pantheon-opencode 1.6.2-beta.2 → 1.6.2-beta.4

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 (57) hide show
  1. package/CHANGELOG.md +48 -0
  2. package/README.md +56 -16
  3. package/bin/pantheon-init.mjs +68 -13
  4. package/package.json +10 -3
  5. package/plugin.json +1 -1
  6. package/prompts/README.md +58 -0
  7. package/prompts/debug-issue.prompt.md +46 -0
  8. package/prompts/dynamic/agora-generated.txt +38 -0
  9. package/prompts/dynamic/orchestrate-generated.txt +51 -0
  10. package/prompts/focus.prompt.md +42 -0
  11. package/prompts/implement-feature.prompt.md +39 -0
  12. package/prompts/optimize-database.prompt.md +58 -0
  13. package/prompts/orchestrate-with-zeus.prompt.md +441 -0
  14. package/prompts/plan-architecture.prompt.md +59 -0
  15. package/prompts/quick-discovery-large-codebase.prompt.md +294 -0
  16. package/prompts/quick-plan-large-feature.prompt.md +197 -0
  17. package/prompts/semantic-summarize.md +50 -0
  18. package/prompts/sketch.prompt.md +91 -0
  19. package/prompts/subtask.prompt.md +131 -0
  20. package/pyproject.toml +1 -1
  21. package/scripts/decision-router-budget.mjs +781 -0
  22. package/scripts/decision-router.mjs +297 -0
  23. package/scripts/install/component-selection.mjs +9 -4
  24. package/scripts/install/decision-router-config.mjs +432 -0
  25. package/scripts/install/health-check.mjs +20 -11
  26. package/scripts/install/interactive.mjs +78 -14
  27. package/scripts/install/interrupt.mjs +4 -3
  28. package/scripts/install/migrate.mjs +123 -72
  29. package/scripts/install/opencode.mjs +162 -114
  30. package/scripts/install/plan.mjs +8 -1
  31. package/scripts/install/state.mjs +12 -0
  32. package/scripts/install/strings.mjs +10 -0
  33. package/scripts/install/venv.mjs +34 -17
  34. package/scripts/restore.mjs +22 -10
  35. package/scripts/test-opencode-v2-sandbox.sh +10 -6
  36. package/src/agents/zeus.md +12 -0
  37. package/src/pantheon/local-decision-agents.ts +170 -0
  38. package/src/pantheon/local-decision-budget.ts +1043 -0
  39. package/src/pantheon/local-decision-egress-gate.ts +264 -0
  40. package/src/pantheon/local-decision-prompt.ts +123 -0
  41. package/src/pantheon/local-decision-provider-matrix.ts +92 -0
  42. package/src/pantheon/local-decision-provider-protocol.ts +335 -0
  43. package/src/pantheon/local-decision-provider.internal.ts +260 -0
  44. package/src/pantheon/local-decision-router.ts +225 -0
  45. package/src/pantheon/local-decision-skills.ts +171 -0
  46. package/src/pantheon/local-decision-tool.ts +126 -0
  47. package/src/pantheon/local-decision-turn.ts +338 -0
  48. package/src/pantheon/local-decision-usage-evidence.internal.ts +207 -0
  49. package/src/pantheon/local-orchestration.ts +322 -0
  50. package/src/pantheon/pantheon-decision-catalog.ts +101 -0
  51. package/src/pantheon/v2-tools.ts +15 -14
  52. package/src/plugin-v2.ts +52 -19
  53. package/src/plugin.ts +4 -0
  54. package/src/plugins/tui/dist/tui-v2.js +3 -3
  55. package/src/plugins/tui/dist/tui.js +1 -1
  56. package/src/plugins/tui/package-lock.json +2 -2
  57. package/src/plugins/tui/package.json +1 -1
package/CHANGELOG.md CHANGED
@@ -10,6 +10,54 @@ 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
+ ## [v1.6.2-beta.4] - 2026-10-31
14
+
15
+ ### Added
16
+
17
+ - **install** — Added a user-level `decision-router` admin CLI for enablement, selection, consent, per-user/provider/model caps, and period confirmation; config is private and non-secret.
18
+ - **install** — Added `decision-router holds` to list unresolved reservations without reserving, reconciling, releasing, or pruning.
19
+ - **docs** — Added an offline protocol for separately evaluating agent routing and task complexity; no benchmark results or provider selection are claimed.
20
+ - **routing** — Added a default-off, local-only recommendation API for exact canonical agent roles and structured complexity dimensions; it does not dispatch agents or change gates.
21
+ - **routing** — Added a private V2 per-turn selector: allowlisted skills, at most one specialist, one call per session/message, and deterministic local fallback.
22
+ - **routing** — Added a private fixed-endpoint adapter for `gpt-6-luna` and `jev-1.14.0` with no caller-supplied transport, retry, or fan-out; Console/Free and Bifrost stay ineligible.
23
+ - **routing** — Provider contract tests run under the real `tsx` runner (`npm run test:ts`) for the exact payload, gate denials, usage reconciliation, and one call per turn; no live call was made.
24
+ - **routing** — Added a user-level `enabled` switch, canonical provider/model matrix, and a single ledger mapping; provider egress also requires the host `permission.assert` gate.
25
+ - **routing** — Added a user-scoped SQLite ledger with explicit UTC reservations and no default caps. Admission caps are local safeguards, not invoice guarantees.
26
+ - **docs** — `decision-router status`/`holds` are read paths: `status` may create the SQLite ledger, and both may migrate it, but neither reserves, reconciles, releases, or prunes.
