ultra-builder-pro-cli 0.5.3 → 0.7.0

package/.ultra-template/changes/active/.gitkeep

@@ -0,0 +1 @@
+

package/.ultra-template/changes/archive/.gitkeep

@@ -0,0 +1 @@
+

package/CHANGELOG.md

@@ -7,6 +7,83 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.7.0] — 2026-07-17
+
+### Added
+
+- Added `ultra-change` and `ultra-doctor` as native workflows on Claude Code,
+  OpenCode, and Codex. Daily fixes, features, redesigns, and incidents now stay
+  attached to an explicit change packet after the initial baseline delivery.
+- Added seven `change.*` MCP tools and `system.doctor`, advancing the live MCP
+  contract to 29 tools and `.ultra/state.db` to schema `9.0` with 15 tables for
+  changes, artifacts, context snapshots, trace links, incidents, projection
+  jobs, and durable event-consumer cursors.
+- Added deterministic convergence and archive gates for task completion, spec
+  deltas, tests, documentation impact, review evidence, context freshness, and
+  baseline reconciliation.
+
+### Changed
+
+- `task.init_project` now initializes `state.db` immediately and creates active
+  and archived change roots. `tasks.json` remains a generated projection.
+- MCP mutations now commit authority first, enqueue a durable projection job,
+  process it, and expose projection status in MCP metadata instead of silently
+  swallowing projection failures.
+- Session-start and edit hooks now understand initialized baselines and active
+  changes. Health checks surface incidents, projection lag, orphan sessions,
+  and missing change artifacts; direct writes to the task projection remain
+  blocked even outside a command-scoped workflow.
+- Memory and code-graph content are exclusively owned by separately installed
+  providers. Ultra accepts only provider metadata references in compiled change
+  contexts and never stores recalled content, graph payloads, or embeddings.
+
+### Fixed
+
+- Wired spec-staleness consumption, crash recovery, projection processing, and
+  durable cursors into the real orchestrator loop instead of leaving production
+  helpers unconsumed.
+- Kept `intent.md` synchronized with authoritative change updates, linked task
+  status events to their newly assigned change, and included `plan.export`
+  events in projection accounting.
+- Added atomic projection claims and backup-first recovery for failed or stale
+  interrupted jobs so a crashed projector cannot remain invisibly `running`.
+- Kept all three native plugin bundles, MCP launchers, agents, hooks, commands,
+  and user-handbook renderings on one explicit Ultra-owned asset allowlist.
+
+### Removed
+
+- Removed the internal filesystem graph watcher and its dependency; external
+  graph providers own indexing and refresh.
+- Removed the retired command-proxy detector, installer option, managed handbook
+  block, cache artifacts, prompts, configuration, and documentation surface.
+- Stopped publishing the archived pre-CLI manual in the npm package; it remains
+  repository history and cannot be mistaken for an active plugin contract.
+
+## [0.6.0] — 2026-07-17
+
+### Changed
+
+- Reduced the supported host matrix to Claude Code, OpenCode, and Codex.
+- Reworked `ultra-verify` into independent two-model verification: the current
+  host remains primary, one read-only external advisor supplies a second view,
+  and evidence rather than model voting determines confidence.
+- Restricted orchestrator routing, MCP session/event inputs, telemetry writes,
+  state schema enums, installer flags, package metadata, and handbooks to the
+  three supported runtimes.
+
+### Removed
+
+- Removed the retired fourth-runtime adapter, installer surface, extension and
+  conformance packaging, pricing entries, collaboration companion, and all
+  active Ultra prompts and documentation that offered it as an advisor.
+
+### Fixed
+
+- Added a retirement regression contract so unsupported runtimes cannot
+  re-enter active source, package, schema, prompt, or generated plugin surfaces.
+- Rejects unsupported runtime labels before state, event, or telemetry writes,
+  including old databases whose historical SQLite constraints were broader.
+
 ## [0.5.3] — 2026-07-17
 
