feat(arch): composite skill architecture + SEO reference implementation

docs/CODEX-PARITY-HANDOFF.md

@@ -2,7 +2,7 @@
 
 This document is generated from the runtime-neutral registry in [scripts/lib/runtime-targets.ts](../scripts/lib/runtime-targets.ts).
 
-Current snapshot: 80 agents, 41 commands, 17 hooks, 13 templates, 28 tests.
+Current snapshot: 80 agents, 41 commands, 17 hooks, 14 templates, 28 tests.
 
 ## Runtime Targets
 

skills/pos-ads/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: pos-ads
+description: "Ads composite — audit campaigns, create ad copy, optimize bids, and report performance with persistent campaign memory. Replaces 20 fragmented ads skills."
+argument-hint: "[audit|create|optimize|report] [platform or campaign]"
+---
+
+# pos-ads
+
+Ads composite — audit campaigns, create ad copy, optimize bids, and report performance with persistent campaign memory. Replaces 20 fragmented ads skills.
+
+**Replaces:** ads-apple, ads-audit, ads-budget, ads-competitor, ads-create, ads-creative, ads-dna, ads-generate, ads-google, ads-landing, ads-linkedin, ads-meta, ads-microsoft, ads-photoshoot, ads-plan, ads-tiktok, ads-youtube, unified-ads, paid-ads, competitive-ads-extractor
+
+## Actions
+
+| Action | What |
+|--------|------|
+| `audit [platform]` | Audit campaign performance across platforms |
+| `create [type]` | Generate ad copy, creatives, and landing page concepts |
+| `optimize` | Budget reallocation and bid optimization recommendations |
+| `report` | Cross-platform performance report with ROI analysis |
+
+## Routing
+
+1. Parse action from first argument
+2. Load domain memory from `~/.productionos/domains/pos-ads/`
+3. If first run: auto-detect project context, create profile
+4. Dispatch to `sub-skills/{action}.md`
+5. Score against `evaluation/rubric.yml`, update memory
+
+## Domain Memory
+
+Stored at `~/.productionos/domains/pos-ads/`:
+profile.yml (platforms, budgets, audiences, goals), campaign-history.jsonl (past campaigns with metrics), creative-bank.jsonl (tested ad copy and results), learnings.jsonl
+
+## Inputs
+
+| Parameter | Values | Default | Description |
+|-----------|--------|---------|-------------|
+| `action` | `audit`, `create`, `optimize`, `report` | `audit` | Sub-skill to run |
+| `target` | path, URL, or description | required | What to operate on |
+
+## Guardrails
+
+1. Memory is append-only. Never delete history.
+2. Every action scored against rubric. No silent completion.
+3. Profile is user-controlled. Never change without asking.
+4. Evidence required. No fabricated metrics or claims.
+5. Learn from history. Compare current output to past executions.

skills/pos-ads/evaluation/rubric.yml

@@ -0,0 +1,55 @@
+domain: ads
+version: 1
+
+actions:
+  audit:
+    passing_score: 8.0
+    dimensions:
+      - name: platform_coverage
+        weight: 0.3
+        passing: 85
+      - name: metric_accuracy
+        weight: 0.35
+        passing: 90
+      - name: recommendation_quality
+        weight: 0.35
+        passing: 85
+  create:
+    passing_score: 8.0
+    dimensions:
+      - name: hook_quality
+        weight: 0.3
+        passing: 80
+      - name: cta_clarity
+        weight: 0.25
+        passing: 85
+      - name: audience_targeting
+        weight: 0.25
+        passing: 80
+      - name: platform_fit
+        weight: 0.2
+        passing: 85
+  optimize:
+    passing_score: 8.0
+    dimensions:
+      - name: data_grounding
+        weight: 0.4
+        passing: 90
+      - name: budget_justification
+        weight: 0.3
+        passing: 85
+      - name: expected_impact
+        weight: 0.3
+        passing: 80
+  report:
+    passing_score: 8.0
+    dimensions:
+      - name: metric_completeness
+        weight: 0.3
+        passing: 85
+      - name: trend_analysis
+        weight: 0.35
+        passing: 80
+      - name: roi_calculation
+        weight: 0.35
+        passing: 90