27
+ - **docs** — The `local-decision-router` subpath is TypeScript source needing the host runtime or a transpiler; the published CLI uses the committed ESM bundle.
28
+
29
+ ### Changed
30
+
31
+ - **security** — Decision egress is fail-closed: a missing gate, cap, period, or credential means zero requests; V1 stays local-only.
32
+ - **release** — `prepack` now runs the `check:decision-router` drift gate, so a stale ledger bundle cannot be packed or published.
33
+
34
+ ### Fixed
35
+
36
+ - **install** — `decision-router caps user` without flags now reports `Both --monthly-usd and --per-decision-usd are required` as a one-line usage error instead of a stack trace.
37
+ - **routing** — `permission.assert` denial (thrown or returned, including the host's `Permission denied: decision.egress` message) means zero requests; an explicit `deny` rule is the kill-switch.
38
+ - **routing** — Egress trusts the user-level config; `permission.assert` is only the host kill-switch. A v2.2.0 canary showed an unruled assertion resolves allow and `evaluate` runs only on that path.
39
+ - **routing** — A cached provider decision re-reads the user-level config/consent gate before reuse, so revoking consent takes effect without waiting for cache expiry.
40
+ - **security** — The prompt hook reuses one decision per turn and never logs or persists the raw request; payloads stay allowlisted.
41
+ - **security** — Credential rejection now requires a token boundary and structured `sk-` prefixes, so prose such as `task-force-*` or `disk-usage-*` is no longer treated as a key.
42
+ - **routing** — Reusing a cached provider decision does not re-prompt the same permission gate; a different or missing gate is still evaluated and denies fail-closed.
43
+ - **docs** — Clarified that `decision-router holds` performs no reservation, reconciliation, release, or pruning, while opening an existing ledger may apply a forward-only schema migration.
44
+
45
+ &lt;!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
46
+ move this section to a versioned entry and reset the template below. --&gt;
47
+ ## [v1.6.2-beta.3] - 2026-10-29
48
+
49
+ ### Changed
50
+
51
+ - **install** — Direct `init` uses matching TTY/headless component defaults; Python runtime and prompt templates are opt-in, and `--yes` suppresses prompts without skipping preview or drift checks.
52
+ - **install** — Runtime health checks verify SQLite FTS5 capability and fail when it is unavailable.
53
+
54
+ ### Fixed
55
+
56
+ - **install** — Explicit restore decline exits successfully; missing authorization fails closed, while cancellation exits `130`. Post-write interrupts and runtime failures record the selected target and partial/degraded state.
57
+ - **install** — Migration schema version is tracked separately from the installed package version and remains idempotent across prerelease package versions.
58
+
59
+ &lt;!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
60
+ move this section to a versioned entry and reset the template below. --&gt;
13
61
  ## [v1.6.2-beta.2] - 2026-10-28
14
62
 
15
63
  ### Added
package/README.md CHANGED
@@ -46,22 +46,25 @@ npx pantheon-opencode init
46
46
  opencode
47
47
  ```
48
48
 
49
- For the optional full setup flow, run `npm run setup`. On detected OpenCode V2,
49
+ Direct `init` uses the same component defaults in TTY and headless runs:
50
+ `agents`, `skills`, `instructions`, `commands`, and `plugins`. `runtime` and
51
+ `prompts` are opt-in. A TTY run selects components and target, shows a preview,
52
+ then asks once for final authorization (default **no**). Without a TTY, `init`
53
+ is preview-only unless `--yes` explicitly authorizes the apply; `--yes` skips
54
+ prompts but still prints the preview and checks for drift.
55
+
56
+ `npm run setup` is a separate repository setup wrapper. On detected OpenCode V2
50
57
  it reviews the global config before installer init: legacy permission rules can
51
- be migrated after explicit review. Verified Pantheon plugin/MCP entries are
52
- reported; JSONC entries move only when order and overrides stay safe, otherwise
53
- the user must choose to keep them in place or abort. Config migration and
54
- installer changes are preflighted before one final authorization; `--yes` alone
55
- does not authorize writes. Overlapping MCP wildcard policies that could allow
56
- an action after a denial are blocked, and ineffective MCP `ask` rules matching
57
- configured or installer-seeded servers require an explicit outcome.
58
- Existing native V2 `permissions` arrays stay authoritative; Pantheon does not
59
- append legacy default grants to them.
60
- Flat third-party MCP entries in `opencode.jsonc` block V2 setup until manually
61
- converted to `mcp.servers` or removed; nested third-party servers are preserved.
62
- Headless/CI/non-TTY runs are preview-only and do not invoke writing init (V1
63
- included). See the [installation guide](docs/INSTALLATION.md) for layer handling,
64
- conflict choices, backups, and recovery.
58
+ be migrated after explicit review. Config migration and installer changes are
59
+ preflighted before one final authorization; `--yes` alone does not authorize
60
+ setup writes, and headless/CI setup remains preview-only. Direct
61
+ `npx pantheon-opencode init` does not run that V2 config-migration review. See
62
+ the [installation guide](docs/INSTALLATION.md) for layer handling, conflict
63
+ choices, backups, and recovery.
64
+ Existing native V2 `permissions` arrays remain authoritative. Unsafe MCP
65
+ wildcard policies and ineffective `ask` rules require explicit resolution;
66
+ third-party flat MCP entries in `opencode.jsonc` must be converted to
67
+ `mcp.servers` or removed before V2 setup.
65
68
 
66
69
  Global `init` and `setup` use `$XDG_CONFIG_HOME/opencode` when
67
70
  `XDG_CONFIG_HOME` is absolute; otherwise they use `~/.config/opencode`.