 ### Fixed
@@ -59,7 +136,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Added first-class native Claude Code and OpenCode plugin builders alongside
   the Codex personal plugin. All hosts build from one explicit runtime asset
-  allowlist; Gemini remains a compatibility extension using the same boundary.
+  allowlist; the then-supported compatibility extension used the same boundary.
 - Added `ubp-handbook` for previewed, backed-up managed-block integration with
   user-level Claude `CLAUDE.md`, Codex `AGENTS.md`, and OpenCode `AGENTS.md`.
 - Added `ultra-tools legacy-memory inspect|archive|prune`; prune requires the
@@ -87,7 +164,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   with the npm package, and fixed `session.subscribe_events` sid filtering.
 - Aligned session/plan input schemas with runtime behavior, including circuit
   breaker output, session spawn intent semantics, and removal of ignored inputs.
-- Bundled Gemini's compatibility MCP runtime durably and restored project-local
+- Bundled the compatibility MCP runtime durably and restored project-local
   `.ultra/state.db` ownership across all adapters.
 - Exported the documented `ultra-tools` executable and included every README
   documentation target in the npm tarball.
@@ -164,8 +241,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - D49 tech-debt sweep: stale Phase comments (`"scheduled for Phase 1"`,
   `"not implemented in Phase 3.1"`) replaced with accurate `unknown verb`
-  messages; `docs/ROADMAP.md` status table synchronized; `.gitignore`
-  excludes `.rtk/` local filter cache.
+  messages; `docs/ROADMAP.md` status table synchronized.
 
 ### Tests
 
@@ -176,13 +252,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- **Phase 6 — monitoring + live code graph** (D44): RTK soft-dependency hook
-  (`adapters/_shared/rtk-detect.cjs`) with `--skip-rtk` flag and install hint;
-  token / cost telemetry with per-runtime pricing table, `telemetry` table +
+- **Phase 6 — monitoring + live code graph** (D44): token / cost telemetry
+  with per-runtime pricing table, `telemetry` table +
   daily `.ultra/telemetry/YYYY-MM-DD.jsonl`; `ultra-tools status --cost`
   panel with by-runtime / by-task / by-session aggregation; chokidar-based
-  code-graph watcher with debounce / awaitWriteFinish / batch thresholds,
-  opt-in via `--with-graph-watcher`.
+  then-experimental incremental graph watcher with debounce and batch thresholds.
 - **Phase 5 — execution resilience** (D43): `recovery.cjs` boot-time orphan
   scan using live PID probe; `circuit_breaker` table with `recordTaskFailure`
   counter and `resetCircuitBreaker` escape hatch; spec-change staleness
@@ -193,8 +267,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Phase 4.6b — full conformance suite** (D45): 4 runtime × 5 capability
   (command / skills / hooks / MCP + no `env._source` leak / install idempotency)
   = 20 additional conformance tests; table-driven `resolve-target.test.cjs`
-  (21 tests) covering `configDir > env > home/cwd` precedence + Gemini
-  extensionRoot append; schema bump 5.2 → 7.1 alongside memory store.
+  (21 tests) covering `configDir > env > home/cwd` precedence and extension
+  root resolution; schema bump 5.2 → 7.1 alongside memory store.
 
 ### Changed
 
@@ -213,9 +287,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   admission control with takeover / resume / abandon strategies (D33).
 - **Phase 4 — cross-runtime distribution** (D41, `5aa1fd0`): shared adapter
   toolkit (`adapters/_shared/` — file-ops / frontmatter / settings-merge /
-  path-rewrite / md-to-toml); 4 runtime adapters (Claude sentinel-block
+  path-rewrite / md-to-toml); the initial runtime adapters (Claude sentinel-block
   settings.json merge; OpenCode `opencode.json.mcp`; Codex `config.toml`
-  marker-block + prompts/; Gemini `extensions/ultra-builder-pro/` package);
+  marker-block + prompts/; plus the then-supported compatibility package);
   `docs/RUNTIME-COMPAT-MATRIX.md` with 10-section capability matrix;
   install / uninstall round-trip tests + idempotency.
 - **Phase 3 — thin-shell commands** (D40, `b3d1797`): 9 commands migrated
@@ -232,7 +306,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   skill manifest, CLI protocol + mapping table; 5 spec validators.
 - **Phase 0 — skeleton**: multi-runtime installer scaffolding.
 
-[Unreleased]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.5.3...HEAD
+[Unreleased]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.7.0...HEAD
+[0.7.0]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.6.0...v0.7.0
+[0.6.0]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.5.3...v0.6.0
 [0.5.3]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.5.2...v0.5.3
 [0.5.2]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.5.1...v0.5.2
 [0.5.1]: https://github.com/rocky2431/ultra-builder-pro-cli/compare/v0.5.0...v0.5.1

package/README.md

@@ -2,13 +2,13 @@
 
 Multi-runtime plugin suite + autonomous coding factory for the Ultra Builder Pro
 agent engineering system. It ships native plugins for **Claude Code · OpenCode ·
-Codex**, retains a compatibility adapter for Gemini CLI, and orchestrates PRD →
+Codex** and orchestrates PRD →
 dependency graph → parallel session execution → auto-merge with a single
 authoritative `.ultra/state.db`.
 