skills/pos-build/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: pos-build
+description: "Build composite — brainstorm, plan, implement, and test with project memory. Replaces 6 fragmented build skills."
+argument-hint: "[brainstorm|plan|implement|test] [feature or task]"
+---
+
+# pos-build
+
+Build composite — brainstorm, plan, implement, and test with project memory. Replaces 6 fragmented build skills.
+
+**Replaces:** build, build-productionos, tdd, auto-mode, feature-dev, feature-development
+
+## Actions
+
+| Action | What |
+|--------|------|
+| `brainstorm [idea]` | Explore idea before building (delegates to /pos plan brainstorm) |
+| `plan [feature]` | Create step-by-step implementation plan |
+| `implement [task]` | Execute implementation with TDD discipline |
+| `test [target]` | Write tests for existing code or verify implementation |
+
+## Routing
+
+1. Parse action from first argument
+2. Load domain memory from `~/.productionos/domains/pos-build/`
+3. If first run: auto-detect project context, create profile
+4. Dispatch to `sub-skills/{action}.md`
+5. Score against `evaluation/rubric.yml`, update memory
+
+## Domain Memory
+
+Stored at `~/.productionos/domains/pos-build/`:
+profile.yml (tech stack, test framework, conventions), build-log.jsonl (past builds with outcomes), test-patterns.jsonl (what test patterns work for this codebase), learnings.jsonl
+
+## Inputs
+
+| Parameter | Values | Default | Description |
+|-----------|--------|---------|-------------|
+| `action` | `brainstorm`, `plan`, `implement`, `test` | `brainstorm` | Sub-skill to run |
+| `target` | path, URL, or description | required | What to operate on |
+
+## Guardrails
+
+1. Memory is append-only. Never delete history.
+2. Every action scored against rubric. No silent completion.
+3. Profile is user-controlled. Never change without asking.
+4. Evidence required. No fabricated metrics or claims.
+5. Learn from history. Compare current output to past executions.

skills/pos-build/evaluation/rubric.yml

@@ -0,0 +1,49 @@
+domain: build
+version: 1
+
+actions:
+  plan:
+    passing_score: 8.0
+    dimensions:
+      - name: completeness
+        weight: 0.3
+        passing: 85
+      - name: sequencing
+        weight: 0.25
+        passing: 80
+      - name: risk_identification
+        weight: 0.25
+        passing: 85
+      - name: test_strategy
+        weight: 0.2
+        passing: 90
+  implement:
+    passing_score: 8.0
+    dimensions:
+      - name: correctness
+        weight: 0.35
+        passing: 90
+      - name: test_coverage
+        weight: 0.25
+        passing: 80
+      - name: minimal_change
+        weight: 0.2
+        passing: 85
+      - name: style_consistency
+        weight: 0.2
+        passing: 80
+  test:
+    passing_score: 8.0
+    dimensions:
+      - name: coverage
+        weight: 0.3
+        passing: 85
+      - name: behavioral_not_structural
+        weight: 0.3
+        passing: 85
+      - name: edge_cases
+        weight: 0.2
+        passing: 80
+      - name: readability
+        weight: 0.2
+        passing: 85

skills/pos-content/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: pos-content
+description: "Content composite — strategy, writing, audit, and refresh with brand voice memory. Replaces 10 fragmented content skills."
+argument-hint: "[strategy|write|audit|refresh] [topic or URL]"
+---
+
+# pos-content
+
+Content composite — strategy, writing, audit, and refresh with brand voice memory. Replaces 10 fragmented content skills.
+
+**Replaces:** content-gap-analysis, content-ops, content-quality-auditor, content-refresher, content-research-writer, content-strategy, unified-content, copywriting, copy-editing, social-content
+
+## Actions
+
+| Action | What |
+|--------|------|
+| `strategy` | Content strategy with audience analysis and calendar planning |
+| `write [topic]` | Create content (article, social post, email, landing page) |
+| `audit [url]` | Audit existing content for quality, freshness, and SEO alignment |
+| `refresh [url]` | Update stale content with current data and improved structure |
+
+## Routing
+
+1. Parse action from first argument
+2. Load domain memory from `~/.productionos/domains/pos-content/`
+3. If first run: auto-detect project context, create profile
+4. Dispatch to `sub-skills/{action}.md`
+5. Score against `evaluation/rubric.yml`, update memory
+
+## Domain Memory
+
+Stored at `~/.productionos/domains/pos-content/`:
+profile.yml (brand voice, audience, tone, platforms), content-calendar.jsonl (published + planned content), performance.jsonl (what content performed well), learnings.jsonl
+
+## Inputs
+
+| Parameter | Values | Default | Description |
+|-----------|--------|---------|-------------|
+| `action` | `strategy`, `write`, `audit`, `refresh` | `strategy` | Sub-skill to run |
+| `target` | path, URL, or description | required | What to operate on |
+
+## Guardrails
+
+1. Memory is append-only. Never delete history.
+2. Every action scored against rubric. No silent completion.
+3. Profile is user-controlled. Never change without asking.
+4. Evidence required. No fabricated metrics or claims.
+5. Learn from history. Compare current output to past executions.

