ultra-builder-pro-cli 0.5.2 → 0.6.0

package/adapters/_shared/tests/runtime-assets.test.cjs

@@ -9,6 +9,7 @@ const REPO_ROOT = path.resolve(__dirname, '..', '..', '..');
 const {
   CORE_PUBLIC_SKILLS,
   INTERNAL_AGENT_SKILLS,
+  SUPPORTED_RUNTIMES,
   COLLAB_SKILLS_BY_RUNTIME,
   RETIRED_SKILLS,
   WORKFLOW_HOOK_FILES,
@@ -51,19 +52,17 @@ test('runtime asset manifest exposes only Ultra-owned core and internal skills',
   assert.deepEqual(RETIRED_SKILLS, RETIRED);
 
   assert.deepEqual(COLLAB_SKILLS_BY_RUNTIME.claude, [
-    'codex-collab', 'gemini-collab', 'ultra-verify',
+    'codex-collab', 'ultra-verify',
   ]);
   assert.deepEqual(COLLAB_SKILLS_BY_RUNTIME.codex, [
-    'cc-collab', 'gemini-collab', 'ultra-verify',
+    'cc-collab', 'ultra-verify',
   ]);
   assert.deepEqual(COLLAB_SKILLS_BY_RUNTIME.opencode, [
-    'cc-collab', 'codex-collab', 'gemini-collab', 'ultra-verify',
-  ]);
-  assert.deepEqual(COLLAB_SKILLS_BY_RUNTIME.gemini, [
     'cc-collab', 'codex-collab', 'ultra-verify',
   ]);
+  assert.deepEqual(SUPPORTED_RUNTIMES, ['claude', 'opencode', 'codex']);
 
-  for (const runtime of ['claude', 'codex', 'opencode', 'gemini']) {
+  for (const runtime of SUPPORTED_RUNTIMES) {
     const names = skillsForRuntime(runtime);
     for (const retired of RETIRED) assert.ok(!names.includes(retired));
     assert.ok(!names.some((name) => /impeccable/i.test(name)));
@@ -89,6 +88,18 @@ test('workflow hook allowlist contains no memory, prompt capture, or generic pol
   assert.doesNotMatch(WORKFLOW_HOOK_FILES.join('\n'), /memory|recall|journal|observation|prompt|dangerous|post_edit/);
 });
 
+test('bundled agents never instruct workers to own persistent memory', () => {
+  const agentsDir = path.join(REPO_ROOT, 'agents');
+  const agentText = fs.readdirSync(agentsDir)
+    .filter((name) => name.endsWith('.md'))
+    .map((name) => fs.readFileSync(path.join(agentsDir, name), 'utf8'))
+    .join('\n');
+
+  assert.doesNotMatch(agentText, /^## Memory$/m);
+  assert.doesNotMatch(agentText, /Update your project memory|Consult memory before starting work/);
+  assert.doesNotMatch(agentText, /\.ultra\/memory|\/recall(?:\s|`|$)/m);
+});
+
 test('npm publish list uses the same explicit skill boundary', () => {
   const pkg = JSON.parse(fs.readFileSync(path.join(REPO_ROOT, 'package.json'), 'utf8'));
   const publishedSkills = pkg.files

package/adapters/opencode.js

@@ -94,7 +94,7 @@ function openCodeTextTransform(input, assetName = '') {
   text = text.replaceAll('Claude runtime', 'OpenCode runtime');
   text = text.replaceAll('Claude-only', 'OpenCode-only');
 