 <div align="center">
 
-[![Version](https://img.shields.io/badge/version-0.5.3-blue)](./CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-0.7.0-blue)](./CHANGELOG.md)
 [![Tests](https://img.shields.io/badge/tests-passing-brightgreen)](#verification)
 [![License: MIT](https://img.shields.io/badge/license-MIT-green)](./LICENSE)
 [![Node](https://img.shields.io/badge/node-%E2%89%A522-informational)](./package.json)
@@ -23,18 +23,23 @@ authoritative `.ultra/state.db`.
 
 - **Builds host-native plugins from one allowlist.** Claude Code, OpenCode, and
   Codex receive their own command/skill, agent, hook, and MCP representation.
-  Gemini remains a compatibility adapter; uninstall is symmetric.
+  Install and uninstall are symmetric on every supported host.
 - **Shares state across runtimes.** `.ultra/state.db` (SQLite + WAL) is the
-  authoritative source for tasks, sessions, events, and telemetry. `tasks.json`
-  and context markdown are generated projections, not handwritten.
+  authoritative source for changes, tasks, sessions, events, incidents,
+  projection jobs, and telemetry. `tasks.json` and context markdown are
+  generated projections, not handwritten.
 - **Keeps memory ownership explicit.** Ultra Builder Pro does not collect
   prompts, transcripts, observations, summaries, or cross-session memory.
   Install cloud-mem/claude-mem separately if persistent memory is wanted.
 - **Runs real PRDs end-to-end.** `task.parse_prd` → `lib/topo.cjs` waves →
   `.ultra/execution-plan.json` → parallel worktree sessions → auto-merge back.
-- **Observes without overhead.** Per-task / per-session / per-runtime token
-  and cost telemetry; live code-review-graph watcher on file save;
-  subscribe-events cursor for real-time dashboards.
+- **Keeps daily work convergent.** `/ultra-change` creates a bounded delta and
+  context manifest; `/ultra-deliver` requires docs/test/review evidence and
+  baseline reconciliation before archive. `/ultra-doctor` reports incidents,
+  projection lag, orphan sessions, and backup-first mechanical recovery.
+- **Keeps external context external.** Memory and code-graph providers retain
+  their own content; Ultra stores only provider/project/revision/status metadata
+  references in a compiled change context.
 
 ## Quickstart
 
@@ -43,7 +48,6 @@ authoritative `.ultra/state.db`.
 npx ultra-builder-pro-cli --claude   --local
 npx ultra-builder-pro-cli --opencode --local
 npx ultra-builder-pro-cli --codex    --local
-npx ultra-builder-pro-cli --gemini   --local
 
 # Or blanket-install to every supported runtime you have
 npx ultra-builder-pro-cli --all --local
@@ -58,9 +62,8 @@ npx ultra-builder-pro-cli --all --local --uninstall
 After install, start a new host session/task and point it at the project.
 Claude Code and OpenCode expose native command forms. Codex exposes the same workflows as namespaced plugin
 skills such as `$ultra-builder-pro:ultra-init`, `$ultra-builder-pro:ultra-plan`,
-and `$ultra-builder-pro:ultra-dev`; `command-map.json` records the nine legacy
-command mappings (`ultra-review` remains a directly invocable skill). Gemini
-retains its compatibility command form.
+and `$ultra-builder-pro:ultra-dev`; `command-map.json` records the eleven legacy
+command mappings (`ultra-review` remains a directly invocable skill).
 See [`docs/RUNTIME-COMPAT-MATRIX.md`](./docs/RUNTIME-COMPAT-MATRIX.md)
 for per-runtime capabilities.
 