@@ -94,6 +97,14 @@ see [Quick Start](docs/QUICKSTART.md).
94
97
  and documenting work.
95
98
  - **5 local MCP servers** for project memory, persistence, resources, confined
96
99
  script execution, and vision.
100
+ - **A selective, advisory decision router** with structured local results and an
101
+ optional per-turn skill/agent recommendation. Provider egress is fail-closed: it
102
+ requires explicit enablement, consent, one cap-valid `openai-decisions` or
103
+ `typesafe-jev` candidate, a confirmed period, and the host permission gate.
104
+ Console/Free and Bifrost are never eligible. This statement applies to this
105
+ decision subsystem—not to all Pantheon egress; quota and Vision integrations
106
+ retain their separately documented external requests. See the
107
+ [router reference](docs/reference/local-decision-router.md).
97
108
  - **An optional TUI** with a dispatch sidebar block and a full-width
98
109
  `/dispatch` panel. The V2 TUI also provides a quota overview in the
99
110
  sidebar/footer and a `/quota` details panel. Quota display is enabled by
@@ -137,6 +148,11 @@ see [Quick Start](docs/QUICKSTART.md).
137
148
  the
138
149
  [quota provider coverage table](docs/tui-v2-quota-providers.md) for limits and
139
150
  troubleshooting.
151
+
152
+
153
+ Quota status requests are separate from the decision subsystem and continue to use
154
+ their documented provider APIs. Removing decision-provider transport does not
155
+ disable these quota requests.
140
156
  `/quota-go-check` is a separate manual, confirmation-gated diagnostic: one
141
157
  GET to `https://opencode.ai/zen/go/v1/usage` with the configured Go API key
142
158
  (not OAuth), at most one attempt per plugin runtime. It never refreshes quota,
@@ -226,6 +242,30 @@ fan-in or job recovery without an exposed mechanism.
226
242
  The delegation model, caller/target matrix, and approval gates are documented
227
243
  in [Architecture](docs/ARCHITECTURE.md) and
228
244
  [Permissions, Gates & Delegation](docs/PERMISSIONS-QUALITY-DELEGATION.md).
245
+ The [local decision router](docs/reference/local-decision-router.md) is
246
+ consultative: it narrows an advisory skill list and can recommend at most one
247
+ specialist, but it never intercepts native child-session dispatch or changes
248
+ permissions. The standalone API is default-off
249
+ (`PANTHEON_LOCAL_DECISION_ROUTER=1`); per-turn selection additionally requires
250
+ `enabled: true` in the user-level config. Provider egress is fail-closed and
251
+ requires every local gate plus a non-denying host `ctx.permission.assert`. The
252
+ host is not deny-by-default for `decision.egress` (an unruled assertion resolved
253
+ allow on the measured canary), so the user-level config/consent gate is the
254
+ source of truth and an explicit host `deny` rule is the kill-switch. A canary
255
+ measured the assertion contract but made no live provider call, so egress remains
256
+ unverified and off by default. Native child-session dispatch remains the final
257
+ routing action.
258
+
259
+ The installer may retain a legacy target-specific preference; it does not control
260
+ runtime routing. `npx pantheon-opencode decision-router ...` manages user-level
261
+ enablement, selection, candidates, consent, caps, and period confirmation. The
262
+ separate SQLite ledger has no default caps. Caps limit local admission but are not
263
+ invoice guarantees. Installation does not contact a model provider or inspect
264
+ credentials. The published CLI imports a committed ESM ledger bundle:
265
+ `npm run build:decision-router` regenerates it and `npm run check:decision-router`
266
+ (the `prepack`/`prepublishOnly` drift gate) fails when it is stale. The
267
+ `pantheon-opencode/local-decision-router` subpath itself is TypeScript source, so
268
+ it loads only in the host runtime or under a transpiler such as `tsx`.
229
269
 
230
270
  ## Documentation
231
271
 
@@ -237,7 +277,7 @@ in [Architecture](docs/ARCHITECTURE.md) and
237
277
  | [Compatibility](docs/COMPATIBILITY.md) | [OpenCode V2](docs/opencode-v2.md) |
238
278
  | [Agents & Skills](docs/agents/README.md) | [Platforms](docs/PLATFORMS.md) |
239
279
  | [Release Process](docs/RELEASING.md) | [Branch Protection](docs/BRANCH-PROTECTION.md) |
240
- | [Model Routing](docs/reference/model-routing.md) | [Env Vars](docs/reference/env-vars.md) · [Code Mode](docs/reference/code-mode.md) |
280
+ | [Model Routing](docs/reference/model-routing.md) | [Local Decision Router](docs/reference/local-decision-router.md) · [Offline Evaluation Protocol](docs/reference/offline-evaluation.md) · [Env Vars](docs/reference/env-vars.md) · [Code Mode](docs/reference/code-mode.md) |
241
281
  | [Contributing](CONTRIBUTING.md) | [Roadmap](ROADMAP.md) · [Changelog](CHANGELOG.md) |
242
282
 
243
283
  The release install gate's tool canary loads the installed tarball's V2 plugin
@@ -84,6 +84,7 @@ function printUsage() {
84
84
  ' npx pantheon-opencode init --merge-dev-config <path> # Merge a dev config (off by default)',
85
85
  )
86
86
  console.log(' npx pantheon-opencode set-tier <name> # Set active model preset (global)')
87
+ console.log(' npx pantheon-opencode decision-router status|select|enable|consent|caps|period')
87
88
  console.log(
88
89
  ' npx pantheon-opencode update [--stable] # Update package + re-run init (default: beta channel)',
89
90
  )
@@ -192,6 +193,22 @@ async function main() {
192
193
  process.exit(result.status ?? 1)
193
194
  }
194
195
 
196
+ if (command === 'decision-router') {
197
+ const { DecisionRouterUsageError, runDecisionRouterCommand } = await import(
198
+ '../scripts/decision-router.mjs'
199
+ )
200
+ try {
201
+ await runDecisionRouterCommand(args.slice(1))
202
+ } catch (error) {
203
+ // A malformed command line is user input, not a crash: print one line and
204
+ // skip the stack the top-level handler would otherwise dump.