-  if (assetName === 'codex-collab' || assetName === 'gemini-collab') {
+  if (assetName === 'codex-collab') {
     text = text.replaceAll('within Claude Code', 'from OpenCode');
     text = text.replaceAll('Claude Code remains primary', 'OpenCode remains primary');
     text = text.replaceAll('Claude orchestrates', 'OpenCode remains primary and orchestrates');
@@ -104,13 +104,8 @@ function openCodeTextTransform(input, assetName = '') {
   }
   if (assetName === 'ultra-verify') {
     text = text.replaceAll('claude-analysis.md', 'opencode-analysis.md');
-    text = text.replaceAll('Claude + Gemini + Codex', 'OpenCode + Gemini + Codex');
     text = text.replaceAll('Claude Code', 'OpenCode');
     text = text.replaceAll('Claude', 'OpenCode');
-    text = text.replace(
-      /Orchestrate OpenCode \+ Gemini \+ Codex for independent three-way analysis\./,
-      'OpenCode remains primary while Gemini and Codex provide independent read-only analyses.',
-    );
   }
   if (assetName === 'learn') {
     text = text.replaceAll(

package/adapters/tests/claude.test.cjs

@@ -52,7 +52,6 @@ test('Claude plugin collaboration and learn workflows are safe native plugin ass
     const read = (name) => fs.readFileSync(path.join(target, 'skills', name, 'SKILL.md'), 'utf8');
     const learn = read('learn');
     const codexCollab = read('codex-collab');
-    const geminiCollab = read('gemini-collab');
     const verify = read('ultra-verify');
     const review = read('ultra-review');
     const plan = read('ultra-plan');
@@ -61,8 +60,6 @@ test('Claude plugin collaboration and learn workflows are safe native plugin ass
     assert.match(learn, /~\/.claude\/skills\/learned-<pattern-slug>-unverified\/SKILL\.md/);
     assert.doesNotMatch(learn, /_unverified\.md/);
     assert.match(codexCollab, /-s read-only/);
-    assert.match(geminiCollab, /--approval-mode plan/);
-    assert.match(verify, /--approval-mode plan/);
     assert.match(verify, /-s read-only/);
     assert.match(verify, /\$CLAUDE_PLUGIN_ROOT\/skills\/ultra-verify\/scripts\/verify_wait\.py/);
     assert.match(review, /\$CLAUDE_PLUGIN_ROOT\/skills\/ultra-review\/scripts\/review_wait\.py/);
@@ -71,7 +68,7 @@ test('Claude plugin collaboration and learn workflows are safe native plugin ass
     assert.match(status, /never fall back .*tasks\.json/i);
     assert.doesNotMatch(plan, /ultra-tools task create/);
 
-    for (const [name, contents] of Object.entries({ codexCollab, geminiCollab, verify })) {
+    for (const [name, contents] of Object.entries({ codexCollab, verify })) {
       assert.doesNotMatch(contents, /--yolo|--full-auto|\/codex:/, name);
     }
   } finally {

package/adapters/tests/codex.test.cjs

@@ -89,7 +89,7 @@ test('install builds one Codex-native plugin with complete skill and command cov
 
     const expectedSkills = skillsForRuntime('codex').sort();
     assert.deepEqual(skillNames(layout.pluginRoot), expectedSkills);
-    assert.equal(expectedSkills.length, 17);
+    assert.equal(expectedSkills.length, 16);
     assert.ok(!fs.existsSync(path.join(layout.pluginRoot, 'skills', 'codex-collab')));
     assert.ok(!fs.existsSync(path.join(layout.pluginRoot, 'skills', 'learned')));
 
@@ -117,7 +117,7 @@ test('every generated skill is Codex-valid, UI-visible, and free of Claude host
       assert.equal(fm.name, name);
       assert.doesNotMatch(
         text,
-        /~\/\.claude|CLAUDE\.md|AskUserQuestion|(^|[\s`(>])\/(?:ultra-[a-z-]+|recall|learn|gemini-collab|codex-collab)(?=$|[\s`,.;):])/m,
+        /~\/\.claude|CLAUDE\.md|AskUserQuestion|(^|[\s`(>])\/(?:ultra-[a-z-]+|recall|learn|codex-collab)(?=$|[\s`,.;):])/m,
       );
       assert.doesNotMatch(text, /TaskCreate|TaskUpdate|TaskList|TaskOutput|run_in_background|~\/\.codex\/skills|mcp__claude/);
 