@@ -70,30 +73,31 @@ for per-runtime capabilities.
 |-------|---------|----------------|
 | **skill** (`skills/ultra-*/`) | Knowledge — prompts, workflows, prose | Runtime's native skill/prompt loader picks them up after install |
 | **MCP** (`mcp-server/`) | Authoritative state — reads/writes `.ultra/state.db` via stdio JSON-RPC | Primary path for task / session / event / plan operations |
-| **CLI** (`ultra-tools`, `bin/*`) | Shell fallback for CI and non-MCP contexts | `ultra-tools task init-project`, `ubp-orchestrator run`, `ultra-tools status --cost` |
+| **CLI** (`ultra-tools`, `bin/*`) | Selected initialization, recovery, diagnostics, and orchestration surfaces | `ultra-tools task init-project`, `ultra-tools system doctor`, `ubp-orchestrator run` |
 
-The three layers share one `.ultra/state.db`. See
+The layers share one `.ultra/state.db`, but the CLI is not a mirror of every MCP
+tool: continuous change mutations are MCP-only and fail closed. See
 [`docs/ARCHITECTURE.md`](./docs/ARCHITECTURE.md) for the full contract and
-[`spec/cli-protocol.md`](./spec/cli-protocol.md) for the 21 live contracts.
+[`spec/cli-protocol.md`](./spec/cli-protocol.md) for the 29 live contracts.
 Review, impact discovery, skill resolution, and user interaction stay on each
 Host's native surfaces; the generated Codex capability map documents those
 replacements without advertising non-existent MCP tools.
 
-The package boundary is deliberate: ten public Ultra workflow skills, four
+The package boundary is deliberate: twelve public Ultra workflow skills, four
 internal review-rule skills, and host-specific collaboration companions. Browser,
 deployment, skill-discovery, and framework guidance belong to their original plugins.
 
 ## Runtime capability matrix
 
-| Feature                      | Claude Code | OpenCode | Codex CLI | Gemini CLI |
-|------------------------------|:-----------:|:--------:|:---------:|:----------:|
-| Custom commands              | ✅          | ✅       | ✅        | ✅         |
-| Skill loader                 | ✅          | ✅       | ✅ (personal plugin) | ✅ |
-| MCP server (stdio)           | ✅          | ✅       | ✅ (plugin `.mcp.json`) | ✅ |
-| Workflow hooks               | ✅ (native plugin) | ✅ (native JS plugin) | ✅ (native plugin) | ⚠︎ compatibility only |
-| Sub-agents                   | ✅          | ✅       | ✅ (9 native TOML agents) | ⚠︎ |
-| Session worktree isolation   | ✅ (all runtimes; driven by `orchestrator/session-runner.cjs`) | ✅ | ✅ | ✅ |
-| Parallel dispatch + auto-merge | ✅ (`ubp-orchestrator run`) | ✅ | ✅ | ✅ |
+| Feature                      | Claude Code | OpenCode | Codex CLI |
+|------------------------------|:-----------:|:--------:|:---------:|
+| Custom commands              | ✅          | ✅       | ✅        |
+| Skill loader                 | ✅          | ✅       | ✅ (personal plugin) |
+| MCP server (stdio)           | ✅          | ✅       | ✅ (plugin `.mcp.json`) |
+| Workflow hooks               | ✅ (native plugin) | ✅ (native JS plugin) | ✅ (native plugin) |
+| Sub-agents                   | ✅          | ✅       | ✅ (9 native TOML agents) |
+| Session worktree isolation   | ✅ (driven by `orchestrator/session-runner.cjs`) | ✅ | ✅ |
+| Parallel dispatch + auto-merge | ✅ (`ubp-orchestrator run`) | ✅ | ✅ |
 
 Full details in [`docs/RUNTIME-COMPAT-MATRIX.md`](./docs/RUNTIME-COMPAT-MATRIX.md)
 (10 sections, with per-runtime deviations called out).
@@ -101,30 +105,33 @@ Full details in [`docs/RUNTIME-COMPAT-MATRIX.md`](./docs/RUNTIME-COMPAT-MATRIX.m
 ## Typical workflow
 
 ```bash
-# 1. Initialize a project (writes .ultra/ skeleton; state.db stays lazy)
+# 1. Initialize a project (writes .ultra/ skeleton and initializes state.db)
 ultra-tools task init-project --name myapp
 
 # 2. Turn a PRD into a task graph + execution plan (human-gate via dry-run)
 #    (invoked by /ultra-plan skill or via MCP task.parse_prd + plan.export)
 
 # 3. Run the plan — parallel sessions, auto-merge back to main on success
-ubp-orchestrator run --with-graph-watcher
+ubp-orchestrator run
 
-# 4. Monitor cost and progress
+# 4. Monitor cost, progress, and runtime health
 ultra-tools status
 ultra-tools status --cost --since 24h
 ultra-tools session list --json
+ultra-tools system doctor
 ```
 
 Or let the skills drive it. In Codex, invoke
 `$ultra-builder-pro:ultra-plan` → `$ultra-builder-pro:ultra-dev` →
-`$ultra-builder-pro:ultra-status`; other runtimes retain their native command form.
+`$ultra-builder-pro:ultra-status`; after the baseline is delivered, daily work
+starts with `$ultra-builder-pro:ultra-change`. Other runtimes retain their
+native command form.
 