205
+ if (!(error instanceof DecisionRouterUsageError)) throw error
206
+ console.error(`Error: ${error.message}`)
207
+ process.exit(1)
208
+ }
209
+ return
210
+ }
211
+
195
212
  // beta.5: one-command update — check npm dist-tag, install, re-init.
196
213
  // Default channel is `beta` (the prerelease line); `--stable` uses `latest`.
197
214
  if (command === 'update') {
@@ -316,9 +333,15 @@ async function main() {
316
333
  !Array.isArray(options.components) ||
317
334
  options.components.some(
318
335
  (component) =>
319
- !['agents', 'skills', 'instructions', 'commands', 'plugins', 'runtime'].includes(
320
- component,
321
- ),
336
+ ![
337
+ 'agents',
338
+ 'skills',
339
+ 'instructions',
340
+ 'prompts',
341
+ 'commands',
342
+ 'plugins',
343
+ 'runtime',
344
+ ].includes(component),
322
345
  ) ||
323
346
  options.clean !== false ||
324
347
  !['v1', 'v2', 'auto'].includes(options.version) ||
@@ -482,8 +505,8 @@ async function main() {
482
505
  process.exit(1)
483
506
  }
484
507
 
485
- // Resolve the user's explicit selection separately from the defaults. An
486
- // unscoped install gets runtime by default; explicit --components does not.
508
+ // Resolve the user's explicit selection separately from the defaults.
509
+ // Runtime and prompt templates are opt-in in both TTY and headless mode.
487
510
  let componentSelection
488
511
  try {
489
512
  componentSelection = resolveComponentSelection(args, {
@@ -525,6 +548,9 @@ async function main() {
525
548
  // installer's own default (`~/.config/opencode`), so the record lands next
526
549
  // to the config the next run will read.
527
550
  const fallbackTarget = isProject ? process.cwd() : resolveGlobalConfigDirectory(process.env)
551
+ let chosenTarget = reviewedPlan?.target.path ?? fallbackTarget
552
+ let hasAppliedWrites = false
553
+ let recoveryBackup = null
528
554
  /**
529
555
  * Persist the one-line record of this run. Bookkeeping is best-effort: a
530
556
  * failure is reported, never fatal, and never changes the exit code.
@@ -532,7 +558,7 @@ async function main() {
532
558
  * @param {{status: string, reason?: string|null, cancelPoint?: string|null, applied: boolean, target?: string}} outcome
533
559
  */
534
560
  const recordRun = (outcome) => {
535
- const recordTarget = outcome.target ?? result?.target ?? fallbackTarget
561
+ const recordTarget = outcome.target ?? result?.target ?? chosenTarget
536
562
  const record = buildRunRecord({
537
563
  status: outcome.status,
538
564
  reason: outcome.reason ?? null,
@@ -610,16 +636,20 @@ async function main() {
610
636
  // consumes Ctrl-C in raw mode while a question is on screen (that path is
611
637
  // the interactive cancel above), so this covers the rest of the run.
612
638
  const onInterrupt = createInterruptHandler({
613
- message: S.canceledAt(S.cancelPoints.interrupt),
639
+ message: () =>
640
+ hasAppliedWrites
641
+ ? S.interruptedAfterWrite(recoveryBackup)
642
+ : S.canceledAt(S.cancelPoints.interrupt),
614
643
  json: isJson,
615
644
  endGaugeLine,
616
645
  onInterrupt: () => {
617
646
  if (!isDryRun) {
618
647
  recordRun({
619
- status: 'canceled',
620
- reason: 'canceled',
648
+ status: hasAppliedWrites ? 'partial' : 'canceled',
649
+ reason: hasAppliedWrites ? 'interrupted' : 'canceled',
621
650
  cancelPoint: 'interrupt',
622
- applied: false,
651
+ applied: hasAppliedWrites,
652
+ target: chosenTarget,
623
653
  })
624
654
  }
625
655
  },
@@ -688,7 +718,21 @@ async function main() {
688
718
  quiet: isJson,
689
719
  verbose: args.includes('--verbose') && !isJson,
690
720
  configPreview,
721
+ onTargetResolved: (resolvedTarget) => {
722
+ chosenTarget = resolvedTarget
723
+ },
724
+ onWrite: ({ target: writeTarget, backupDir }) => {
725
+ hasAppliedWrites = true
726
+ chosenTarget = writeTarget
727
+ if (backupDir) recoveryBackup = backupDir
728
+ },
691
729
  })
730
+ // File writers and the runtime installer use synchronous APIs. If SIGINT
731
+ // arrives during one of those calls, Node queues the signal callback
732
+ // until the event loop runs again. Keep the handler armed for one turn
733
+ // after the installer returns so a queued post-write interrupt is still
734
+ // recorded as partial instead of being lost to the success path.