@@ -135,13 +135,12 @@ test('every generated skill is Codex-valid, UI-visible, and free of Claude host
     const verify = fs.readFileSync(path.join(layout.pluginRoot, 'skills', 'ultra-verify', 'SKILL.md'), 'utf8');
     assert.match(verify, /Codex owns the task/);
     assert.match(verify, /claude --safe-mode/);
-    assert.match(verify, /gemini --approval-mode plan/);
     assert.match(verify, /codex-analysis\.md/);
     assert.doesNotMatch(verify, /codex-output\.md|Claude synthesizes|Claude-only/);
 
     const waiter = fs.readFileSync(path.join(layout.pluginRoot, 'skills', 'ultra-verify', 'scripts', 'verify_wait.py'), 'utf8');
-    assert.match(waiter, /"claude": "claude-output\.md"/);
-    assert.doesNotMatch(waiter, /"codex": "codex-output\.md"/);
+    assert.match(waiter, /ADVISOR = "claude"/);
+    assert.match(waiter, /OUTPUT_FILE = "claude-output\.md"/);
 
     const review = fs.readFileSync(path.join(layout.pluginRoot, 'skills', 'ultra-review', 'SKILL.md'), 'utf8');
     assert.match(review, /native Codex custom agents/);
@@ -179,7 +178,7 @@ test('install converts all nine Claude agents into native Codex agent TOML', ()
       assert.match(text, /^developer_instructions = """/m);
       assert.doesNotMatch(
         text,
-        /model = "opus"|^tools =|^memory =|maxTurns|CLAUDE\.md|AskUserQuestion|(^|[\s`(>])\/(?:ultra-[a-z-]+|recall|learn|gemini-collab|codex-collab)(?=$|[\s`,.;):])/m,
+        /model = "opus"|^tools =|^memory =|maxTurns|CLAUDE\.md|AskUserQuestion|(^|[\s`(>])\/(?:ultra-[a-z-]+|recall|learn|codex-collab)(?=$|[\s`,.;):])/m,
       );
     }
   } finally {

package/adapters/tests/opencode.test.cjs

@@ -60,7 +60,6 @@ test('install performs content-level OpenCode adaptation for commands, skills, r
     const learn = fs.readFileSync(path.join(target, 'skills', 'learn', 'SKILL.md'), 'utf8');
     const review = fs.readFileSync(path.join(target, 'skills', 'ultra-review', 'SKILL.md'), 'utf8');
     const codexCollab = fs.readFileSync(path.join(target, 'skills', 'codex-collab', 'SKILL.md'), 'utf8');
-    const geminiCollab = fs.readFileSync(path.join(target, 'skills', 'gemini-collab', 'SKILL.md'), 'utf8');
     const verify = fs.readFileSync(path.join(target, 'skills', 'ultra-verify', 'SKILL.md'), 'utf8');
 
     assert.deepEqual(Object.keys(parseFm(plan).fm), ['name', 'description']);
@@ -68,9 +67,9 @@ test('install performs content-level OpenCode adaptation for commands, skills, r
     assert.doesNotMatch(learn, /_unverified\.md/);
     assert.match(review, /~\/.config\/opencode\/skills\/ultra-review\/scripts\/review_wait\.py/);
     assert.match(codexCollab, /OpenCode remains primary/);
-    assert.match(geminiCollab, /OpenCode remains primary/);
     assert.match(verify, /OpenCode remains primary/);
     assert.match(verify, /opencode-analysis\.md/);
+    assert.match(verify, /codex exec -s read-only/);
     assert.match(plan, /LEGACY_STATE_MIGRATION_REQUIRED/);
     assert.match(plan, /never read or write .*tasks\.json/i);
     assert.doesNotMatch(plan, /ultra-tools task create/);