 ## CLI surface
 
 | Binary | Purpose |
 |--------|---------|
-| `ultra-builder-pro-cli` / `ubp` | Installer — `--claude / --opencode / --codex / --gemini / --all`, `--local / --global`, `--uninstall`, `--skip-rtk` |
+| `ultra-builder-pro-cli` / `ubp` | Installer — `--claude / --opencode / --codex / --all`, `--local / --global`, `--uninstall` |
 | `ubp-orchestrator` | Session dispatch daemon — `run`, `start`, `stop`, `status` |
 | `ultra-tools` | State-layer CLI — `task`, `session`, `status`, `db`, `migrate`; explicit `legacy-memory` archive/prune migration |
 | `ubp-handbook` | Preview/apply the managed Ultra contract in `CLAUDE.md` / `AGENTS.md`, with backup |
@@ -171,7 +178,7 @@ Individual suites: `test:state`, `test:orch`, `test:spec`, `test:rest`.
 
 | Doc | What's in it |
 |-----|--------------|
-| [`docs/PLAN.zh-CN.md`](./docs/PLAN.zh-CN.md) | Authoritative 12-phase execution plan (zh-CN) — decisions, risks, timeline |
+| [`docs/PLAN.zh-CN.md`](./docs/PLAN.zh-CN.md) | Historical 12-phase plan and decision log with a current-boundary overlay |
 | [`docs/ROADMAP.md`](./docs/ROADMAP.md) | One-page English roadmap + phase status |
 | [`docs/ARCHITECTURE.md`](./docs/ARCHITECTURE.md) | Single-page system architecture |
 | [`docs/AGENT-CONTEXT.md`](./docs/AGENT-CONTEXT.md) | Canonical runtime context contract |
@@ -179,7 +186,6 @@ Individual suites: `test:state`, `test:orch`, `test:spec`, `test:rest`.
 | [`docs/RUNTIME-COMPAT-MATRIX.md`](./docs/RUNTIME-COMPAT-MATRIX.md) | Per-runtime capability matrix |
 | [`docs/STATE-DB-ACCESS-POLICY.md`](./docs/STATE-DB-ACCESS-POLICY.md) | Multi-process write contract |
 | [`docs/COMMIT-HASH-BACKFILL.md`](./docs/COMMIT-HASH-BACKFILL.md) | Two-commit task-completion flow |
-| [`docs/LEGACY-HERMES.md`](./docs/LEGACY-HERMES.md) | Archived pre-CLI "Hermes 6.6" documentation |
 | [`CHANGELOG.md`](./CHANGELOG.md) | v0.1 → current release notes |
 
 ## License

package/adapters/_shared/codex-assets.cjs

@@ -66,7 +66,6 @@ const COMMAND_NAMES = Object.freeze(CORE_PUBLIC_SKILLS.filter((name) => name !==
 const SKILL_REFERENCE_NAMES = Object.freeze([
   ...COMMAND_NAMES,
   'cc-collab',
-  'gemini-collab',
   'ultra-review',
   'ultra-verify',
 ]);
@@ -119,52 +118,12 @@ Do not call native Codex subagents through this skill; use Codex subagent orches
 parallel work and reserve \`cc-collab\` for an explicitly requested cross-model perspective.
 `;
 
