code-ai-installer 4.0.1-b → 4.0.1-c

package/domains/development/locales/en/prompt-examples.md

@@ -1,152 +1,66 @@
-# Prompt Examples for Pipeline Launch
+# Prompt examples — Development domain
 
-> Copy the relevant prompt, replace placeholders with your data, paste into chat.
+> Copy the prompt you need, fill in your data, send it to the chat.
 
----
+## Choosing the mode
 
-## 🔵 Full Pipeline (`/start-task`) — new feature / refactoring
+| Task | Command |
+|------|---------|
+| New feature / refactor / new API or screen | `/start-task` — Full (PM → UX → ARCH → DEV → REV → OPS → TEST → RG) |
+| Bug in existing functionality (> 2 files or changes an API contract) | `/bugfix` (DEV → REV → TEST) |
+| Small fix, 1–2 files, blast radius ≈ 0 | `/hotfix` (DEV → RG) |
+| Unsure about the mode | describe the task — Conductor classifies it via the Decision Tree |
 
-### New Feature
-```
-@AGENTS.md Task:
-Implement [feature description].
-Full pipeline!
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md from the protocol skills → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate or skill = BLOCKER
-3. Do not proceed without Approved
-```
-
-### Refactoring
-```
-@AGENTS.md Task:
-Refactor [description: what exactly we're refactoring and why].
-Full pipeline!
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md from the protocol skills → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate or skill = BLOCKER
-3. Do not proceed without Approved
-```
-
----
+## Full Pipeline — `/start-task`
 
-## 🟢 Bugfix (`/bugfix`) — bug fix (4 gates)
-
-### Standard Bug
-```
-@AGENTS.md Task:
-Fix bug: [bug description, reproduction steps, expected vs actual behavior].
-Bugfix pipeline!
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md from the protocol skills → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate or skill = BLOCKER
-3. Do not proceed without Approved
-```
-
-### Bug with File References
+### New feature
 ```
-@AGENTS.md Task:
-@[path/to/file1] @[path/to/file2]
-Bug: [description]. Root cause likely in the referenced files.
-Bugfix pipeline!
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate = BLOCKER
+/start-task Implement [feature description]. Acceptance criteria: [what must work].
 ```
 
----
-
-## 🟡 Hotfix (`/hotfix`) — minor fix (2 gates)
-
-### Typo / CSS / Single Line
+### Refactor
 ```
-@AGENTS.md Task:
-Hotfix: [fix description, e.g.: "fix typo in heading" or "change button color from #333 to #555"].
-Hotfix pipeline!
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate = BLOCKER
+/start-task Refactor [what exactly and why]. Target state: [...]. Do not touch: [...].
 ```
 
----
-
-## 🎨 Design Only (Google Stitch)
-
-### UI Screen Generation
+### Feature with UI
 ```
-@AGENTS.md Task:
-Create UI screens for [description: which screens, which application].
-Use Google Stitch via $stitch-prompting-guide (workflow) + $stitch-api-reference (MCP API).
-
-SESSION RULES:
-1. view_file agents/ux_ui_designer.md → view_file $stitch-prompting-guide → deliverable → notify_user → wait for "Approved"
+/start-task [screen/component description]. UX: [interface requirements, a11y]. @[path to the mockup or current component]
 ```
 
----
-
-## 📄 Documentation Only
-
-### Document Adaptation / Creation
+## Bugfix — `/bugfix`
 ```
-@AGENTS.md Task:
-[Documentation task description, e.g.: "adapt files X, Y, Z for global use"].
-Bugfix pipeline! (or Full pipeline — depending on complexity)
-
-SESSION RULES:
-1. Each gate: view_file agents/<role>.md → view_file on EVERY SKILL.md → deliverable → notify_user → wait for "Approved"
-2. Skipping a gate = BLOCKER
+/bugfix Bug: [description]. Repro steps: [...]. Expected vs actual: [...]. @[path to the file — if you know where]
 ```
 
