geneseed 5.6.0 → 6.0.1

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 (114) hide show
  1. package/CHANGELOG.md +30 -0
  2. package/README.md +2 -2
  3. package/SHIPPED.md +2 -2
  4. package/adapters/opencode/README.md +7 -4
  5. package/docs/concepts/agents.md +10 -0
  6. package/docs/concepts/exclusions.md +1 -1
  7. package/docs/concepts/foreman-mode.md +4 -2
  8. package/docs/concepts/kits.md +20 -2
  9. package/docs/concepts/loops.md +5 -0
  10. package/docs/concepts/skills.md +1 -1
  11. package/docs/concepts/superpowers-kit.md +79 -0
  12. package/docs/extending.md +11 -4
  13. package/docs/guides/install-scripted.md +1 -1
  14. package/docs/reference/cli.md +2 -0
  15. package/docs/reference/profiles.md +10 -5
  16. package/harness.config.json +1 -1
  17. package/js/README.md +3 -0
  18. package/js/build/args.mjs +4 -2
  19. package/js/build/driver.mjs +8 -8
  20. package/js/build/emit-claude.mjs +5 -3
  21. package/js/build/emit-opencode.mjs +2 -2
  22. package/js/build/generate.mjs +1 -1
  23. package/js/build/profile.mjs +15 -7
  24. package/js/build/render.mjs +81 -10
  25. package/js/build/source.mjs +75 -1
  26. package/js/cli-table.json +17 -2
  27. package/js/hosts/installs.mjs +50 -4
  28. package/js/hosts/native.mjs +125 -9
  29. package/js/hosts/opencode.mjs +13 -15
  30. package/js/inspect/checks-authoring.mjs +24 -1
  31. package/js/inspect/checks-build.mjs +43 -2
  32. package/js/inspect/checks-counts.mjs +72 -1
  33. package/js/inspect/diff.mjs +17 -3
  34. package/js/inspect/doctor.mjs +23 -2
  35. package/js/inspect/status.mjs +5 -1
  36. package/js/inspect/validate.mjs +1 -1
  37. package/js/loop/catalog.mjs +10 -0
  38. package/js/maintain/kit-migration.mjs +109 -0
  39. package/js/maintain/plugin-detect.mjs +150 -0
  40. package/js/maintain/profiles.mjs +46 -5
  41. package/js/maintain/setup.mjs +45 -11
  42. package/js/maintain/update.mjs +1 -1
  43. package/js/web/actions.mjs +28 -7
  44. package/js/web/agent-models-api.mjs +172 -0
  45. package/js/web/api.mjs +3 -1
  46. package/js/web/profiles-api.mjs +11 -7
  47. package/js/web/routes.mjs +5 -1
  48. package/package.json +1 -1
  49. package/profiles/autonomous.json +1 -0
  50. package/profiles/corporate.json +1 -1
  51. package/profiles/mentor.json +1 -0
  52. package/profiles/minimal.json +2 -1
  53. package/profiles/starter.json +1 -1
  54. package/src/AGENT.md.tmpl +1 -6
  55. package/src/agents/_template.md +4 -0
  56. package/src/agents/advocate.md +1 -0
  57. package/src/agents/architect.md +1 -0
  58. package/src/agents/developer.md +12 -4
  59. package/src/agents/empiricist.md +1 -0
  60. package/src/agents/explorer.md +1 -0
  61. package/src/agents/framer.md +1 -0
  62. package/src/agents/historian.md +1 -0
  63. package/src/agents/operator.md +1 -0
  64. package/src/agents/performance.md +2 -1
  65. package/src/agents/pragmatist.md +1 -0
  66. package/src/agents/react-reviewer.md +1 -0
  67. package/src/agents/researcher.md +1 -0
  68. package/src/agents/reviewer.md +1 -0
  69. package/src/agents/security.md +1 -0
  70. package/src/agents/skeptic.md +1 -0
  71. package/src/agents/spring-reviewer.md +1 -0
  72. package/src/agents/spring-security.md +1 -0
  73. package/src/agents/steward.md +1 -0
  74. package/src/agents/user-advocate.md +1 -0
  75. package/src/agents/visionary.md +1 -0
  76. package/src/bricks/ci-fix.md +12 -2
  77. package/src/kits/superpowers/missing.md +3 -0
  78. package/src/kits/superpowers/provider.md +7 -0
  79. package/src/kits/superpowers/rules.md +6 -0
  80. package/src/kits/superpowers.json +45 -0
  81. package/src/modes/foreman.md +18 -12
  82. package/src/skills/parallel-agents.md +2 -2
  83. package/src/skills/spring-boot/cloud.md +8 -2
  84. package/src/skills/spring-boot.md +3 -1
  85. package/src/skills/spring-testing.md +3 -3
  86. package/src/upstreams.json +1 -1
  87. package/web/dist/assets/{ActiveLoops-DOJCPPPc.js → ActiveLoops-6p44eD86.js} +1 -1
  88. package/web/dist/assets/{Activity-bUSPM3Dl.js → Activity-CIgrq5R1.js} +1 -1
  89. package/web/dist/assets/{ActivityDetail-AHRoXeFN.js → ActivityDetail-Brqv1u4f.js} +1 -1
  90. package/web/dist/assets/{Bonsai-B5XY9pid.js → Bonsai-8NtlFrPO.js} +1 -1
  91. package/web/dist/assets/{CopyButton-C0wwRR9H.js → CopyButton-BHqKgdjJ.js} +1 -1
  92. package/web/dist/assets/{Dashboard-DPAA96EI.js → Dashboard-CoveJ_-U.js} +2 -2
  93. package/web/dist/assets/{Docs-CpGDq165.js → Docs-C1jzrFnD.js} +1 -1
  94. package/web/dist/assets/{Factories-cNhETa5Q.js → Factories-D5pLnt5S.js} +1 -1
  95. package/web/dist/assets/{GreenhouseView-DO3ADmT9.js → GreenhouseView-ofAQ-V5h.js} +1 -1