package/agents/tdd-runner.md

@@ -86,7 +86,8 @@ Scan test files for forbidden patterns:
 ALL PASS / X FAILURES NEED ATTENTION
 ```
 
-## Memory
+## Handoff
 
-Update your project memory with test patterns, common failure causes, and framework
-quirks for this project. Consult memory before starting work.
+Return reusable test patterns, common failure causes, and framework quirks to the
+parent agent as explicit evidence. Do not create or update a private Ultra memory
+store; persistent memory belongs to the host's separately installed provider.

package/bin/install.js

@@ -4,14 +4,14 @@
  * ultra-builder-pro-cli — multi-runtime installer.
  *
  * Distributes Ultra Builder Pro assets — commands, agents, skills, hooks,
- * MCP server — to Claude Code, OpenCode, Codex CLI, and Gemini CLI via
+ * MCP server — to Claude Code, OpenCode, and Codex CLI via
  * runtime-specific adapters under adapters/. Install is idempotent and
  * uses atomic writes; uninstall reverses via sentinel/manifest blocks.
  *
  * Usage:
  *   npx ultra-builder-pro-cli [options]
  *
- *   --claude / --opencode / --codex / --gemini   select runtime(s)
+ *   --claude / --opencode / --codex              select runtime(s)
  *   --all                                         install to all supported runtimes
  *   -g, --global                                  install to runtime's global config dir
  *   -l, --local                                   install into current working directory
@@ -29,8 +29,7 @@ const os = require('node:os');
 
 const pkg = require('../package.json');
 const { validateConfigDir } = require('../adapters/_shared/validate.cjs');
-
-const SUPPORTED_RUNTIMES = ['claude', 'opencode', 'codex', 'gemini'];
+const { SUPPORTED_RUNTIMES } = require('../adapters/_shared/runtime-assets.cjs');
 
 const COLORS = {
   reset: '\x1b[0m',
@@ -58,7 +57,7 @@ function printBanner() {
     paint('cyan', '   ╚═════╝ ╚═════╝ ╚═╝         ╚═════╝╚══════╝╚═╝'),
     '',
     `  ${paint('bold', 'Ultra Builder Pro CLI')} ${paint('dim', 'v' + pkg.version)}`,
-    `  ${paint('dim', 'Multi-runtime installer for Claude Code, OpenCode, Codex, Gemini')}`,
+    `  ${paint('dim', 'Multi-runtime installer for Claude Code, OpenCode, and Codex')}`,
     '',
   ];
   console.log(banner.join('\n'));
@@ -71,7 +70,6 @@ function printHelp() {
     ${paint('cyan', '--claude')}           Claude Code
     ${paint('cyan', '--opencode')}         OpenCode
     ${paint('cyan', '--codex')}            Codex CLI (OpenAI)
-    ${paint('cyan', '--gemini')}           Gemini CLI (Google)
     ${paint('cyan', '--all')}              all supported runtimes
 
   ${paint('yellow', 'Scope:')}
@@ -89,7 +87,7 @@ function printHelp() {
     ${paint('dim', '# Install to Claude Code globally')}
     npx ultra-builder-pro-cli --claude --global
 
-    ${paint('dim', '# Install to all 4 runtimes locally in this repo')}
+    ${paint('dim', '# Install to all supported runtimes locally in this repo')}
     npx ultra-builder-pro-cli --all --local
 
     ${paint('dim', '# Uninstall from OpenCode')}
@@ -116,7 +114,6 @@ function parseArgs(argv) {
       case '--claude': runtimes.add('claude'); break;
       case '--opencode': runtimes.add('opencode'); break;
       case '--codex': runtimes.add('codex'); break;
-      case '--gemini': runtimes.add('gemini'); break;
       case '--all':
         SUPPORTED_RUNTIMES.forEach(r => runtimes.add(r));
         break;
@@ -193,7 +190,7 @@ async function main() {
   }
 
   if (runtimes.length === 0) {