----
-
-## ⚡ Gate Transitions (User Responses)
-
-### Approve and Move to Next Gate
+## Hotfix — `/hotfix`
 ```
-Approved. Next gate: agents/<role>.md. Skills: view_file required. Full format.
+/hotfix [small fix: typo in a heading / button color #333 → #555 / one line].
 ```
 
-### Approve with Trigger (Wix / Shopify)
-```
-Approved. TEST gate. Wix.
-```
+## How the work flows
 
-### Reject / Return for Rework
-```
-NO-GO. Fix: [list of issues]. Return to [gate] gate.
-```
+- The pipeline is executed by the **code-ai MCP** machine. On every gate: `current_gate` → `classify_gate` → role via `load_role` + required skills via `get_skill` → deliverable + `submit_artifact` → `sign_off` → `advance_gate`.
+- **Silent-by-default:** you are asked only on judgment gates — **PM / UX / ARCH / RG** — and on a `fork` escalation (via `request_decision`). DEV/REV/OPS/TEST pass on their own; OPS/TEST are signed after a green auto-check (`build` / `run_tests`).
+- **Fix Cycle:** a red auto-check or a review/test FAIL → rollback to DEV. 2 rollbacks in a row across the DEV/OPS/REV/TEST band → circuit breaker: an architect audit in a fresh session.
 
----
+> Full rules — in `pipeline-rules.md`.
 
-## 🔧 Utility Prompts
+## Gate replies
 
-### Check Status
 ```
-Show the current Master Checklist and status of all gates.
+Approved
 ```
-
-### Run a Specific Agent
 ```
-Run agents/<role>.md. Context: [brief description]. Skills: view_file required.
+Approved. [clarification / condition]
 ```
-
-### Update Version
 ```
-Update the version in package.json to X.Y.Z
+NO-GO. Fix: [list of issues]. Return to gate [PM/UX/ARCH/...].
 ```
 
----
-
-## 📝 Tips for Writing Prompts
+## Tips
 
-1. **Always start with `@AGENTS.md`** — this loads the agent system
-2. **Reference files with `@[path]`** — the agent immediately sees the context
-3. **Name the pipeline mode** — "Full pipeline!", "Bugfix pipeline!", "Hotfix pipeline!"
-4. **Add SESSION RULES** — this guarantees protocol compliance
-5. **Write "Approved" explicitly** — without this word the agent will not proceed to the next gate
-6. **Be specific** — the more precise the description, the better the result
+1. **Name the mode** explicitly (`/start-task` / `/bugfix` / `/hotfix`) or let the Conductor classify it.
+2. **Reference files via `@path`** — the agent sees the context right away.
+3. **The more precise the description and acceptance criteria** — the fewer Fix Cycle iterations.
+4. **On judgment gates** (PM/UX/ARCH/RG) the decision is yours; the machine doesn't pull you into routine.