  96. package/web/dist/assets/Installs-DpWgyT5j.js +3 -0
  97. package/web/dist/assets/{JournalView--W-Lw_V9.js → JournalView-CsuN3_B-.js} +1 -1
  98. package/web/dist/assets/{Kits-ChfehxY6.js → Kits-P4xspDuN.js} +1 -1
  99. package/web/dist/assets/{Laws-CNxYmV36.js → Laws-DK3OwGDl.js} +1 -1
  100. package/web/dist/assets/{LibRows-CiPvsRYB.js → LibRows-CGO25X3D.js} +1 -1
  101. package/web/dist/assets/Library-e-94YED6.js +3 -0
  102. package/web/dist/assets/{Loops-DAblKkyq.js → Loops-C9DSBj-A.js} +1 -1
  103. package/web/dist/assets/{OperatorHudView-QtSaSyYm.js → OperatorHudView-Cc8mlViW.js} +1 -1
  104. package/web/dist/assets/{Personal-DxSGv638.js → Personal-DVh2jYPO.js} +1 -1
  105. package/web/dist/assets/{Settings-CgTbhvRv.js → Settings-CknBI-Fx.js} +1 -1
  106. package/web/dist/assets/{Tabs-Cf4X1hdz.js → Tabs-C30JrkJH.js} +1 -1
  107. package/web/dist/assets/api-DxwG8y94.js +1 -0
  108. package/web/dist/assets/index-CUJ8AaAF.css +1 -0
  109. package/web/dist/assets/{index-B_pTwZra.js → index-MB7E-va_.js} +3 -3
  110. package/web/dist/index.html +3 -3
  111. package/web/dist/assets/Installs-D3dDyq10.js +0 -3
  112. package/web/dist/assets/Library-BxQvbwVU.js +0 -3
  113. package/web/dist/assets/api-CZ-x0iWG.js +0 -1
  114. package/web/dist/assets/index-CzfImiYb.css +0 -1
@@ -0,0 +1,172 @@
1
+ /**
2
+ * The console's view of the sub-agent models — `GET`/`POST /api/agent-models`.
3
+ *
4
+ * What it reads and writes is `agent-overrides.json` of ONE Claude Code install: the `tiers` map
5
+ * (`standard`, `light`) and each agent's `agents.<name>.model` pin. The precedence (pin > map >
6
+ * the tier's default > inherit) is `claudeAgentModel` in `js/hosts/native.mjs`, the same function
7
+ * the emit writes agent frontmatter with, so this page can never promise a model the build will
8
+ * not write. The value validation is `overrideProblem` and the map reader `loadTierMap`, there too.
9
+ *
10
+ * THE FILE IS THE USER'S. A write changes only the keys it names, key by key, in the order the
11
+ * file already has; it refuses a file that does not parse, or a `tiers`/`agents` that is not an
12
+ * object, rather than replace what it cannot read. Nothing here rebuilds: the model is written
13
+ * into the agent files at emit, so a save answers `rebuild: true` and the page offers the rebuild
14
+ * through the install action.
15
+ *
16
+ * A tier's value: a string sets it (`''` is the user choosing inherit, as `loadTierMap` reads it),
17
+ * `null` removes the key (back to the tier's default). A pin: a string sets it, `''` or `null`
18
+ * removes `model` and leaves the rest of the agent's entry alone.
19
+ *
20
+ * Pins and the map are Claude Code's: OpenCode records pins but this page does not edit them, and
21
+ * a host that is not Claude Code answers each agent's tier and no model.
22
+ */
23
+ import path from 'node:path';
24
+
25
+ import { renderAll } from '../build/render.mjs';
26
+ import { loadKits, makeCfg } from '../build/source.mjs';
27
+ import { EMIT_HOST_SCOPE, tiersOfDir } from '../hosts/installs.mjs';
28
+ import {
29
+ CLAUDE_TIER_DEFAULTS, claudeAgentModel, loadAgentOverrides, loadTierMap, overrideProblem,
30
+ } from '../hosts/native.mjs';
31
+ import { isFile, readText, writeTextAtomic } from '../lib/fs.mjs';
32
+ import { formatValue, isDict, isTruthy, jsonDumpsIndent, parseJson } from '../lib/json.mjs';
33
+ import { bget } from './actions.mjs';
34
+
35
+ const FILE = 'agent-overrides.json';
36
+ /** The tiers that take a model; `strong` is the session's own and has no key. */
37
+ const MAP_TIERS = ['standard', 'light'];
38
+ const MAX_MODEL = 200;
39
+
40
+ let tierTable = null;
41
+ /**
42
+ * `Map<agent name, 'strong'|'standard'|'light'>` off the SOURCE: the render strips an agent's
43
+ * `<!-- tier: -->` line from every output, so an installed file cannot say it. Every kit's agents
44
+ * are included, so a kit's seat is not read as `standard` by default. Built once per process.
45
+ */
46
+ export function agentTiers() {
47
+ if (tierTable) return tierTable;
48
+ const out = new Map();
49
+ const { items } = renderAll(makeCfg({ kits: [...loadKits().keys()] }), 'neutral');
50
+ for (const { rel, tier = 'standard' } of items) {
51
+ const m = /^agents\/([^_/][^/]*)\.md$/.exec(rel);
52
+ if (m) out.set(m[1], tier);
53
+ }
54
+ tierTable = out;
55
+ return out;
56
+ }
57
+
58
+ /** `{ exists, data, problem }`: the file parsed as an object, else why it cannot be edited. */
59
+ function readOverrides(dir) {
60
+ const file = path.join(dir, FILE);
61
+ if (!isFile(file)) return { file, exists: false, data: {}, problem: null };