-    bail('no runtime selected; use --claude / --opencode / --codex / --gemini / --all');
+    bail('no runtime selected; use --claude / --opencode / --codex / --all');
   }
 
   const scope = resolveScope(flags);

package/bin/orchestrator.js

@@ -22,6 +22,7 @@
 const fs = require('node:fs');
 const path = require('node:path');
 const { spawn } = require('node:child_process');
+const { isSupportedRuntime } = require('../adapters/_shared/runtime-assets.cjs');
 
 const REPO_ROOT = process.env.UBP_REPO_ROOT || process.cwd();
 const PIDFILE = path.join(REPO_ROOT, '.ultra', 'orchestrator.pid');
@@ -46,8 +47,13 @@ function optInAllowed(settings) {
 }
 
 function parseRuntimes() {
-  const raw = process.env.UBP_ORCH_RUNTIMES || 'claude,opencode,codex,gemini';
-  return raw.split(',').map((s) => s.trim()).filter(Boolean);
+  const raw = process.env.UBP_ORCH_RUNTIMES || 'claude,opencode,codex';
+  const runtimes = [...new Set(raw.split(',').map((s) => s.trim()).filter(Boolean))];
+  const invalid = runtimes.filter((runtime) => !isSupportedRuntime(runtime));
+  if (invalid.length > 0) {
+    throw new Error(`unsupported orchestrator runtime(s): ${invalid.join(', ')}`);
+  }
+  return runtimes;
 }
 
 function cmdRun(opts = {}) {
@@ -61,11 +67,12 @@ function cmdRun(opts = {}) {
   }
   const { initStateDb } = require('../mcp-server/lib/state-db.cjs');
   const { runDaemon } = require('../orchestrator/daemon.cjs');
+  const runtimes = parseRuntimes();
   const { db } = initStateDb(path.join(REPO_ROOT, '.ultra', 'state.db'));
   const handle = runDaemon({
     db,
     repoRoot: REPO_ROOT,
-    runtimes: parseRuntimes(),
+    runtimes,
     pollMs: Number(process.env.UBP_ORCH_POLL_MS || 1000),
     onError: (err) => process.stderr.write(`orchestrator error: ${err.message}\n`),
   });
@@ -102,7 +109,7 @@ function cmdRun(opts = {}) {
   };
   process.on('SIGTERM', () => shutdown('SIGTERM'));
   process.on('SIGINT', () => shutdown('SIGINT'));
-  process.stderr.write(`orchestrator running (pollMs=${Number(process.env.UBP_ORCH_POLL_MS || 1000)}, runtimes=${parseRuntimes().join(',')})\n`);
+  process.stderr.write(`orchestrator running (pollMs=${Number(process.env.UBP_ORCH_POLL_MS || 1000)}, runtimes=${runtimes.join(',')})\n`);
   // Keep process alive while the setInterval is unref'd.
   setInterval(() => {}, 60000);
 }

package/docs/AGENT-CONTEXT.md

@@ -1,8 +1,7 @@
 # Ultra Builder Pro — Agent Context
 
 Current shared runtime contract for the native Claude Code, OpenCode, and Codex
-plugins. Gemini CLI remains a compatibility adapter and is not the reference
-shape for the three first-class plugins.
+plugins.
 
 ## 1. Ownership boundary
 
@@ -66,9 +65,9 @@ fixtures. Do not add an ad-hoc server handler first.
 
 | Host | Plugin form | Workflow entry | Hook form | Collaboration companions |
 |---|---|---|---|---|
-| Claude Code | `.claude-plugin/plugin.json`, native commands/skills/agents, `.mcp.json` | `/ultra-*`, `/learn` | native `hooks/hooks.json` | `codex-collab`, `gemini-collab` |
-| Codex | personal plugin with `.codex-plugin/plugin.json`, namespaced skills, TOML agents, `.mcp.json` | `$ultra-builder-pro:ultra-*`, `$ultra-builder-pro:learn` | native `hooks/hooks.json` through the Codex wire adapter | `cc-collab`, `gemini-collab` |
-| OpenCode | config bundle plus native JavaScript plugin | `/ultra-*`, `/learn` | `event`, system transform, compaction, and tool lifecycle handlers | `cc-collab`, `codex-collab`, `gemini-collab` |