package/domains/development/pipeline.yaml

@@ -1,135 +1,150 @@
-# pipeline.yaml — development domain state machine
-#
-# Source of truth for code-ai-mcp advance_gate. Schema:
-#   the shared types module (src/shared) → PipelineYaml zod.
-#
-# 3 modes match docs/.agents/workflows/{pipeline-rules,bugfix,hotfix}.md.
-# CONDUCTOR isn't a GateName — it's the orchestrator role that owns mode-routing
-# (classified outside this state machine via the conductor agent).
-#
-# All gates currently use sign_off_policy=den because the Phase-2 auto-check
-# tools (run_tests, check_lint, build, dependency_supply_chain, ...) are still
-# stubs. Once those tools are wired, OPS/REV/TEST gates flip to mcp_auto_pass
-# or either.
-#
-# CIRCUIT BREAKER CONVENTION: when dev_rollback_threshold trips, the audit is
-# expected to run in a FRESH Claude Code session — not as a subagent inside the
-# stuck session. Provenance: wing_codex_product/decisions 2026-05-23, "SUBAGENTS
-# FOR CONSISTENCY CHECKS — qualitatively stronger than self-check, clean context,
-# no my bias." Same Claude that hit the wall reproduces the blind spot that
-# caused the rollback. advance_gate's error message carries this recommendation.
-
-schema_version: 1
-domain: development
-default_mode: full
-default_signer: den
-
-modes:
-  - mode: full
-    gates:
-      - gate: PM
-        produced_by: [product_manager]
-        signs_off: [product_manager, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: UX
-        produced_by: [ux_ui_designer]
-        signs_off: [ux_ui_designer, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: ARCH
-        produced_by: [architect]
-        signs_off: [architect, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: DEV
-        produced_by: [senior_full_stack]
-        signs_off: [senior_full_stack, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: REV
-        produced_by: [reviewer]
-        signs_off: [reviewer, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: OPS
-        produced_by: [devops]
-        signs_off: [devops, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: TEST
-        produced_by: [tester]
-        signs_off: [tester, conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: RG
-        produced_by: [conductor]
-        signs_off: [conductor]
-        sign_off_policy: den
-        classify_required: true
-        retryable: false
-
-  - mode: bugfix
-    gates:
-      - gate: DEV
-        produced_by: [senior_full_stack]
-        signs_off: [senior_full_stack]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: REV
-        produced_by: [reviewer]
-        signs_off: [reviewer]
-        sign_off_policy: den
-        classify_required: true
-        retryable: true
-      - gate: TEST
-        produced_by: [tester]
-        signs_off: [tester]
-        sign_off_policy: den
-        classify_required: true
-        retryable: false
-
-  - mode: hotfix
-    gates:
-      # Combined DEV+TEST: senior_full_stack writes the fix AND the test;
-      # tester co-signs to confirm verification ran. Single gate keeps the
-      # touchpoint_budget.hotfix = 1 invariant.
-      - gate: DEV
-        produced_by: [senior_full_stack]
-        signs_off: [senior_full_stack, tester]
-        sign_off_policy: den
-        classify_required: true
-        retryable: false
-      - gate: RG
-        produced_by: [conductor]
-        signs_off: [conductor]
-        sign_off_policy: den
-        classify_required: false
-        retryable: false
-
-circuit_breaker:
-  enabled: true
-  dev_rollback_threshold: 2
-  audit_role: architect
-  audit_required_skills:
-    - system-design-checklist
-    - current-state-analysis
-    - design-patterns-reference
-  reset_on:
-    - rg_pass
-    - new_arch_adr
-
-touchpoint_budget:
-  hotfix: 1
-  bugfix_midsize: [1, 2]
-  feature_backend: [1, 3]
-  feature_ui_with_arch_fork: [3, 5]
-  complex_3plus_fc: [3, 6]
+# pipeline.yaml — development domain state machine
+#
+# Source of truth for code-ai-mcp advance_gate. Schema:
+#   the shared types module (src/shared) → PipelineYaml zod.
+#
+# 3 modes match domains/development/.agents/workflows/{pipeline-rules,bugfix,hotfix}.md.
+# CONDUCTOR isn't a GateName — it's the orchestrator role that owns mode-routing
+# (classified outside this state machine via the conductor agent).
+#
+# SIGN-OFF MODEL (reduced intervention, silent-by-default):
+#   PM / UX / ARCH / RG → sign_off_policy=user. The judgment gates the user owns:
+#     what/why we build (PM/UX/ARCH) and the final release (RG).