735
+ await new Promise((resolve) => setImmediate(resolve))
692
736
  } catch (err) {
693
737
  installError = err
694
738
  } finally {
@@ -711,8 +755,15 @@ async function main() {
711
755
  `${S.installFailed(installError.message)}\n${S.failHintNoMcp}\n${S.failHintForce}`,
712
756
  )
713
757
  }
714
- // No record for a thrown failure: the error itself is the explanation,
715
- // and a run that failed has nothing to remember beyond its message.
758
+ if (!isDryRun) {
759
+ const outcome = installError.installOutcome ?? {}
760
+ recordRun({
761
+ status: outcome.applied ? 'partial' : 'failed',
762
+ reason: 'failed',
763
+ applied: outcome.applied === true,
764
+ target: outcome.target ?? chosenTarget,
765
+ })
766
+ }
716
767
  process.exitCode = 1
717
768
  return
718
769
  }
@@ -812,6 +863,7 @@ async function main() {
812
863
  totals,
813
864
  mcp_servers: mcpServers,
814
865
  mcp_server_count: mcpServers.length,
866
+ runtime_effects: result.runtimeEffects ?? null,
815
867
  files: files.map(({ path: filePath, action, bytes }) => ({
816
868
  path: path.relative(result.target, filePath) || '.',
817
869
  action,
@@ -881,7 +933,10 @@ async function main() {
881
933
  const doctorStartedAt = Date.now()
882
934
  const { spawnSync } = await import('node:child_process')
883
935
  const doctorScript = path.join(ROOT, 'scripts', 'doctor.mjs')
884
- const doctor = spawnSync(process.execPath, [doctorScript], { stdio: 'inherit', cwd: ROOT })
936
+ const doctor = spawnSync(process.execPath, [doctorScript, '--target', result.target], {
937
+ stdio: 'inherit',
938
+ cwd: ROOT,
939
+ })
885
940
  const doctorSeconds = ((Date.now() - doctorStartedAt) / 1000).toFixed(1)
886
941
  // Never leave "Running health check..." as the last word: report the
887
942
  // outcome and duration, and propagate the child's exit code so CI sees
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "pantheon-opencode",
3
- "version": "1.6.2-beta.2",
3
+ "version": "1.6.2-beta.4",
4
4
  "description": "Multi-agent orchestration for OpenCode — 14 specialized agents, TDD, Themis review gates.",
5
5
  "type": "module",
6
6
  "workspaces": [
@@ -11,6 +11,7 @@
11
11
  "registry": "https://registry.npmjs.org/"
12
12
  },
13
13
  "exports": {
14
+ "./local-decision-router": "./src/pantheon/local-decision-router.ts",
14
15
  "./plugin": "./src/plugin.ts",
15
16
  "./plugin-v2": "./src/plugin-v2.ts",
16
17
  "./v2-bridge": "./src/pantheon/v2-bridge.ts"
@@ -20,6 +21,8 @@
20
21
  "prepare": "husky",
21
22
  "setup": "node scripts/setup.mjs",
22
23
  "build:agents-md": "node scripts/build-agents-md.mjs",
24
+ "build:decision-router": "node scripts/build-decision-router.mjs",
25
+ "check:decision-router": "node scripts/build-decision-router.mjs --check",
23
26
  "skills:lock": "node scripts/update-skills-lock.mjs",
24
27
  "setup:lite": "node bin/pantheon-init.mjs init --yes --clean --components agents,skills,instructions",
25
28
  "doctor": "node scripts/doctor.mjs",
@@ -27,7 +30,8 @@
27
30
  "init": "node bin/pantheon-init.mjs init --yes",
28
31
  "opencode": "OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS=true opencode",
29
32
  "start": "OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS=true opencode",
30
- "prepublishOnly": "node scripts/validate-routing.mjs",
33
+ "prepack": "npm run check:decision-router",
34
+ "prepublishOnly": "node scripts/validate-routing.mjs && npm run check:decision-router",
31
35
  "postpublish": "node -e \"console.log('✅ Published pantheon-opencode v' + require('./package.json').version + ' to npm')\"",
32
36
  "version:status": "node scripts/versioning.mjs status",
33
37
  "version:apply": "node scripts/versioning.mjs apply",
@@ -35,7 +39,7 @@
35
39
  "release:dry-run": "node scripts/versioning.mjs status && node scripts/release-notes.mjs && node scripts/package-evidence.mjs",
36
40
  "package:evidence": "node scripts/package-evidence.mjs",
37
41
  "engine:check": "node scripts/validate-node-engine-policy.mjs",
38
- "package:validate": "node scripts/validate-package.mjs && npm run engine:check",
42
+ "package:validate": "node scripts/validate-package.mjs && npm run check:decision-router && npm run engine:check",
39
43
  "lint": "biome ci .",
40
44
  "docs:lint": "node scripts/docs-link-check.mjs",
41
45
  "docs:media": "node scripts/docs-media.mjs",
@@ -87,6 +91,7 @@
87
91
  "src/instructions/*.instructions.md",
88
92
  "src/mcp/*.mjs",
89
93
  "scripts/*.mjs",
94
+ "!scripts/build-decision-router.mjs",
90
95
  "src/mcp/*.sh",
91
96
  "src/routing.yml",
92
97
  "src/pantheon/**",
@@ -100,6 +105,7 @@
100
105
  "src/plugins/tui/**",
101
106
  "!src/plugins/tui/node_modules/**",
102
107
  "src/skills/**",
108
+ "prompts/**",
103
109
  "!**/__pycache__/**",
104
110
  "!**/*.pyc"
105
111
  ],
@@ -115,6 +121,7 @@
115
121
  "@commitlint/cli": "21.2.3",
116
122
  "@commitlint/config-conventional": "21.2.3",
117
123
  "ajv": "8.20.0",
124
+ "esbuild": "0.28.2",
118
125
  "husky": "^9.1.7",
119
126
  "tsx": "4.23.15",
120
127
  "typescript": "7.0.2"
package/plugin.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "pantheon",
3
3
  "publisher": "ils15",
4
- "version": "1.6.2-beta.2",
4
+ "version": "1.6.2-beta.4",
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",
@@ -0,0 +1,58 @@
1
+ # Prompt Files
2
+
3
+ Reusable prompt templates for common development workflows. Each prompt
4
+ specifies a target agent, required tools, and structured instructions.