+| Claude Code | `.claude-plugin/plugin.json`, native commands/skills/agents, `.mcp.json` | `/ultra-*`, `/learn` | native `hooks/hooks.json` | `codex-collab`, `ultra-verify` |
+| Codex | personal plugin with `.codex-plugin/plugin.json`, namespaced skills, TOML agents, `.mcp.json` | `$ultra-builder-pro:ultra-*`, `$ultra-builder-pro:learn` | native `hooks/hooks.json` through the Codex wire adapter | `cc-collab`, `ultra-verify` |
+| OpenCode | config bundle plus native JavaScript plugin | `/ultra-*`, `/learn` | `event`, system transform, compaction, and tool lifecycle handlers | `cc-collab`, `codex-collab`, `ultra-verify` |
 
 The current host remains primary. Collaboration skills call another runtime
 only when explicitly requested, use it as a read-only advisor, and return the

package/docs/ARCHITECTURE.md

@@ -10,8 +10,7 @@
 
 `ultra-builder-pro-cli` is a multi-runtime plugin suite that **distributes** the
 Ultra Builder Pro engineering loop (skills + commands + workflow hooks) as
-native Claude Code, OpenCode, and Codex plugins, with a Gemini compatibility
-adapter, and
+native Claude Code, OpenCode, and Codex plugins, and
 **runs** that loop with isolated sessions sharing one authoritative state
 store (`.ultra/state.db`).
 
@@ -24,7 +23,7 @@ the supported hosts share.
                        ┌──────────────────────────┐
                        │       runtime CLI        │
                        │  (claude / opencode /    │
-                       │   codex / gemini)        │
+                       │   codex)                 │
                        └────────────┬─────────────┘
                                     │
                 ┌───────────────────┼────────────────────┐
@@ -53,7 +52,7 @@ the supported hosts share.
 
 | Layer        | Role                                         | Form                                                  |
 |--------------|----------------------------------------------|-------------------------------------------------------|
-| **skill**    | knowledge carrier; tells the runtime *what to do* | `skills/<name>/SKILL.md` discovered natively by all four runtimes |
+| **skill**    | knowledge carrier; tells the runtime *what to do* | `skills/<name>/SKILL.md` discovered natively by all supported runtimes |
 | **MCP**      | authoritative workflow-state API             | stdio MCP server exposing seven declared tool families in [`spec/mcp-tools.yaml`](../spec/mcp-tools.yaml); 21 tools are live and nine scheduled contracts map to host-native surfaces |
 | **CLI**      | the lowest common denominator hooks can call  | `ultra-tools <family> <verb>` — same contract, last-line JSON envelope (see [`spec/cli-protocol.md`](../spec/cli-protocol.md)) |
 

package/docs/LEGACY-HERMES.md

@@ -117,8 +117,7 @@ If ANY component is fake/mocked/simulated -> Quality = 0
 | Skill | Purpose | User-Invocable |
 |-------|---------|----------------|
 | `ultra-review` | Parallel code review with 6 agents + coordinator | Yes |
-| `ultra-verify` | Three-way AI verification (Claude + Gemini + Codex) | Yes |
-| `gemini-collab` | Gemini CLI as independent sub-agent for review, analysis, opinions | Yes |
+| `ultra-verify` | Independent cross-model verification | Yes |
 | `codex-collab` | OpenAI Codex CLI as independent sub-agent for review, analysis | Yes |
 | `recall` | Cross-session memory search, save summaries, tags | Yes |
 | `agent-browser` | Browser automation CLI for AI agents | Yes |
@@ -206,54 +205,6 @@ Step 4.5 of `/ultra-dev` runs `/ultra-review all` (forced full coverage) as a ma
 
 ---
 