62
+ let data;
63
+ try { data = parseJson(readText(file)); } catch {
64
+ return { file, exists: true, data: null, problem: `${FILE} is not valid JSON (comments are not allowed) — fix it by hand` };
65
+ }
66
+ if (!isDict(data)) return { file, exists: true, data: null, problem: `${FILE} is not a JSON object — fix it by hand` };
67
+ return { file, exists: true, data, problem: null };
68
+ }
69
+
70
+ /** The view's host and the install's directory; `claude` false for every host but Claude Code. */
71
+ function viewOf(state) {
72
+ const [host] = EMIT_HOST_SCOPE.get(state.emit || '') ?? ['opencode'];
73
+ return { host, claude: host === 'claude', dir: state.target };
74
+ }
75
+
76
+ export function apiAgentModels(state, env = process.env) {
77
+ const { host, claude, dir } = viewOf(state);
78
+ const tiers = tiersOfDir(state.root ?? dir, host);
79
+ const known = agentTiers();
80
+ const names = state.inventory.agents.map((a) => a.name);
81
+ if (!claude) {
82
+ return { host, claude, tiers, agents: names.map((name) => ({ name, tier: known.get(name) ?? null })) };
83
+ }
84
+ const { file, problem } = readOverrides(dir);
85
+ // The loaders take a no-op reporter: a console read prints nothing on stderr.
86
+ const quiet = () => {};
87
+ const map = problem ? {} : loadTierMap(dir, quiet);
88
+ const pins = problem ? {} : loadAgentOverrides(dir, quiet);
89
+ const on = tiers === 'auto' ? { map, env } : null;
90
+ const warnings = new Set();
91
+ const agents = names.map((name) => {
92
+ const tier = known.get(name) ?? null;
93
+ const pin = Object.hasOwn(pins, name) && isTruthy(pins[name]?.model) ? formatValue(pins[name].model) : null;
94
+ // A personal agent (no shipped source) is not Geneseed's to tier: only its pin counts.
95
+ const r = claudeAgentModel(tier ?? 'none', pin, tier ? on : null);
96
+ // The alias warning is the tier map's, whether or not the routing is installed yet: the
97
+ // console's editor shows with a picked `auto`, before the rebuild that turns it on.
98
+ const w = tier ? claudeAgentModel(tier, pin, { map, env }).warning : null;
99
+ if (w) warnings.add(w);
100
+ return { name, tier, pin, model: r.model, from: r.from };
101
+ });
102
+ return {
103
+ host, claude, tiers, file: { path: file, problem },
104
+ map: Object.fromEntries(Object.entries(map).map(([k, v]) => [k, formatValue(v)])),
105
+ defaults: { ...CLAUDE_TIER_DEFAULTS },
106
+ warnings: [...warnings], agents,
107
+ };
108
+ }
109
+
110
+ /**
111
+ * Apply one edit to the parsed file in place; the problems found, none applied when there are some.
112
+ * `edit` is `{ tiers?: {standard?, light?}, agents?: {<name>: model} }`, `names` the installed agents.
113
+ */
114
+ export function editOverrides(data, edit, names) {
115
+ const problems = [];
116
+ const asModel = (where, v) => {
117
+ if (v === null) return null;
118
+ if (typeof v !== 'string') { problems.push(`${where}: a model name, or null`); return undefined; }
119
+ const s = v.trim();
120
+ const why = overrideProblem(s);
121
+ if (why) { problems.push(`${where} ${why}`); return undefined; }
122
+ if (s.length > MAX_MODEL) { problems.push(`${where} is longer than ${MAX_MODEL} characters`); return undefined; }
123
+ return s;
124
+ };
125
+ const tiers = isDict(edit.tiers) ? edit.tiers : {};
126
+ const agents = isDict(edit.agents) ? edit.agents : {};
127
+ const tierOps = [];
128
+ for (const [k, v] of Object.entries(tiers)) {
129
+ if (!MAP_TIERS.includes(k)) { problems.push(`tiers.${k} is not a tier that takes a model (${MAP_TIERS.join(', ')})`); continue; }
130
+ const m = asModel(`tiers.${k}`, v);
131
+ if (m !== undefined) tierOps.push([k, m]);
132
+ }
133
+ const pinOps = [];
134
+ for (const [name, v] of Object.entries(agents)) {
135
+ if (name === '__proto__' || !names.includes(name)) { problems.push(`agents.${name} is not an installed agent`); continue; }
136
+ const m = asModel(`agents.${name}.model`, v);
137
+ if (m !== undefined) pinOps.push([name, m === '' ? null : m]);
138
+ }
139
+ if (!tierOps.length && !pinOps.length && !problems.length) problems.push('nothing to save');
140
+ if (tierOps.length && Object.hasOwn(data, 'tiers') && !isDict(data.tiers)) problems.push(`${FILE}: "tiers" is not an object — fix it by hand`);
141
+ if (pinOps.length && Object.hasOwn(data, 'agents') && !isDict(data.agents)) problems.push(`${FILE}: "agents" is not an object — fix it by hand`);
142
+ if (problems.length) return problems;
143
+
144
+ for (const [k, m] of tierOps) {
145
+ if (m === null) { if (isDict(data.tiers)) delete data.tiers[k]; continue; }