-const GEMINI_COLLAB_BODY = `# Gemini Collaboration for Codex
-
-Use Gemini as an advisory second model. Codex remains responsible for scope, source verification,
-decisions, and the final answer.
-
-## Preflight
-
-1. Confirm \`gemini\` is installed and capture \`gemini --version\`.
-2. Confirm the target workspace and requested scope.
-3. Form Codex's initial analysis before reading Gemini's answer when independence matters.
-
-## Invoke safely
-
-Run Gemini in non-interactive, read-only plan mode:
-
-\`\`\`bash
-gemini --approval-mode plan --output-format text -p "<bounded prompt>"
-\`\`\`
-
-Use a timeout appropriate to the task. Redirect large output to a temporary file when needed, then
-read only the relevant result. Do not use auto-edit or yolo modes. Do not place secrets, tokens,
-unrelated files, or unbounded repository content in the prompt.
-
-For repository analysis, give Gemini the exact files, diff range, question, and expected response
-shape. If the CLI is missing, unauthenticated, times out, or returns empty output, report the degraded
-path and continue with Codex evidence rather than blocking.
-
-## Verify and synthesize
-
-Treat Gemini output as an untrusted analysis artifact:
-
-1. Check every consequential claim against the current checkout, runtime, or primary source.
-2. Separate agreements, useful dissent, and unsupported claims.
-3. Resolve scope or version mismatches before comparing conclusions.
-4. Present Codex's final judgment, not a transcript of two models.
-
-Read the bundled references only when a mode or CLI detail is needed. Never let model agreement
-substitute for tests, runtime evidence, or authoritative documentation.
-`;
-
 const ULTRA_VERIFY_BODY = String(function ultraVerifyBodyTemplate() { /*
-# Ultra Verify — Codex-Primary Three-Way Verification
+# Ultra Verify — Codex-Primary Two-Model Verification
 
 Codex owns the task, writes the first independent analysis, verifies every consequential claim,
-and produces the final synthesis. Claude Code and Gemini are read-only external advisors. Use this
-workflow only when the user asks for cross-model or three-way verification.
+and produces the final synthesis. Claude Code is a read-only external advisor. Use this workflow
+only when the user asks for cross-model verification.
 
 ## Modes
 
@@ -175,13 +134,11 @@ workflow only when the user asks for cross-model or three-way verification.
 
 ## Preconditions
 
-1. Confirm `claude --version` and `gemini --version`.
+1. Confirm `claude --version` plus authentication.
 2. Define the exact workspace, scope, evidence standard, and expected answer shape.
-3. Do not send secrets, unrelated files, or an unbounded home directory to either advisor.
-
-## Workflow
+3. Do not send secrets, unrelated files, or an unbounded home directory to the advisor.
 
-### 1. Create the session and write Codex's independent view
+## 1. Create the session and write Codex's independent view
 
 ```bash
 SESSION_ID="$(date +%Y%m%d-%H%M%S)-verify-<mode>"
@@ -189,63 +146,47 @@ SESSION_PATH=".ultra/collab/${SESSION_ID}"
 mkdir -p "${SESSION_PATH}"
 ```
 
-Before invoking either advisor, write the evidence-backed Codex analysis to
-`${SESSION_PATH}/codex-analysis.md`. This ordering is mandatory because it prevents the external
-answers from priming the primary analysis.
-
-### 2. Launch Claude Code and Gemini concurrently
+Write the evidence-backed Codex analysis to `${SESSION_PATH}/codex-analysis.md` before reading
+the external answer. Record the mode, scope, checkout, and evidence boundary.
 
-Give both advisors the same bounded raw question and evidence, without Codex's conclusions.
+## 2. Launch Claude Code read-only
 
-Claude Code:
+Give Claude Code the bounded raw question and evidence without Codex's conclusions.
 
 ```bash
-claude --safe-mode -p "<BOUNDED_PROMPT>" \\
-  --permission-mode plan \\
-  --tools "Read,Grep,Glob,Bash" \\
-  --output-format text \\
-  --no-session-persistence \\
-  > "${SESSION_PATH}/claude-output.md" \\
+claude --safe-mode -p "<BOUNDED_PROMPT>" \
+  --permission-mode plan \
+  --tools "Read,Grep,Glob,Bash" \
+  --output-format text \
+  --no-session-persistence \
+  > "${SESSION_PATH}/claude-output.md" \
   2> "${SESSION_PATH}/claude-error.log"
 ```
 
-Gemini:
-
-```bash
-gemini --approval-mode plan --output-format text -p "<BOUNDED_PROMPT>" \\
-  > "${SESSION_PATH}/gemini-output.md" \\
-  2> "${SESSION_PATH}/gemini-error.log"
-```
-
-Start the two commands with parallel `exec_command` calls and a short initial yield. If a command
-returns a live session id, poll it with `write_stdin` in bounded intervals. Do not use YOLO,
-auto-edit, permission bypass, or write-capable external tools.
+Run the command in a yielded shell session. Do not enable permission bypass, mutation tools,
+background writes, or session persistence.
 