-## Ultra Verify — Three-Way AI Cross-Verification
-
-### Overview
-
-`/ultra-verify` orchestrates Claude + Gemini + Codex for independent three-way analysis. Each AI works independently, then Claude synthesizes with a confidence score based on consensus.
-
-### Prerequisites
-
-- Gemini CLI: `npm install -g @google/gemini-cli` + authenticated
-- Codex CLI: `npm install -g @openai/codex` + `codex login`
-
-### Usage
-
-```
-/ultra-verify decision <question>    # Architecture/design decision
-/ultra-verify diagnose <symptoms>    # Bug diagnosis — three sets of hypotheses
-/ultra-verify audit <scope>          # Code audit — findings ranked by consensus
-/ultra-verify estimate <task>        # Effort estimation
-```
-
-### How It Works
-
-1. **Claude answers first** (writes to file before reading external AI output — prevents contamination)
-2. **Gemini + Codex run in parallel** (background tasks, 5 min timeout)
-3. **Claude reads all three outputs** via Read tool
-4. **Confidence scoring**: Consensus (3/3) > Majority (2/3) > No Consensus
-5. **Synthesis** written to `.ultra/collab/<session-id>/synthesis.md`
-
-### Degraded Operation
-
-- **One AI fails**: Two-way comparison, confidence capped at Majority
-- **Two AIs fail**: Claude-only with explicit warning
-- Never blocks the workflow on external AI failures
-
-### Architecture: Shared Base + Thin Skills
-
-```
-ai-collab-base/         # Canonical shared files (non-user-invocable)
-  ├── references/        # collab-protocol.md, collaboration-modes.md, prompt-templates.md
-  └── sync.sh            # Copies canonical files to consumer skills
-
-gemini-collab/           # Thin skill: Gemini-specific CLI reference + shared files
-codex-collab/            # Thin skill: Codex-specific CLI reference + shared files
-ultra-verify/            # Orchestration: parallel execution + confidence scoring
-```
-
----
-
 ## Cross-Session Memory
 
 ### Overview
@@ -479,14 +430,12 @@ Automated enforcement via Python hooks in `hooks/`. **Hooks are deterministic
 |   |   |-- templates/        # Output templates (reserved)
 |   |-- ultra-review/         # Parallel review orchestration
 |   |   |-- scripts/          # review_wait.py, review_verdict_update.py
-|   |-- ultra-verify/         # Three-way AI verification (Claude+Gemini+Codex)
+|   |-- ultra-verify/         # Independent cross-model verification
 |   |   |-- references/       # cross-verify-modes, confidence-system, orchestration-flow
 |   |   |-- evals/            # skill-creator eval tests
 |   |-- ai-collab-base/       # Shared collab protocol (non-user-invocable)
 |   |   |-- references/       # collab-protocol, collaboration-modes, prompt-templates
 |   |   |-- sync.sh           # Sync canonical files to consumer skills
-|   |-- gemini-collab/        # Gemini CLI as independent sub-agent
-|   |   |-- references/       # gemini-cli-reference, gemini-prompts, shared files
 |   |   |-- evals/            # skill-creator eval tests
 |   |-- codex-collab/         # OpenAI Codex CLI as independent sub-agent
 |   |   |-- references/       # codex-cli-reference, codex-prompts, shared files
@@ -532,7 +481,6 @@ Automated enforcement via Python hooks in `hooks/`. **Hooks are deterministic
 |   |-- collab/                      # AI collaboration sessions (auto-managed)
 |   |   |-- <session-id>/           # Per-session output
 |   |       |-- claude-analysis.md
-|   |       |-- gemini-output.md
 |   |       |-- codex-output.md
 |   |       |-- synthesis.md
 |   |       |-- metadata.json