5
+
6
+ ## Available Prompts (10)
7
+
8
+ | Prompt | Agent | Purpose |
9
+ |---|---|---|
10
+ | **orchestrate-with-zeus** | Zeus | Full feature orchestration — planning, implementation, review, deployment |
11
+ | **implement-feature** | Zeus | End-to-end feature implementation with TDD and quality gates |
12
+ | **plan-architecture** | Athena | Strategic architecture planning with research and TDD-driven plan |
13
+ | **debug-issue** | Apollo | Rapid debugging with parallel file discovery and analysis |
14
+ | **optimize-database** | Demeter | Database schema, query, and performance optimization |
15
+ | **focus** | Zeus | Pin a session goal to keep all work aligned with a single objective |
16
+ | **sketch** | Athena | Interview a rough idea into a structured feature spec (3–5 questions) |
17
+ | **subtask** | Zeus | Delegate a bounded child task to a specialist and receive a structured result |
18
+ | **quick-discovery-large-codebase** | Apollo | Fast-path bounded discovery for large codebases (8-minute cap) |
19
+ | **quick-plan-large-feature** | Athena | Fast-path bounded planning for large features (5-minute cap) |
20
+
21
+ ## Usage
22
+
23
+ Prompt templates are **not** slash commands. OpenCode slash commands live in
24
+ `commands/` and are invoked in chat:
25
+
26
+ ```text
27
+ /pantheon Add CSV export to the reports page, including tests and a review.
28
+ /pantheon-audit Changes in src/auth/
29
+ /pantheon-focus Implement email verification end-to-end with TDD
30
+ /pantheon-deepwork Migrate the billing module
31
+ /pantheon-model status
32
+ /pantheon-optimize
33
+ /pantheon-vision
34
+ /pantheon-consolidate
35
+ ```
36
+
37
+ The templates in this directory are loaded by the command or agent that needs
38
+ them; they are not invoked directly by filename. In particular:
39
+
40
+ - `focus.prompt.md` is used by `/pantheon-focus`.
41
+ - The audit guidance is used by `/pantheon-audit`.
42
+ - `sketch.prompt.md` has **no matching command file**, so it is available as a
43
+ template only. There is currently no `/sketch` command.
44
+
45
+ For the full command list, see the commands directory in the repository. For
46
+ agent roles, see `docs/agents/README.md`.
47
+
48
+ ## Placeholder contract
49
+
50
+ OpenCode expands `$ARGUMENTS` and positional values such as `$1` in slash
51
+ commands. Those are command-runtime variables; prompt templates do not expand
52
+ `$input`, shell variables, angle-bracket examples, or JSON braces.
53
+
54
+ Manual prompt-template inputs use `{{UPPER_SNAKE_CASE}}`, for example
55
+ `{{TASK_INPUT}}` or `{{TARGET_AGENT}}`. The command or agent that loads the
56
+ template fills these fields before dispatch. This repository does not perform
57
+ automatic placeholder substitution or escaping, so callers must insert values
58
+ as data and preserve their intended quoting and boundaries.
@@ -0,0 +1,46 @@
1
+ ---
2
+ name: debug-issue
3
+ description: "Rapidly debug issues with parallel file discovery and analysis"
4
+ agent: apollo
5
+ tools: ['search', 'usages']
6
+ ---
7
+
8
+ # Debug Quickly (Apollo - Scout)
9
+
10
+ ## Adaptive Search Strategy
11
+
12
+ Choose only the independent searches that fit the report. These are candidate
13
+ angles, not a required fan-out; a focused bug may need only one or two:
14
+
15
+ Search 1: Exact error message
16
+ Search 2: Mentioned function/class
17
+ Search 3: Related file patterns
18
+ Search 4: Similar test cases
19
+ Search 5: Related issues/PRs
20
+ Search 6: Stack trace patterns
21
+ Search 7: Config files
22
+ Search 8: Logging statements
23
+ Search 9: Mentioned dependencies
24
+ Search 10: Recent changes in area
25
+
26
+ ## Analysis & Synthesis
27
+ - Eliminate duplicates
28
+ - Identify file relationships
29
+ - Extract patterns
30
+ - Return prioritized findings
31
+
32
+ ## Recommendations
33
+
34
+ After analysis, @apollo returns:
35
+ 1. **Root Cause Hypothesis**: What's probably wrong
36
+ 2. **Key Files**: Most relevant files
37
+ 3. **Reproduction Steps**: How to reproduce
38
+ 4. **Suggested Fix**: Solution direction
39
+ 5. **Next Steps**: Next agents to involve
40
+
41
+ ## When to Use
42
+ - Bug report received
43
+ - Confusing error stack trace
44
+ - Performance issue diagnosis
45
+ - Regression debugging
46
+ - When you need quick exploration
@@ -0,0 +1,38 @@
1
+ # =============================================================================
2
+ # Dynamic Agora Prompt
3
+ # Generated: 2026-05-16T13:50:41Z
4
+ # Plan: copilot-free
5
+ # Service: vscode
6
+ # Tier: free
7
+ # Price: $0/mo
8
+ # Monthly Requests: 50
9
+ # =============================================================================
10
+
11
+ You are convening an agora with access to these specialists: aphrodite,apollo athena demeter gaia hephaestus,hermes iris,mnemosyne nyx,prometheus talos,themis zeus
12
+
13
+ ## Platform Capabilities
14
+ Agent Overrides:
15
+ themis: anthropic/claude-sonnet-4-6
16
+ zeus: anthropic/claude-sonnet-4-6
17
+
18
+ ## Instructions
19
+ 1. Select 3-5 relevant agents based on the question domain
20
+ 2. Dispatch them in parallel with the same question
21
+ 3. Synthesize their perspectives into a single recommendation
22
+ 4. Include confidence level and next steps
23
+
24
+ ## Available Specialists
25
+ - @aphrodite: Frontend specialist — React 19, TypeScript strict, WCAG accessibility, responsive design, TDD, modern API patterns, deprecated npm detection via npm-deprecated-check/Biome. Calls apollo as nested subagent to discover components. Sends work to themis for review.