146
+ if (!isDict(data.tiers)) data.tiers = {};
147
+ data.tiers[k] = m;
148
+ }
149
+ for (const [name, m] of pinOps) {
150
+ if (m === null) {
151
+ if (isDict(data.agents) && Object.hasOwn(data.agents, name) && isDict(data.agents[name])) delete data.agents[name].model;
152
+ continue;
153
+ }
154
+ if (!isDict(data.agents)) data.agents = {};
155
+ if (!Object.hasOwn(data.agents, name) || !isDict(data.agents[name])) data.agents[name] = {};
156
+ data.agents[name].model = m;
157
+ }
158
+ return [];
159
+ }
160
+
161
+ /** `POST /api/agent-models`: `ok: false` (a 409) when nothing was written, with the reason. */
162
+ export function apiAgentModelsSave(state, body) {
163
+ const { claude, dir } = viewOf(state);
164
+ if (!claude) return { ok: false, error: 'Models are set on Claude Code installs only' };
165
+ const { file, data, problem } = readOverrides(dir);
166
+ if (problem) return { ok: false, error: problem };
167
+ const names = state.inventory.agents.map((a) => a.name);
168
+ const problems = editOverrides(data, { tiers: bget(body, 'tiers'), agents: bget(body, 'agents') }, names);
169
+ if (problems.length) return { ok: false, error: problems.join('; ') };
170
+ writeTextAtomic(file, `${jsonDumpsIndent(data, { ensureAscii: false })}\n`);
171
+ return { ok: true, rebuild: true, ...apiAgentModels(state) };
172
+ }
package/js/web/api.mjs CHANGED
@@ -24,7 +24,7 @@ import { factCount } from '../hosts/memory-files.mjs';
24
24
  import { autostartState } from '../hosts/shim.mjs';
25
25
  import {
26
26
  footprintOfDir, gatesOfDir, installedDefaults, installState, installTargets, leftoversOf, modeOfDir,
27
- trustOfDir, sharedTierOfDir, contextWindowOfDir, strandedGatesOf,
27
+ trustOfDir, sharedTierOfDir, contextWindowOfDir, tiersOfDir, strandedGatesOf,
28
28
  postureOfDir, readJsonMaybe, readMaybe, themeOfDir,
29
29
  } from '../hosts/installs.mjs';
30
30
  import { PRESETS } from '../loop/score.mjs';
@@ -247,6 +247,8 @@ export function apiInstalls(state) {
247
247
  // The context window the install is built for (`--context-window`); it writes settings on
248
248
  // Claude Code only. A row not installed yet offers the machine's, as a new install takes it.
249
249
  contextWindow: installState(root, host, scope) === 'absent' ? machineWindow() : contextWindowOfDir(root, host),
250
+ // The tier routing the install was built with (`--tiers`); off for a row not installed yet.
251
+ tiers: installState(root, host, scope) === 'absent' ? 'off' : tiersOfDir(root, host),
250
252
  // Files here that are the user's where Geneseed ships one, and retired leftovers: what a
251
253
  // forced rebuild would replace or remove (`leftoversOf`). Read off the emit's own record.
252
254
  leftovers: leftoversOf(root, host, scope),
@@ -15,10 +15,10 @@ import path from 'node:path';
15
15
  import { ROOT } from '../build/source.mjs';
16
16
  import { installProfile } from '../build/profile.mjs';
17
17
  import { HOSTS, expanduser } from '../hosts/hosts.mjs';
18
- import { installState, installTargets } from '../hosts/installs.mjs';
18
+ import { installState, installTargets, tiersOfDir } from '../hosts/installs.mjs';
19
19
  import {
20
20
  BASE_PROFILE, isGitClone, listProfiles, loadProfile, profileBuildArgs, profileDiff, profileDirs,
21
- profileFromInstall, profileHosts,
21
+ profileFromInstall, profileHosts, profileKits,
22
22
  } from '../maintain/profiles.mjs';
23
23
  import { checkProjectFolder, globalLayering, markerCollision } from '../maintain/profile-scope.mjs';
24
24
  import { hostTools, profileChanges } from '../maintain/setup.mjs';
@@ -50,9 +50,13 @@ function activeTargets() {
50
50
  * `starter`, which lists four hosts, could match no install). `sharedTier` counts only on a
51
51
  * project install: a global apply ignores it (D37), so a global install cannot differ on it.
52
52
  */
53
- export function followsProfile(installP, scope, p) {
53
+ export function followsProfile(installP, scope, p, root = null) {
54
54
  const skip = new Set(['hosts', 'login', ...(scope === 'global' ? ['sharedTier'] : [])]);
55
- return profileDiff(installP, p).every((d) => skip.has(d.key));
55
+ // A kit chosen by detection is the kit an apply would have built for the install's own host: an
56
+ // install has the kits it has, never a kit "by detection", so compare the resolved list.
57
+ const resolved = p.kitsByDetection?.length
58
+ ? { ...p, kits: profileKits(p, installP.hosts?.[0] ?? '', scope === 'project' ? root : null), kitsByDetection: [] } : p;
59
+ return profileDiff(installP, resolved).every((d) => skip.has(d.key));
56
60
  }
57
61
 