-### 3. Collect only completed outputs
+## 3. Collect completed output
 
-Wait for both CLI sessions to finish or reach their explicit timeout. The bundled
-`scripts/verify_wait.py` can verify file stability for automation, but Codex must not perform one
-blocking wait longer than 60 seconds. Read an output only when it is non-empty; retain the matching
-error log as evidence when a CLI fails.
+The bundled `scripts/verify_wait.py` checks output stability. Poll yielded sessions in bounded
+intervals and never hold one blocking tool call longer than 60 seconds. Read the output only when
+it is non-empty; retain the error log when the advisor fails.
 
-### 4. Verify and synthesize
+## 4. Verify and synthesize
 
-Compare `codex-analysis.md`, `claude-output.md`, and `gemini-output.md`:
+Compare `codex-analysis.md` and `claude-output.md`:
 
 1. Verify claims against the current checkout, runtime, tests, or primary documentation.
-2. Separate consensus, majority views, useful dissent, and unsupported assertions.
-3. Explain scope, version, or assumption differences before scoring agreement.
+2. Separate verified agreement, useful dissent, and unsupported assertions.
+3. Explain scope, version, or assumption differences before judging the result.
 4. Write `${SESSION_PATH}/synthesis.md` and `metadata.json`, then present one Codex-owned answer.
 
-Use the bundled confidence rules. Three agreeing, independently verified views are consensus; two
-are a majority; three materially different answers are no consensus and require decomposition or
-more evidence. Model agreement never overrides failing tests or authoritative runtime evidence.
+Model agreement never overrides failing tests or authoritative runtime evidence.
 
 ## Degraded operation
 
-- One advisor fails: continue with Codex plus the available advisor and name the missing view.
-- Both advisors fail: return Codex-only analysis with an explicit single-source warning.
-- Never block the user's task solely because an external CLI is absent, unauthenticated, or slow.
+- Advisor failure: return Codex-only analysis with an explicit single-source warning.
+- Never block the user's task solely because the external CLI is absent, unauthenticated, or slow.
 