+#   DEV / REV → sign_off_policy=either. Claude signs as 'mcp' from agent judgment
+#     (no deterministic tool — DEV is authoring, REV is a review judgment). the user is
+#     not pulled in routinely; only via request_decision for a genuine fork.
+#   OPS / TEST → sign_off_policy=either WITH auto_check. Claude runs the named MCP
+#     check tool (build / run_tests); green → signs 'mcp' silently; red → classify
+#     + rollback to DEV (Fix Cycle). the user sees these only if escalated.
+#
+# RED / ROLLBACK / CIRCUIT BREAKER: a failure on DEV/OPS/REV/TEST rolls back to DEV
+# (Fix Cycle), NOT to the user. advance_gate counts each rollback-to-DEV; on the 2nd
+# (dev_rollback_threshold) the breaker REFUSES and routes to an architect deep audit
+# instead — expected to run in a FRESH Claude Code session (clean-context principle;
+# the same Claude that hit the wall reproduces the blind spot that caused the
+# rollback). Provenance: wing_codex_product/decisions 2026-05-23 "SUBAGENTS FOR
+# CONSISTENCY CHECKS". Reset on rg_pass or a new ARCH ADR (record_decision).
+
+schema_version: 1
+domain: development
+default_mode: full
+default_signer: user
+
+modes:
+  - mode: full
+    gates:
+      - gate: PM
+        produced_by: [product_manager]
+        signs_off: [product_manager, conductor]
+        sign_off_policy: user
+        classify_required: true
+        retryable: true
+      - gate: UX
+        produced_by: [ux_ui_designer]
+        signs_off: [ux_ui_designer, conductor]
+        sign_off_policy: user
+        classify_required: true
+        retryable: true
+      - gate: ARCH
+        produced_by: [architect]
+        signs_off: [architect, conductor]
+        sign_off_policy: user
+        classify_required: true
+        retryable: true
+      - gate: DEV
+        produced_by: [senior_full_stack]
+        signs_off: [senior_full_stack, conductor]
+        sign_off_policy: either
+        classify_required: true
+        retryable: true
+      - gate: REV
+        produced_by: [reviewer]
+        signs_off: [reviewer, conductor]
+        sign_off_policy: either
+        classify_required: true
+        retryable: true
+      - gate: OPS
+        produced_by: [devops]
+        signs_off: [devops, conductor]
+        sign_off_policy: either
+        auto_check:
+          tool: build
+        classify_required: true
+        retryable: true
+      - gate: TEST
+        produced_by: [tester]
+        signs_off: [tester, conductor]
+        sign_off_policy: either
+        auto_check:
+          tool: run_tests
+        classify_required: true
+        retryable: true
+      - gate: RG
+        produced_by: [conductor]
+        signs_off: [conductor]
+        sign_off_policy: user
+        classify_required: true
+        retryable: false
+
+  - mode: bugfix
+    gates:
+      - gate: DEV
+        produced_by: [senior_full_stack]
+        signs_off: [senior_full_stack]
+        sign_off_policy: either
+        classify_required: true
+        retryable: true
+      - gate: REV
+        produced_by: [reviewer]
+        signs_off: [reviewer]
+        sign_off_policy: either
+        classify_required: true
+        retryable: true
+      - gate: TEST
+        produced_by: [tester]
+        signs_off: [tester]
+        sign_off_policy: either
+        auto_check:
+          tool: run_tests
+        classify_required: true
+        retryable: false
+
+  - mode: hotfix
+    gates:
+      # Combined DEV+TEST: senior_full_stack writes the fix AND the test;
+      # tester co-signs to confirm verification ran. Single gate keeps the
+      # touchpoint_budget.hotfix = 1 invariant. auto_check=run_tests lets Claude
+      # sign 'mcp' on green; PM/UX/ARCH judgment gates are skipped in hotfix.
+      - gate: DEV
+        produced_by: [senior_full_stack]
+        signs_off: [senior_full_stack, tester]
+        sign_off_policy: either
+        auto_check:
+          tool: run_tests
+        classify_required: true
+        retryable: false
+      - gate: RG
+        produced_by: [conductor]
+        signs_off: [conductor]
+        sign_off_policy: user
+        classify_required: false
+        retryable: false
+
+circuit_breaker:
+  enabled: true
+  dev_rollback_threshold: 2
+  audit_role: architect
+  audit_required_skills:
+    - system-design-checklist
+    - current-state-analysis
+    - design-patterns-reference
+  reset_on:
+    - rg_pass
+    - new_arch_adr
+
+touchpoint_budget:
+  hotfix: 1
+  bugfix_midsize: [1, 2]
+  feature_backend: [1, 3]
+  feature_ui_with_arch_fork: [3, 5]
+  complex_3plus_fc: [3, 6]