comman

commands/implement.md

@@ -0,0 +1,152 @@
+---
+description: Implement iOS/SwiftUI features from eval specs with automatic verification
+argument-hint: <eval-name>
+allowed-tools: Read, Write, Edit, Bash, Grep, Glob, Task
+---
+
+# /implement Command
+
+Implement iOS features from eval specs. Spawns verifier agent for checking - never tests own work.
+
+## Usage
+
+```
+/implement <eval-name>
+```
+
+## Flow
+
+```
+Read eval spec -> Check SwiftUI guide -> Implement -> Spawn verifier -> Fix if needed -> Done
+```
+
+## Process
+
+### 1. Load Eval Spec
+
+Read `.claude/evals/<name>.yaml` and extract `building_spec`:
+- description
+- requirements
+- constraints
+- files
+
+If no eval exists:
+```
+No eval spec found for "<name>".
+Create one first with: /eval <name>
+```
+
+### 2. Check SwiftUI Guide
+
+Read `.claude/axiom-skills-guide.md` for:
+- Project-specific iOS patterns
+- Architecture preferences
+- SwiftUI conventions
+
+If missing, use standard SwiftUI best practices.
+
+### 3. Implement
+
+Build the feature following the spec. iOS/SwiftUI guidelines:
+- `@State`, `@Binding`, `@ObservedObject` appropriately
+- `async/await` over completion handlers
+- `@MainActor` for UI updates
+- Small, composable Views
+- Proper error handling
+
+### 4. Spawn Verifier (MANDATORY)
+
+**NEVER verify own work.** Always spawn eval-verifier agent:
+
+```python
+verifier_result = spawn_agent("eval-verifier", {
+    "spec": f".claude/evals/{name}.yaml"
+})
+```
+
+The verifier:
+- Runs all checks from `verification_spec`
+- Collects evidence
+- Reports pass/fail
+
+### 5. Handle Results
+
+**Pass:** Report success and files changed.
+
+**Fail:** Read failure feedback, fix specific issues, re-spawn verifier.
+
+Max 5 iterations.
+
+## Orchestration Logic
+
+```python
+max_iterations = 5
+iteration = 0
+
+# Read spec
+spec = read_yaml(f".claude/evals/{name}.yaml")
+guide = read_file(".claude/axiom-skills-guide.md")  # optional
+
+# Implement
+implement(spec.building_spec, guide)
+
+while iteration < max_iterations:
+    # ALWAYS spawn verifier - never test own work
+    verifier_result = spawn_agent("eval-verifier", {
+        "spec": f".claude/evals/{name}.yaml"
+    })
+
+    if verifier_result.all_passed:
+        return success(verifier_result)
+
+    # Fix failures
+    fix_issues(verifier_result.failures)
+    iteration += 1
+
+return failure("Max iterations reached")
+```
+
+## Output
+
+```
+/implement auth
+---
+
+[Loading eval spec: .claude/evals/auth.yaml]
+
+Building Spec:
+  - Login view with email/password
+  - Async authentication service
+  - Error handling with alerts
+
+[Checking .claude/axiom-skills-guide.md...]
+
+[Implementing...]
+
+Files Created:
+  + Sources/Features/Auth/LoginView.swift
+  + Sources/Features/Auth/AuthViewModel.swift
+  + Sources/Services/AuthService.swift
+
+Files Modified:
+  ~ Sources/App/ContentView.swift
+
+[Spawning eval-verifier...]
+
+Verification:
+  ✅ file-exists: LoginView.swift
+  ✅ file-contains: @MainActor
+  ✅ ui-login: Form renders correctly
+     📸 Evidence: 2 screenshots
+
+---
+Implementation complete: 3/3 checks passed
+```
+
+## Critical Rules
+
+1. **NEVER test own work** - always spawn eval-verifier
+2. **NEVER skip verification** - even if "confident"
+3. **NEVER claim pass without verifier confirmation**
+4. **Read the spec** - don't assume requirements
+5. **Minimal fixes** - on retry, fix only what failed

