDFLARE_API_TOKEN detected — pass --account-id at setup-model'
278
+ : 'Workers AI-scoped CLOUDFLARE_API_TOKEN + accountId; setup-model merges the entry'
279
+ }
@@ -2789,6 +2789,26 @@ async function installPass(
2789
2789
  // .ts path is not a registrable entry on this host and must not be moved.
2790
2790
  config.plugin = removePantheonPluginReferences(config.plugin)
2791
2791
  if (!Array.isArray(config.plugins)) config.plugins = []
2792
+ // Preflight: a Pantheon-owned registration whose target directory is
2793
+ // missing means the previous install was invalidated (the npx cache
2794
+ // recycles `_npx/<hash>` dirs — field evidence: a registered copy
2795
+ // vanished and the host silently dropped the plugin, so the whole tool
2796
+ // surface disappeared until the next init). Name the cause loudly
2797
+ // before the re-registration below fixes it.
2798
+ for (const ref of Array.isArray(config.plugins) ? config.plugins : []) {
2799
+ const refPath = typeof ref === 'string' ? ref : ref?.package
2800
+ if (typeof refPath !== 'string' || managedPluginIdentity(refPath) === null) continue
2801
+ const absolute = isAbsolute(refPath) ? refPath : resolve(target, refPath)
2802
+ if (existsSync(absolute)) continue
2803
+ const staleNpx = refPath.includes('_npx/')
2804
+ warning(
2805
+ `Existing Pantheon registration points to ${refPath}, which no longer exists` +
2806
+ (staleNpx
2807
+ ? ' (the npx cache recycled that copy) — re-registering to the freshly installed tree'
2808
+ : ' — re-registering to the freshly installed tree'),
2809
+ )
2810
+ }
2811
+
2792
2812
  const migrated = []
2793
2813
  const blocked = []
2794
2814
  for (const plugin of config.plugin) {
@@ -215,6 +215,23 @@ export function copyPluginFiles(
215
215
  else result.skipped++
216
216
  }
217
217
 
218
+ // CHANGELOG.md — the copy's one-time "What's new" window reads the release
219
+ // highlights for the running version from its own installed root. Best
220
+ // effort: a missing source simply means no window (the TUI fails open).
221
+ // src/plugins/tui → src/plugins → src → the package root, which ships
222
+ // CHANGELOG.md in its npm `files` allow-list.
223
+ const changelogSrc = join(srcDir, '..', '..', '..', 'CHANGELOG.md')
224
+ if (existsSync(changelogSrc)) {
225
+ const changelogStatus = writeIfChanged(
226
+ join(dstDir, 'CHANGELOG.md'),
227
+ readFileSync(changelogSrc, 'utf8'),
228
+ dryRun,
229
+ { recorder },
230
+ )
231
+ if (changelogStatus === 'created') result.created++
232
+ else result.skipped++
233
+ }
234
+
218
235
  return result
219
236
  }
220
237