58
62
  /** The hosts row of the tab: `{ host, label, present }` in HOSTS order. */
@@ -79,7 +83,7 @@ function previewRows(p, project, hosts) {
79
83
  // (`markerCollision`): the row says so and the panel does not offer it.
80
84
  const refused = project ? markerCollision(project, h.host) : null;
81
85
  const changes = h.present && !refused && state === 'active'
82
- ? profileChanges(profileBuildArgs(p, h.host, project)) : [];
86
+ ? profileChanges(profileBuildArgs(p, h.host, project, root ? tiersOfDir(root, h.host) : null)) : [];
83
87
  return {
84
88
  host: h.host, label: h.label, present: h.present, state, root: root ? String(root) : null,
85
89
  changes, layering: project && !refused ? globalLayering(h.host) : null, refused,
@@ -114,14 +118,14 @@ export function apiProfiles(state, url, { src = {}, tools = hostTools } = {}) {
114
118
  ...head, description: p.description ?? '', profile: p,
115
119
  gatesOff: Object.keys(p.gates ?? {}),
116
120
  diff: base.ok ? profileDiff(p, base.profile) : [],
117
- matches: readBacks.filter((rb) => followsProfile(rb.profile, rb.scope, p))
121
+ matches: readBacks.filter((rb) => followsProfile(rb.profile, rb.scope, p, rb.path))
118
122
  .map(({ host, scope, path: at }) => ({ host, scope, path: at })),
119
123
  };
120
124
  });
121
125
  const loaded = profiles.filter((x) => x.profile);
122
126
  const installs = readBacks.map((rb) => ({
123
127
  host: rb.host, scope: rb.scope, path: rb.path,
124
- matches: loaded.filter((x) => followsProfile(rb.profile, rb.scope, x.profile)).map((x) => x.name),
128
+ matches: loaded.filter((x) => followsProfile(rb.profile, rb.scope, x.profile, rb.path)).map((x) => x.name),
125
129
  diff: base.ok ? profileDiff(rb.profile, base.profile) : [],
126
130
  }));
127
131
  const hosts = hostRows(tools);
package/js/web/routes.mjs CHANGED
@@ -28,6 +28,7 @@ import { apiUsage } from './usage-api.mjs';
28
28
  import { apiProfiles } from './profiles-api.mjs';
29
29
  import { fleetSort } from './state.mjs';
30
30
  import { apiGates } from './gates-api.mjs';
31
+ import { apiAgentModels, apiAgentModelsSave } from './agent-models-api.mjs';
31
32
  import { migratePreview } from '../maintain/migrate.mjs';
32
33
  import { apiProfile, apiProfileSave, apiRules, apiRulesDecide, apiRulesMutate, apiRulesPromote } from './user-files.mjs';
33
34
 