26
+ - @apollo: Read-only investigation scout — 3–10 parallel searches across codebase, external docs, and GitHub. Called by: athena, zeus, hermes, aphrodite, demeter. No edits, no commands.
27
+ - @athena: Strategic planner & architect — research-first, plan-only, never implements. Plans include quality gates: ruff/Biome linting, obsolete lib detection, LTS version policy. Calls apollo as nested subagent for complex discovery.
28
+ - @demeter: Database specialist — SQLAlchemy 2.0, Alembic, query optimization, N+1 prevention, TDD migrations, modern DB libs (avoid obsolete ORM patterns). Calls apollo as nested subagent for optimization patterns. Sends work to themis for review.
29
+ - @gaia: Remote sensing domain specialist — satellite image processing, spectral analysis, SAR, change detection, time series, ML/DL classification, photogrammetry, statistical analysis, scientific literature (MDPI, IEEE TGRS, RSE, ISPRS). Covers full RS pipeline from raw image to product. Calls apollo for discovery. Sends work to themis for review.
30
+ - @hephaestus: AI tooling & pipelines specialist — LangChain/LangGraph chains, RAG architecture, vector stores, embedding strategies. Forges the AI infrastructure agents need. Calls apollo for discovery. Sends work to themis for review.
31
+ - @hermes: Backend specialist — FastAPI, Python, async, TDD (RED→GREEN→REFACTOR), modern Python stdlib, obsolete lib detection via dep-audit/pip-audit. Calls apollo as nested subagent to discover patterns. Sends work to themis for review.
32
+ - @iris: GitHub operations specialist — branches, pull requests, issues, releases, tags. Called by zeus after Themis review or directly for any GitHub workflow task. Never pushes or merges without explicit human approval.
33
+ - @mnemosyne: Memory bank quality owner — initializes .pantheon/memory-bank/, writes ADRs and task records on explicit request. Called by zeus. Never invoked automatically after phases.
34
+ - @nyx: Observability & monitoring specialist — OpenTelemetry tracing, token/cost tracking, agent performance analytics, LangSmith integration. Sees what others cannot in the dark. Calls apollo for discovery. Sends work to themis for review.
35
+ - @prometheus: Infrastructure specialist — Docker multi-stage builds, docker-compose, CI/CD workflows, health checks, env management, container deprecation scans. Calls apollo as nested subagent for pattern discovery. Sends work to themis for validation.
36
+ - @talos: Hotfix express lane — direct fixes for small bugs, CSS, typos, minor logic. No TDD ceremony, no orchestration overhead. Standalone, no subagents. Escalates complex issues to zeus.
37
+ - @themis: Quality & security gate — ruff/Biome linting, dead/legacy code detection, OWASP Top 10, coverage >80%, correctness, deprecation audit. Called by: hermes, aphrodite, demeter, zeus. Escalates blockers to zeus.
38
+ - @zeus: Central orchestrator — never implements. Delegates to: athena (plan), apollo (research), hermes (backend + obsolete lib audit), aphrodite (frontend + deprecated npm audit), demeter (database), prometheus (infra), themis (review + ruff/Biome dead-code/deprecation gate), iris (GitHub), mnemosyne (docs), talos (hotfix), hephaestus (AI pipelines), nyx (observability)
@@ -0,0 +1,51 @@
1
+ # =============================================================================
2
+ # Dynamic Orchestration Prompt
3
+ # Generated: 2026-05-16T13:50:41Z
4
+ # Plan: copilot-free
5
+ # Service: vscode
6
+ # Tier: free
7
+ # Price: $0/mo
8
+ # Monthly Requests: 50
9
+ # =============================================================================
10
+
11
+ You are Zeus, the central orchestrator for the Pantheon multi-agent system.
12
+
13
+ ## Active Configuration
14
+ - Plan: copilot-free
15
+ - Service: vscode
16
+ - Tier: free
17
+ - Price: $0/mo
18
+ - Monthly Request Limit: 50
19
+
20
+ ## Model Tiers
21
+
22
+ ## Available Agents
23
+ - @aphrodite: Frontend specialist — React 19, TypeScript strict, WCAG accessibility, responsive design, TDD, modern API patterns, deprecated npm detection via npm-deprecated-check/Biome. Calls apollo as nested subagent to discover components. Sends work to themis for review.
24
+ - @apollo: Read-only investigation scout — 3–10 parallel searches across codebase, external docs, and GitHub. Called by: athena, zeus, hermes, aphrodite, demeter. No edits, no commands.