skills/implement/SKILL.md

@@ -1,148 +0,0 @@
----
-name: implement
-description: Implement iOS/SwiftUI features from eval specs. Use when building iOS features. Triggers on "/implement", "build ios feature", or "implement [feature] for ios".
-allowed-tools: Read, Write, Edit, Bash, Grep, Glob, Task
----
-
-# Implement Skill (iOS/SwiftUI)
-
-I implement iOS features from eval specs. I build - I do NOT verify.
-
-## Critical Rule
-
-**NEVER test or verify my own work.** After implementation, I spawn the `eval-verifier` agent to check my work. This separation ensures honest verification.
-
-## Workflow
-
-```
-Read eval spec -> Implement -> Spawn eval-verifier -> Report results
-```
-
-### Step 1: Find Eval Spec
-
-Look for the eval in `.claude/evals/<name>.yaml`
-
-If no eval exists, tell user to create one first:
-```
-No eval spec found for "<name>".
-Run: /eval <name>
-to create the building and verification spec first.
-```
-
-### Step 2: Read Building Spec
-
-Extract from eval YAML:
-- `building_spec.description` - what to build
-- `building_spec.requirements` - specific requirements
-- `building_spec.constraints` - rules to follow
-- `building_spec.files` - suggested file paths
-
-### Step 3: Reference SwiftUI Guide
-
-Before implementing, check `.claude/axiom-skills-guide.md` for:
-- SwiftUI patterns and conventions
-- Project-specific iOS guidelines
-- Architecture preferences (MVVM, etc.)
-
-If guide doesn't exist, use standard SwiftUI best practices.
-
-### Step 4: Implement
-
-Build the feature following:
-1. Requirements from eval spec
-2. Constraints from eval spec
-3. SwiftUI best practices
-4. Project conventions
-
-**iOS/SwiftUI Guidelines:**
-- Use `@State`, `@Binding`, `@ObservedObject` appropriately
-- Prefer `async/await` over completion handlers
-- Use `@MainActor` for UI updates
-- Keep Views small and composable
-- Extract reusable components
-- Use proper error handling with `Result` or `throws`
-
-### Step 5: Spawn Verifier
-
-After implementation complete, spawn the eval-verifier agent:
-
-```
-Task tool call:
-  subagent_type: general-purpose
-  prompt: |
-    You are the eval-verifier agent. Read the agent instructions from:
-    agents/eval-verifier.md
-
-    Then verify the implementation against:
-    .claude/evals/<name>.yaml
-
-    Run all verification checks, collect evidence, and report pass/fail.
-```
-
-### Step 6: Handle Results
-
-**If verifier passes:**
-```
-Implementation complete: <name>
-All checks passed.
-
-Files created:
-  + path/to/file.swift
-
-Files modified:
-  ~ path/to/existing.swift
-```
-
-**If verifier fails:**
-Read failure feedback, fix specific issues, re-spawn verifier.
-Max 5 iterations.
-
-## Output Format
-
-```
-Implementing: <name>
----
-
-[Reading eval spec...]
-
-Building Spec:
-  - Requirement 1
-  - Requirement 2
-
-[Implementing...]
-
-Files Created:
-  + Sources/Features/Login/LoginView.swift
-  + Sources/Features/Login/LoginViewModel.swift
-
-Files Modified:
-  ~ Sources/App/ContentView.swift
-
-[Spawning verifier...]
-
----
-Verification Results:
-  [verifier output here]
-```
-
-## Example
-
-User: `/implement auth`
-
-1. Read `.claude/evals/auth.yaml`
-2. Check `.claude/axiom-skills-guide.md` for iOS patterns
-3. Implement:
-   - `LoginView.swift` - SwiftUI form
-   - `LoginViewModel.swift` - business logic
-   - `AuthService.swift` - API calls
-4. Spawn eval-verifier
-5. Report results
-
-## What I Do NOT Do
-
-- Skip reading the eval spec
-- Verify my own work (verifier does this)
-- Collect evidence (verifier does this)
-- Generate tests (verifier does this)
-- Add features not in the spec
-- Assume verification passed without running verifier