@@ -64,6 +65,8 @@ export const POST_ROUTES = new Map([
64
65
  // ever consulted — the same shape as `/api/memory/delete` and `/api/activity` above.
65
66
  ['/api/loops/preset', [apiLoopsPresetMutate, false]],
66
67
  ['/api/autostart', [apiAutostart, true]],
68
+ // The `tiers` map and the per-agent model pins of one Claude Code install's agent-overrides.json.
69
+ ['/api/agent-models', [apiAgentModelsSave, true]],
67
70
  ]);
68
71
 
69
72
  /**
@@ -74,7 +77,7 @@ export const POST_ROUTES = new Map([
74
77
  * are the only actions that read the request's install.
75
78
  */
76
79
  export const VIEW_POSTS = new Set(['/api/activity', '/api/memory/delete', '/api/rules',
77
- '/api/rules/promote', '/api/rules/decide', '/api/profile', '/api/actions/restore', '/api/actions/build']);
80
+ '/api/rules/promote', '/api/rules/decide', '/api/profile', '/api/actions/restore', '/api/actions/build', '/api/agent-models']);
78
81
 
79
82
  /**
80
83
  * The GETs that answer for the whole machine, not one install: they IGNORE a `?host=&path=`
@@ -229,6 +232,7 @@ export const STATE_ROUTES = {
229
232
  // answers — a daemon whose POST worked and whose GET 501'd would have no way to obtain one.
230
233
  '/api/rules': apiRules,
231
234
  '/api/loadout': (state) => apiLoadout(state),
235
+ '/api/agent-models': (state) => apiAgentModels(state),
232
236
  '/api/mcp': apiMcp,
233
237
  '/api/loops': apiLoops,
234
238
  '/api/loops/active': apiLoopsActive,
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "//": "Every field here is argued and gated in tests/unit/package_manifest.test.mjs. `files`, `bin`, `exports` and `engines` each have a test that fails when reality drifts from this file; `version` mirrors harness.config.json and is gated too. Read that docblock before editing this one.",
3
3
  "name": "geneseed",
4
- "version": "5.6.0",
4
+ "version": "6.0.1",
5
5
  "description": "A portable, theme-able harness you implant once and use everywhere to grow a disciplined AI coding agent.",
6
6
  "keywords": [
7
7
  "ai",
@@ -8,6 +8,7 @@
8
8
  "trust": "aggressive",
9
9
  "doctrines": "all",
10
10
  "kits": [],
11
+ "kitsByDetection": ["superpowers"],
11
12
  "contextWindow": "1m",
12
13
  "gates": { "ownershipGate": false },
13
14
  "login": true
@@ -7,7 +7,7 @@
7
7
  "mode": "direct",
8
8
  "trust": "prudent",
9
9
  "doctrines": "all",
10
- "kits": [],
10
+ "kits": ["superpowers"],
11
11
  "contextWindow": "200k",
12
12
  "gates": { "ownershipGate": false, "ownBranchPass": false },
13
13
  "login": false
@@ -8,6 +8,7 @@
8
8
  "trust": "prudent",
9
9
  "doctrines": "all",
10
10
  "kits": [],
11
+ "kitsByDetection": ["superpowers"],
11
12
  "contextWindow": "200k",
12
13
  "login": true
13
14
  }
@@ -7,8 +7,9 @@
7
7
  "mode": "direct",
8
8
  "trust": "balanced",
9
9
  "doctrines": ["craft", "rigor"],
10
- "excludeSkills": ["brick-forge", "bruno", "daydream", "factory", "factory-forge", "forge-mcp", "prototype", "react-view-transitions"],
10
+ "excludeSkills": ["brick-forge", "bruno", "daydream", "factory", "factory-forge", "forge-mcp", "react-view-transitions"],
11
11
  "kits": [],
12
+ "kitsByDetection": ["superpowers"],
12
13
  "contextWindow": "200k",
13
14
  "gates": { "ownershipGate": false },
14
15
  "login": true
@@ -7,7 +7,7 @@
7
7
  "mode": "direct",
8
8
  "trust": "balanced",
9
9
  "doctrines": "all",
10
- "kits": [],
10
+ "kits": ["superpowers"],
11
11
  "contextWindow": "200k",
12
12
  "gates": { "ownershipGate": false },
13
13
  "login": true
package/src/AGENT.md.tmpl CHANGED
@@ -232,14 +232,11 @@ the ten council seats are convened by the [council {{SKILL}}]({{DIR_SKILLS}}/cou
232
232
  <!-- CATALOG:begin skills -->
233
233
  | {{SKILL}} | Trigger |
234
234
  | --- | --- |
235
- | [brainstorm]({{DIR_SKILLS}}/brainstorm.md) | a new feature or project with no design yet, or whose goal, scope, or success criteria need pinning down first |
236
- | [plan]({{DIR_SKILLS}}/plan.md) | a task has more than a couple of steps |
237
235
  | [codebase-design]({{DIR_SKILLS}}/codebase-design.md) | designing a module's interface, placing a seam, or needing the deep-module vocabulary; pinning domain terms in GLOSSARY.md or recording a hard-to-reverse decision as an ADR |
238
236
  | [wayfinder]({{DIR_SKILLS}}/wayfinder.md) | a chunk of work too big for one session, still in the fog — chart decision tickets before building |
239
237
  | [parallel-agents]({{DIR_SKILLS}}/parallel-agents.md) | several independent subtasks and a tool that runs subagents |
240
238
  | [develop]({{DIR_SKILLS}}/develop.md) | implementing a feature, fixing a bug, or changing behaviour — failing test first, affected suite green, one committed slice at a time |
241
239
  | [worktree]({{DIR_SKILLS}}/worktree.md) | work must not share a tree with what is in flight — a crew, a second task, another session in this checkout |
242
- | [debug]({{DIR_SKILLS}}/debug.md) | a bug or failure to diagnose before fixing — a red CI run or pre-merge check included, reproduced under CI's conditions |
243
240
  | [prototype]({{DIR_SKILLS}}/prototype.md) | a design question needs a runnable answer before real implementation |
244
241
  | [document-project]({{DIR_SKILLS}}/document-project.md) | missing/drifted docs, or a PRD, design system, architecture doc or AGENTS.md; orienting to the repo, or refreshing its map after the structure changed |
245
242
  | [ingest]({{DIR_SKILLS}}/ingest.md) | a task needs content from a PDF, Office doc, or URL |
@@ -248,8 +245,6 @@ the ten council seats are convened by the [council {{SKILL}}]({{DIR_SKILLS}}/cou
248
245
  | [learn-mode]({{DIR_SKILLS}}/learn-mode.md) | the user wants to learn while building — they design, you code what they approve |
249
246
  | [ownership]({{DIR_SKILLS}}/ownership.md) | the developer must own production code — they decide each part at a granularity the team sets, from the root AGENTS.md |
250
247
  | [commit]({{DIR_SKILLS}}/commit.md) | staging and writing a commit |
251
- | [geneseed-code-review]({{DIR_SKILLS}}/geneseed-code-review.md) | reviewing a diff or PR |
252
- | [review-response]({{DIR_SKILLS}}/review-response.md) | acting on review feedback you have received |
253
248
  | [fresh-eyes]({{DIR_SKILLS}}/fresh-eyes.md) | a finished artifact needs an independent verdict on whether it meets the original spec, from an agent with none of the solving session's context |
254
249
  | [security-audit]({{DIR_SKILLS}}/security-audit.md) | a change touches the security surface — auth, untrusted input, secrets, shell or file access, a new dependency — or "is this safe" |
255
250
  | [dependencies]({{DIR_SKILLS}}/dependencies.md) | auditing dependencies — outdated, vulnerable, abandoned, unused — or upgrading a dependency, framework, or language version one change at a time |
@@ -298,7 +293,7 @@ inventory — that list is the catalogue and it is authoritative, as is its
298
293
  procedure is in [`{{DIR_SKILLS}}/`]({{DIR_SKILLS}}/) and the host loads it on
299
294
  demand.
300
295
  <!-- CATALOG:end -->
301
- {{EXCLUDED_SKILLS_LINE}}{{KITS_LINE}}{{SHARED_TIER_LINE}}{{CONTEXT_WINDOW_LINE}}
296
+ {{EXCLUDED_SKILLS_LINE}}{{KITS_LINE}}{{SHARED_TIER_LINE}}{{CONTEXT_WINDOW_LINE}}{{TIERS_LINE}}
302
297
  When a task repeats and no {{SKILL}} covers it, forge one through the
303
298
  [skill-forge]({{DIR_SKILLS}}/skill-forge.md) {{SKILL}}: copy
304
299
  [`{{DIR_SKILLS}}/_template.md`]({{DIR_SKILLS}}/_template.md), keep it to one coherent
@@ -38,6 +38,10 @@
38
38
  edit, webfetch, and bash. If it must run read-only commands (tests, linters,
39
39
  scanners), add the marker `<!-- bash: allow -->` in this section to gate bash to
40
40
  "ask" instead of denying it outright.
41
+ - Model tier: an unmarked agent is `standard`. For `light` (read-only, read-and-summarise
42
+ seats) or `strong` (judgement seats that must run on the session model), put the marker
43
+ `<!-- tier: light -->` on a line of its own directly above `## When to dispatch`; the build
44
+ strips it from every output. A seat that writes code is never `light`.
41
45
  - Caution: the build detects the phrase "Read-only" anywhere in the file — a
42
46
  write-capable {{AGENT}}'s spec must not contain it, or its emit is locked down.
43
47
 