 ## Session files
 
@@ -254,70 +195,16 @@ more evidence. Model agreement never overrides failing tests or authoritative ru
   codex-analysis.md
   claude-output.md
   claude-error.log
-  gemini-output.md
-  gemini-error.log
   metadata.json
   synthesis.md
 ```
 
-Read the bundled references for mode-specific prompts, scoring, metadata, and automation details.
-*/ }).match(/\/\*([\s\S]*?)\*\//)[1].trim();
-
-const ULTRA_VERIFY_FLOW = String(function ultraVerifyFlowTemplate() { /*
-# Codex-Native Orchestration Flow
-
-## 1. Primary analysis
-
-Create `.ultra/collab/<SESSION_ID>/` and write `codex-analysis.md` before reading any external
-answer. Record the mode, scope, checkout, and evidence boundary.
-
-## 2. Parallel advisors
-
-Launch Claude Code in `--safe-mode --permission-mode plan` and Gemini in
-`--approval-mode plan` through separate `exec_command` calls. Redirect their output to
-`claude-output.md` and `gemini-output.md`. If a call yields a session id, poll with
-`write_stdin` in bounded intervals; do not grant either advisor mutation authority.
-
-## 3. Completion gate
-
-Proceed only after both sessions have completed or timed out. For automation, run:
-
-```bash
-python3 ~/plugins/ultra-builder-pro/skills/ultra-verify/scripts/verify_wait.py \\
-  "${SESSION_PATH}" --timeout 1200
-```
-
-Run the waiter as a yielded exec session and poll it; never hold one blocking tool call for more
-than 60 seconds. Its JSON reports `claude` and `gemini` as `complete`, `failed`, `empty`, or
-`pending`.
-
-## 4. Synthesis
-
-Read `codex-analysis.md` plus each completed advisor output. Verify consequential claims, compute
-consensus using `confidence-system.md`, and write `synthesis.md` plus:
-
-```json
-{
-  "id": "<SESSION_ID>",
-  "agent": "ultra-verify",
-  "mode": "<mode>",
-  "models": {"codex": "<primary>", "claude": "<advisor>", "gemini": "<advisor>"},
-  "scope": "<scope>",
-  "timestamp": "<ISO 8601>",
-  "confidence": "<consensus|majority|no_consensus>",
-  "degraded": false
-}
-```
-
-If one advisor fails, cap the result at majority and name the missing perspective. If both fail,
-set `degraded: true`, `agents_responded: ["codex"]`, and mark the result single-source.
+Read the bundled references for mode-specific evidence and confidence guidance.
 */ }).match(/\/\*([\s\S]*?)\*\//)[1].trim();
-
 function titleCase(name) {
   const special = {
     'ai-collab-base': 'AI Collaboration Base',
     'cc-collab': 'Claude Code Collaboration',
-    'gemini-collab': 'Gemini Collaboration',
     'use-railway': 'Use Railway',
   };
   if (special[name]) return special[name];
@@ -600,8 +487,6 @@ and does not require an interactive MCP or legacy CLI menu.
 }
 
 function adaptUltraVerifyAsset(input, rel) {
-  if (rel === path.join('references', 'orchestration-flow.md')) return ULTRA_VERIFY_FLOW;
-
   let text = adaptHostText(String(input), '');
   if (rel === path.join('scripts', 'verify_wait.py')) {
     text = text.replaceAll('Codex', 'Claude Code');
@@ -621,7 +506,6 @@ function adaptUltraVerifyAsset(input, rel) {
   text = text.replaceAll('"codex": "<model>"', '"claude": "<advisor>"');
   text = text.replaceAll('["claude"]', '["codex"]');
   text = text.replaceAll('Claude Code-only', 'Codex-only');
-  text = text.replaceAll('Gemini and Claude Code in parallel (native parallel execution)', 'Claude Code and Gemini concurrently');
   text = text.replaceAll('run_in_background', 'native concurrent execution');
   text = text.replaceAll('Bash tool', '`exec_command`');
   text = text.replaceAll('Read tool', 'targeted file reads');
@@ -661,7 +545,7 @@ function adaptHostText(input, skillName = '') {
     text = adaptCodexPrimaryText(text, skillName);
   }
 
-  if (skillName === 'gemini-collab' || skillName === 'ai-collab-base') {
+  if (skillName === 'ai-collab-base') {
     text = text.replaceAll("Claude's", "Codex's");
     text = text.replaceAll('Claude-only', 'Codex-only');
     text = text.replaceAll('Claude ', 'Codex ');
@@ -716,8 +600,7 @@ function adaptHostText(input, skillName = '') {
 function adaptedDescription(sourceDescription, targetName) {
   const special = {
     'cc-collab': 'Ask Claude Code for an independent read-only analysis while Codex owns verification and synthesis. Use only when the user explicitly requests CC or Claude Code collaboration.',
-    'gemini-collab': 'Ask Gemini CLI for an independent read-only analysis while Codex verifies and synthesizes. Use only when the user explicitly requests Gemini collaboration.',
-    'ultra-verify': 'Run Codex-primary three-way verification with read-only Claude Code and Gemini advisors, then verify and synthesize the evidence.',
+    'ultra-verify': 'Run Codex-primary two-model verification with a read-only Claude Code advisor, then verify and synthesize the evidence.',
     'ultra-dev': 'Execute one Ultra task with Codex-native TDD, persistent task.update state, workflow checkpoints, and the native Ultra review agents.',
     'learn': 'Extract one reusable pattern from the current Codex task into a valid user skill, with explicit user approval before writing.',
   };
@@ -730,7 +613,6 @@ function buildSkillMarkdown(sourceText, sourceName, targetName) {
   const { fm, body } = parseFrontmatter(sourceText);
   let adaptedBody;
   if (targetName === 'cc-collab') adaptedBody = CC_COLLAB_BODY;
-  else if (targetName === 'gemini-collab') adaptedBody = GEMINI_COLLAB_BODY;
   else if (targetName === 'ultra-verify') adaptedBody = ULTRA_VERIFY_BODY;
   else adaptedBody = adaptHostText(body, targetName);
   return yaml.dump({
@@ -745,7 +627,6 @@ function buildOpenAiYaml(name, description) {
     COMMAND_NAMES.includes(name)
     || INTERNAL_SKILLS.has(name)
     || name === 'cc-collab'
-    || name === 'gemini-collab'
     || name.startsWith('ultra-')
   );
   const short = description.replace(/\s+/g, ' ').trim().slice(0, 63).replace(/[\s.,;:]+$/, '');

package/adapters/_shared/frontmatter.cjs

@@ -1,8 +1,8 @@
 'use strict';
 
 // YAML frontmatter parsing / serialization for markdown assets.
-// Every runtime adapter consumes this — OpenCode lowercases keys,
-// Gemini extracts description for toml, Claude preserves verbatim.
+// Every runtime adapter consumes this — OpenCode lowercases keys while
+// Claude Code and Codex preserve their native metadata contracts.
 
 const yaml = require('js-yaml');