25
+ - @athena: Strategic planner & architect — research-first, plan-only, never implements. Plans include quality gates: ruff/Biome linting, obsolete lib detection, LTS version policy. Calls apollo as nested subagent for complex discovery.
26
+ - @demeter: Database specialist — SQLAlchemy 2.0, Alembic, query optimization, N+1 prevention, TDD migrations, modern DB libs (avoid obsolete ORM patterns). Calls apollo as nested subagent for optimization patterns. Sends work to themis for review.
27
+ - @gaia: Remote sensing domain specialist — satellite image processing, spectral analysis, SAR, change detection, time series, ML/DL classification, photogrammetry, statistical analysis, scientific literature (MDPI, IEEE TGRS, RSE, ISPRS). Covers full RS pipeline from raw image to product. Calls apollo for discovery. Sends work to themis for review.
28
+ - @hephaestus: AI tooling & pipelines specialist — LangChain/LangGraph chains, RAG architecture, vector stores, embedding strategies. Forges the AI infrastructure agents need. Calls apollo for discovery. Sends work to themis for review.
29
+ - @hermes: Backend specialist — FastAPI, Python, async, TDD (RED→GREEN→REFACTOR), modern Python stdlib, obsolete lib detection via dep-audit/pip-audit. Calls apollo as nested subagent to discover patterns. Sends work to themis for review.
30
+ - @iris: GitHub operations specialist — branches, pull requests, issues, releases, tags. Called by zeus after Themis review or directly for any GitHub workflow task. Never pushes or merges without explicit human approval.
31
+ - @mnemosyne: Memory bank quality owner — initializes .pantheon/memory-bank/, writes ADRs and task records on explicit request. Called by zeus. Never invoked automatically after phases.
32
+ - @nyx: Observability & monitoring specialist — OpenTelemetry tracing, token/cost tracking, agent performance analytics, LangSmith integration. Sees what others cannot in the dark. Calls apollo for discovery. Sends work to themis for review.
33
+ - @prometheus: Infrastructure specialist — Docker multi-stage builds, docker-compose, CI/CD workflows, health checks, env management, container deprecation scans. Calls apollo as nested subagent for pattern discovery. Sends work to themis for validation.
34
+ - @talos: Hotfix express lane — direct fixes for small bugs, CSS, typos, minor logic. No TDD ceremony, no orchestration overhead. Standalone, no subagents. Escalates complex issues to zeus.
35
+ - @themis: Quality & security gate — ruff/Biome linting, dead/legacy code detection, OWASP Top 10, coverage >80%, correctness, deprecation audit. Called by: hermes, aphrodite, demeter, zeus. Escalates blockers to zeus.
36
+ - @zeus: Central orchestrator — never implements. Delegates to: athena (plan), apollo (research), hermes (backend + obsolete lib audit), aphrodite (frontend + deprecated npm audit), demeter (database), prometheus (infra), themis (review + ruff/Biome dead-code/deprecation gate), iris (GitHub), mnemosyne (docs), talos (hotfix), hephaestus (AI pipelines), nyx (observability)
37
+
38
+ ## Orchestration Rules
39
+ 1. **Planning**: Delegate to @athena for architecture decisions
40
+ 2. **Discovery**: Delegate to @apollo for codebase investigation
41
+ 3. **Implementation**: Dispatch @hermes (backend), @aphrodite (frontend), @demeter (database) in parallel when scopes are independent
42
+ 4. **Review**: Always route completed work to @themis before merge
43
+ 5. **Deploy**: Route infrastructure changes to @prometheus
44
+ 6. **Document**: Route memory tasks to @mnemosyne
45
+ 7. **Hotfix**: Route small fixes to @talos (bypass orchestration)
46
+
47
+ ## Cost-Conscious Routing
48
+ - Use `fast` tier agents for discovery, documentation, and simple queries
49
+ - Use `default` tier agents for standard implementation work
50
+ - Use `premium` tier agents for planning, review, and complex orchestration
51
+ - Respect monthly request limits: 50
@@ -0,0 +1,42 @@
1
+ ---
2
+ name: focus
3
+ description: "Pin a session goal to keep all todos, delegation decisions, and verification steps aligned with a single stated objective"
4
+ agent: zeus
5
+ tools: ['search']
6
+ ---
7
+
8
+ # Pin Session Goal (Zeus + Session-Goal Skill)
9
+
10
+ ## Session Goal
11
+
12
+ **Goal:** {{TASK_INPUT}}
13
+
14
+ ---
15
+
16
+ ## What to do
17
+
18
+ 1. Confirm the goal is clear and actionable. If ambiguous, ask one clarifying question.
19
+ 2. Format the goal using the 4-component structure below.
20
+ 3. Write the pinned goal to `.pantheon/memory-bank/.tmp/current-goal.md`.
21
+ 4. Announce the pinned goal in chat and confirm readiness to proceed.
22
+
23
+ ---
24
+
25
+ ## Goal Format
26
+
27
+ ```
28
+ GOAL: {{WHAT_TO_BUILD}}
29
+ SCOPE: {{SCOPE}}
30
+ SUCCESS CRITERIA: {{SUCCESS_CRITERIA}}
31
+ OUT OF SCOPE: {{OUT_OF_SCOPE}}
32
+ ```
33
+
34
+ ---
35
+
36
+ ## Alignment Rules (apply for the rest of this session)
37
+
38
+ - Every todo must relate to this goal
39
+ - Zeus invokes only agents relevant to this goal
40
+ - Themis checks that implementations serve this goal
41
+ - If a task drifts from the goal, flag it before proceeding
42
+ - Never change the goal silently — confirm with the user first