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_ADVOCATE}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs the strongest case for the motion.
7
8
  - The user wants the upside of a change, plan, or decision pressed hard rather than hedged.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_ARCHITECT}}
4
4
 
5
+ <!-- tier: strong -->
5
6
  ## When to dispatch
6
7
  - A task is large, ambiguous, or touches multiple subsystems.
7
8
  - There are competing approaches and the trade-off needs to be made explicit.
@@ -48,9 +48,10 @@
48
48
  actually close — report the blocker instead ({{ONTOLOGY}}: {{ONT_EVIDENCE}}).
49
49
 
50
50
  ## Output contract
51
- - The report starts with one line, `STATUS: OK|BLOCKED|OUT_OF_SCOPE|ERROR`: `OK` the contract is met;
51
+ - The report starts with one line, `STATUS: OK|BLOCKED|OUT_OF_SCOPE|ERROR|RELAY`: `OK` the contract is met;
52
52
  `BLOCKED` an input or a decision is missing; `OUT_OF_SCOPE` not this task; `ERROR` a step failed,
53
- red tests included.
53
+ red tests included; `RELAY` a task worker hands off at its ceiling: work in progress committed,
54
+ handoff note follows.
54
55
  - The files changed in the worktree and a one-line summary of what changed and
55
56
  why, ready for the tester's pass.
56
57
  - If the scoped task cannot be implemented as given (missing input, contradictory
@@ -61,8 +62,15 @@
61
62
  ### Task worker (Foreman)
62
63
 
63
64
  Dispatched as Foreman's task worker, there is no tester seat: explore, implement and test
64
- in your own context, and after two repair rounds on your own red tests return red with the
65
- findings ({{DOCTRINE:bound-the-loop}}). As a lead, hold every member's result before you end. Report `T<n> · <task> · files: <write set> · <state>`
65
+ in your own context. While you iterate, run the targeted test file, not the full suite.
66
+ Send its output to a file, and read only the count line and the failing test names. Wait
67
+ for a background run's completion notice; never poll it with a monitor or a sleep loop.
68
+ Count your proof runs (one run of a check command); these bounds are starting values
69
+ ({{DOCTRINE:bound-the-loop}}). After the 3rd run in a row that leaves the same test red,
70
+ return `STATUS: ERROR` with the findings. At about 150k tokens of context, or after about
71
+ 20 proof runs, hand off: commit your work in progress, and return `STATUS: RELAY` with a
72
+ handoff note (the failing tests, what you tried, the next step). A hand-off is no failure.
73
+ As a lead, hold every member's result before you end. Report `T<n> · <task> · files: <write set> · <state>`
66
74
  at start and on each change, as the mode says. Return the branch, its commit list, any
67
75
  uncommitted work, and the raw test + lint output — the parent re-runs it before any merge.
68
76
  That output is a full run to completion, with the runner's own count line, never a
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_EMPIRICIST}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs every claim, for and against, held to evidence.
7
8
  - The debate is running on assertion and intuition; the user wants it anchored to data, tests, or precedent.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_EXPLORER}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A question needs sweeping many files or directories, but only the conclusion
7
8
  belongs in the main context — not the file contents.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_FRAMER}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a test of the framing — is this the right problem.
7
8
  - The motion may be a solution in search of a problem, or the real need sits upstream of what's proposed.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_HISTORIAN}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs precedent and institutional memory.
7
8
  - The motion resembles something tried before; the user wants the track record before repeating it.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_OPERATOR}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a voice for running it in production.
7
8
  - The motion ships to a live system: its behaviour under load, its observability, and the on-call burden all matter.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_PERFORMANCE}}
4
4
 