skills/pos-content/evaluation/rubric.yml

@@ -0,0 +1,55 @@
+domain: content
+version: 1
+
+actions:
+  strategy:
+    passing_score: 8.0
+    dimensions:
+      - name: audience_clarity
+        weight: 0.3
+        passing: 85
+      - name: channel_fit
+        weight: 0.3
+        passing: 80
+      - name: calendar_specificity
+        weight: 0.4
+        passing: 85
+  write:
+    passing_score: 8.0
+    dimensions:
+      - name: quality
+        weight: 0.3
+        passing: 85
+      - name: seo_alignment
+        weight: 0.25
+        passing: 80
+      - name: brand_voice
+        weight: 0.25
+        passing: 85
+      - name: readability
+        weight: 0.2
+        passing: 90
+  audit:
+    passing_score: 8.0
+    dimensions:
+      - name: coverage
+        weight: 0.3
+        passing: 80
+      - name: actionability
+        weight: 0.35
+        passing: 90
+      - name: freshness_check
+        weight: 0.35
+        passing: 85
+  refresh:
+    passing_score: 8.0
+    dimensions:
+      - name: improvement_delta
+        weight: 0.4
+        passing: 80
+      - name: accuracy
+        weight: 0.3
+        passing: 90
+      - name: seo_preserved
+        weight: 0.3
+        passing: 85

skills/pos-debug/SKILL.md

@@ -0,0 +1,47 @@
+---
+name: pos-debug
+description: "Debug composite — reproduce, hypothesize, test, fix with bug pattern memory. Replaces 4 debug skills."
+argument-hint: "[diagnose|fix|verify] [bug or error]"
+---
+
+# pos-debug
+
+Debug composite — reproduce, hypothesize, test, fix with bug pattern memory. Replaces 4 debug skills.
+
+**Replaces:** debug, unified-debug, systematic-debugging, investigate
+
+## Actions
+
+| Action | What |
+|--------|------|
+| `diagnose [bug]` | Reproduce + hypothesize + test — root cause analysis |
+| `fix [finding]` | Apply minimum fix for confirmed root cause |
+| `verify` | Re-run reproduction + regression test after fix |
+
+## Routing
+
+1. Parse action from first argument
+2. Load domain memory from `~/.productionos/domains/pos-debug/`
+3. If first run: auto-detect project context, create profile
+4. Dispatch to `sub-skills/{action}.md`
+5. Score against `evaluation/rubric.yml`, update memory
+
+## Domain Memory
+
+Stored at `~/.productionos/domains/pos-debug/`:
+bug-history.jsonl (past bugs with root causes), patterns.jsonl (recurring bug patterns in this codebase), learnings.jsonl
+
+## Inputs
+
+| Parameter | Values | Default | Description |
+|-----------|--------|---------|-------------|
+| `action` | `diagnose`, `fix`, `verify` | `diagnose` | Sub-skill to run |
+| `target` | path, URL, or description | required | What to operate on |
+
+## Guardrails
+
+1. Memory is append-only. Never delete history.
+2. Every action scored against rubric. No silent completion.
+3. Profile is user-controlled. Never change without asking.
+4. Evidence required. No fabricated metrics.
+5. Learn from history. Compare to past executions.