@@ -648,7 +596,7 @@ Multi-step tasks use the Task system:
 
 ### v6.3.0 (2026-03-09) - Memory System v2
 
-**Structured summaries, real session identity, security hardening**, verified by 3 rounds of ultra-verify (Claude + Gemini + Codex):
+**Structured summaries, real session identity, security hardening**, verified by repeated independent review:
 
 **Schema v2 Migration**:
 - New `session_summaries` table: structured JSON fields (`request`, `completed`, `learned`, `next_steps`)
@@ -681,29 +629,6 @@ Multi-step tasks use the Task system:
 - `hooks/user_prompt_capture.py` (New), `hooks/observation_capture.py` (New)
 - `CLAUDE.md`, `README.md`
 
-### v6.2.0 (2026-03-08) - Multi-AI Collaboration Refactor
-
-**Shared base architecture + three-way AI verification**, verified by 3 rounds of ultra-verify audit (Claude + Gemini + Codex):
-
-- New `ai-collab-base` skill: shared collaboration protocol, modes, prompt templates (non-user-invocable)
-  - `sync.sh` keeps canonical files in sync across 3 consumer skills
-  - Eliminates ~90% structural duplication between gemini-collab and codex-collab
-- New `ultra-verify` skill: three-way AI cross-verification (Claude + Gemini + Codex)
-  - 4 modes: `decision`, `diagnose`, `audit`, `estimate`
-  - Confidence scoring: Consensus (3/3), Majority (2/3), No Consensus
-  - Degraded operation: one AI fails → two-way, two fail → Claude-only with warning
-- Rewritten `gemini-collab`: thin skill pointing to shared base + Gemini-specific CLI reference
-- Rewritten `codex-collab`: thin skill pointing to shared base + Codex-specific CLI reference
-- CLI bug fixes verified against actual `--help` output:
-  - `--commit-title` → `--title` (codex review)
-  - `--full-auto` + `--sandbox read-only` conflict resolved
-  - `$(cat $FILE)` shell injection → stdin pipe pattern
-  - `2>/dev/null` → `error.log` (all files, not just orchestration)
-  - `-o`/`--output-format` correctly documented for Gemini CLI
-  - Hardcoded model names removed (use `-m <model>`)
-- skill-creator compatible `evals/evals.json` for all 3 user-invocable skills
-- Files: `skills/ai-collab-base/`, `skills/ultra-verify/`, `skills/gemini-collab/`, `skills/codex-collab/`
-
 ### v6.1.0 (2026-03-08) - Product Discovery Round 0
 
 **Product Discovery & Strategy phase** — fills the gap between vague ideas and technical specification. Inspired by [phuryn/pm-skills](https://github.com/phuryn/pm-skills) frameworks:
@@ -723,7 +648,6 @@ Multi-step tasks use the Task system:
 
 **System consolidation, cleanup, and Multi-AI collaboration**:
 
-- New `gemini-collab` skill: Gemini CLI as sub-agent for review, project analysis, second opinions
 - New `codex-collab` skill: OpenAI Codex CLI as sub-agent with built-in `codex review` integration
 - Removed codex skill and all references (CLAUDE.md, README.md, skills/codex/)
 - Stop hook hardening: removed main branch bypass, fixed git status path truncation
@@ -1048,7 +972,7 @@ Multi-step tasks use the Task system:
 
 **Removed (redundant - Claude handles natively)**:
 - Agents: build-error-resolver, doc-updater, e2e-runner, frontend-developer, refactor-cleaner
-- Skills: gemini, promptup, skill-creator
+- Skills: promptup, skill-creator
 - Hooks: user_prompt_agent.py (routing), agent_reminder.py (routing)
 
 **Improved**:
@@ -1149,7 +1073,7 @@ Multi-step tasks use the Task system:
 
 **Core Changes**:
 - Unified Priority Stack in CLAUDE.md
-- Codex and Gemini skill integration
+- Codex skill integration
 - Anti-Pattern Detection in `/ultra-test`
 
 ---