5
+ <!-- tier: strong -->
5
6
  ## When to dispatch
6
7
  - The user reports something slow, heavy or timing out — a latency, throughput,
7
8
  memory or startup complaint, or a regression against an earlier version.
@@ -13,7 +14,7 @@
13
14
  ## When NOT to dispatch
14
15
  - No one has measured anything and nothing is slow — speculative tuning is the
15
16
  waste this {{AGENT}} exists to refuse.
16
- - The code is wrong, not slow — that is the debug {{SKILL}}.
17
+ - The code is wrong, not slow — that is the [debug {{SKILL}}](../{{DIR_SKILLS}}/debug.md).
17
18
 
18
19
  ## Inputs
19
20
  - The symptom or the diff, the workload it runs under (data size, call rate,
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_PRAGMATIST}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a judgement of whether the motion is worth its cost.
7
8
  - The user wants the real cost, effort, and complexity of a change surfaced against its payoff.
@@ -6,6 +6,7 @@
6
6
 
7
7
  React-specific review only. Generic correctness, naming and structure belong to [reviewer](reviewer.md); defer to it and do not repeat its checks.
8
8
 
9
+ <!-- tier: strong -->
9
10
  ## When to dispatch
10
11
  - A change touches `.tsx` / `.jsx` files, hooks or React tests and needs a framework-aware pass.
11
12
 
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_RESEARCHER}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A question needs verified answers from *outside* the repository. The open web, a
7
8
  vendor's docs, an upgrade guide, a specification: the main context gets the verified
@@ -4,6 +4,7 @@
4
4
 
5
5
  {{CREDIT:superpowers:skills/requesting-code-review/code-reviewer.md}}
6
6
 
7
+ <!-- tier: strong -->
7
8
  ## When to dispatch
8
9
  - A change is complete and about to be committed or opened as a PR.
9
10
  - The user asks for a review, second opinion, or pre-merge check.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_SECURITY}}
4
4
 
5
+ <!-- tier: strong -->
5
6
  ## When to dispatch
6
7
  - A change touches authentication, authorization, input handling, file/network
7
8
  I/O, secrets, cryptography, or dependencies.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_SKEPTIC}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a devil's advocate to attack the motion.
7
8
  - The user wants a change, plan, or claim stress-tested for how it fails before committing.
@@ -6,6 +6,7 @@
6
6
 
7
7
  Spring-specific review only. Generic correctness, naming and structure belong to [reviewer](reviewer.md); defer to it and do not repeat its checks.
8
8
 
9
+ <!-- tier: strong -->
9
10
  ## When to dispatch
10
11
  - A change touches Spring Boot code (controllers, services, repositories, configuration, tests) and needs a framework-aware pass.
11
12
 
@@ -6,6 +6,7 @@
6
6
 
7
7
  Spring Security 7 audit only. The generic audit (injection, secrets, dependencies beyond Spring) belongs to [security](security.md); defer to it.
8
8
 
9
+ <!-- tier: strong -->
9
10
  ## When to dispatch
10
11
  - A Spring change touches authentication, authorization, CORS, CSRF, headers, TLS or secrets handling.
11
12
 
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_STEWARD}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a defence of the long term against short-term wins.
7
8
  - The user wants the architectural and maintenance consequences of a change weighed before committing.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_USER_ADVOCATE}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs a voice for whoever consumes the outcome.
7
8
  - The motion touches a user-facing surface: UX, an API's ergonomics, docs, error messages, or a downstream developer's workflow.
@@ -2,6 +2,7 @@
2
2
 
3
3
  > {{DESC_VISIONARY}}
4
4
 
5
+ <!-- tier: light -->
5
6
  ## When to dispatch
6
7
  - A council debate (the council {{SKILL}}) needs the bold, transformative version argued against incrementalism.
7
8
  - The debate risks under-reaching: the user wants the ambitious upside of a change pushed, not just the safe slice.
@@ -5,8 +5,18 @@ effect: mutate
5
5
  skill: debug
6
6
  outcomes: pass, fail
7
7
  ---
8
- Follow the debug skill's CI branch against the failing check the card names — read its actual
9
- log, not a guess from the check's title, before changing anything.
8
+ The steps below stand alone. If a debug skill is installed, or the superpowers plugin's
9
+ systematic-debugging skill, load it for the general procedure; if neither is, carry on without.
10
+ Work against the failing check the card names:
11
+
12
+ 1. Read its actual log, not a guess from the check's title, before changing anything. Find the
13
+ first failing job and step, and the command that produced the error.
14
+ 2. Classify before touching anything: a real failure, an environment difference (version, path,
15
+ line endings, a binary missing in CI), an infra flake, or a gate on the report itself.
16
+ 3. Reproduce under CI's own conditions (its script, flags and toolchain). Green locally and red in
17
+ CI has not reproduced yet. Never retry a run blindly.
18
+ 4. Fix at the cause. Make the repo deterministic rather than CI more permissive; fix a flake at
19
+ its source or quarantine it with an issue and an owner, never skip it silently.
10
20
 
11
21
  Touch only what that failure requires; this is a repair, not a chance to improve the pipeline
12
22
  along the way.
@@ -0,0 +1,3 @@
1
+ {{PROVIDER_HEADING}}
2
+
3
+ The `superpowers` kit is off, and this host has no superpowers plugin to stand in for it, so the brainstorm, debug, plan, code-review and review-response {{SKILLS}} are not installed. Work such a task from the general method. If no superpowers skill is in your inventory, tell the user once: turn on the `superpowers` kit.