feat: add 13 Claude Code harness skills

skills/claude-api/SKILL.md

@@ -0,0 +1,183 @@
+---
+name: claude-api
+description: Build, debug, and optimize Claude API / Anthropic SDK applications. Covers prompt caching, thinking, tool use, batch processing, streaming, file APIs, citations, and model migrations. Trigger when code imports anthropic or @anthropic-ai/sdk.
+origin: claude-code-harness
+tools: Read, Write, Edit, Bash, Glob, Grep
+---
+
+# Claude API
+
+Build, debug, and optimize applications using the Claude API (Anthropic SDK). All apps built with this skill should include prompt caching.
+
+## When to Activate
+
+**Trigger on:**
+- Code that imports `anthropic` or `@anthropic-ai/sdk`
+- User asks about the Claude API, Anthropic SDK, or Managed Agents
+- User adds or modifies Claude features: caching, thinking, tool use, batch, files, citations, memory
+- User asks about a specific model (Opus, Sonnet, Haiku)
+- Questions about prompt caching or cache hit rates in an Anthropic SDK project
+- Migrating between Claude model versions
+
+**Skip when:**
+- Code imports `openai` or another provider SDK
+- Filename contains `-openai.py`, `-generic.py`, or similar
+- Provider-neutral code with no Anthropic-specific patterns
+
+## Current Models (as of 2025)
+
+| Model | ID | Best For |
+|-------|-----|----------|
+| Claude Opus 4.7 | `claude-opus-4-7` | Most capable, complex reasoning |
+| Claude Sonnet 4.6 | `claude-sonnet-4-6` | Balanced capability and speed |
+| Claude Haiku 4.5 | `claude-haiku-4-5-20251001` | Fast, cost-efficient |
+
+Default to the latest and most capable model for new applications.
+
+## Prompt Caching (Required)
+
+Every app built with this skill must include prompt caching. Caching reduces cost and latency for repeated system prompts, tool definitions, and long documents.
+
+### Python
+
+```python
+import anthropic
+
+client = anthropic.Anthropic()
+
+response = client.messages.create(
+    model="claude-opus-4-7",
+    max_tokens=1024,
+    system=[
+        {
+            "type": "text",
+            "text": "Your long system prompt here...",
+            "cache_control": {"type": "ephemeral"}
+        }
+    ],
+    messages=[{"role": "user", "content": "Hello"}]
+)
+```
+
+### TypeScript
+
+```typescript
+import Anthropic from "@anthropic-ai/sdk";
+
+const client = new Anthropic();
+
+const response = await client.messages.create({
+  model: "claude-opus-4-7",
+  max_tokens: 1024,
+  system: [
+    {
+      type: "text",
+      text: "Your long system prompt here...",
+      cache_control: { type: "ephemeral" },
+    },
+  ],
+  messages: [{ role: "user", content: "Hello" }],
+});
+```
+
+### Cache Hit Detection
+
+```python
+# Check cache performance
+print(response.usage.cache_creation_input_tokens)  # tokens written to cache
+print(response.usage.cache_read_input_tokens)       # tokens read from cache
+```
+
+## Tool Use
+
+```python
+tools = [
+    {
+        "name": "get_weather",
+        "description": "Get current weather for a location",
+        "input_schema": {
+            "type": "object",
+            "properties": {
+                "location": {"type": "string", "description": "City name"}
+            },
+            "required": ["location"]
+        }
+    }
+]
+
+response = client.messages.create(
+    model="claude-opus-4-7",
+    max_tokens=1024,
+    tools=tools,
+    messages=[{"role": "user", "content": "What's the weather in Paris?"}]
+)
+```
+
+## Extended Thinking
+
+```python
+response = client.messages.create(
+    model="claude-opus-4-7",
+    max_tokens=16000,
+    thinking={
+        "type": "enabled",
+        "budget_tokens": 10000
+    },
+    messages=[{"role": "user", "content": "Solve this complex problem..."}]
+)
+```
+
+## Streaming
+
+```python
+with client.messages.stream(
+    model="claude-opus-4-7",
+    max_tokens=1024,
+    messages=[{"role": "user", "content": "Tell me a story"}]
+) as stream:
+    for text in stream.text_stream:
+        print(text, end="", flush=True)
+```
+
+## Batch Processing
+
+```python
+# Create a batch
+batch = client.messages.batches.create(
+    requests=[
+        {"custom_id": "req-1", "params": {"model": "claude-opus-4-7", "max_tokens": 100, "messages": [...]}},
+        {"custom_id": "req-2", "params": {"model": "claude-opus-4-7", "max_tokens": 100, "messages": [...]}},
+    ]
+)
+
+# Poll for results
+import time
+while True:
+    batch = client.messages.batches.retrieve(batch.id)
+    if batch.processing_status == "ended":
+        break
+    time.sleep(60)
+```
+
+## Model Migration
+
+When migrating between model versions:
+
+1. Update the `model` string to the new ID
+2. Check for deprecated parameters (some older params removed in newer models)
+3. Test with the new model — behavior may differ even with the same prompt
+4. Update any hardcoded model-specific limits (context window, max output tokens)
+
+## Common Issues
+
+| Issue | Cause | Fix |
+|-------|-------|-----|
+| Low cache hit rate | System prompt changes between calls | Make system prompt static; move dynamic content to user turn |
+| `anthropic.APIStatusError: 529` | Overloaded | Implement exponential backoff |
+| Tool call not executed | Model returned `tool_use` block | Handle `stop_reason: "tool_use"` and call the tool |
+| Thinking not appearing | Budget too low | Increase `budget_tokens`; minimum ~1000 |
+
+## Related Skills
+
+- `agent-architecture-audit` — Audit multi-layer agent applications built on the Claude API
+- `agent-harness-construction` — Build full agent harnesses using the Claude API

skills/code-review/SKILL.md

@@ -0,0 +1,121 @@
+---
+name: code-review
+description: Review the current diff for correctness bugs and reuse/simplification/efficiency cleanups at a given effort level. Pass --comment to post findings as inline PR comments, or --fix to apply findings to the working tree. Supports low/medium/high/max/ultra effort levels.
+origin: claude-code-harness
+tools: Read, Bash, Glob, Grep
+---
+
+# Code Review
+
+Review the current branch diff for correctness bugs, reuse opportunities, simplification wins, and efficiency improvements. Produces ranked findings with optional inline PR comments or automatic fixes.
+
+## When to Use
+
+- User asks for a code review of the current branch or a specific PR
+- User wants findings posted as inline PR comments (`--comment`)
+- User wants findings automatically applied to the working tree (`--fix`)
+- User specifies an effort level: low, medium, high, max, or ultra
+
+**Invoke via:** `/code-review [effort] [--comment] [--fix] [PR#]`
+
+**Ultra mode** (`/code-review ultra` or `/code-review ultra <PR#>`): launches a multi-agent cloud review. Billed and user-triggered only — do not invoke via Bash.
+
+## Effort Levels
+
+| Level | Coverage | Best For |
+|-------|----------|----------|
+| `low` | High-confidence, obvious bugs only | Quick sanity check before merge |
+| `medium` | Moderate coverage, clear wins | Standard PR review |
+| `high` | Broader coverage, may include uncertain findings | Important features or risky changes |
+| `max` | Maximum breadth, includes speculative findings | Critical systems, security-sensitive changes |
+| `ultra` | Deep multi-agent cloud review | Large PRs, pre-release audits |
+
+Default effort: `medium`.
+
+## Review Workflow
+
+### 1. Identify the Diff
+
+```bash
+# Current branch vs main
+git diff main...HEAD
+
+# Specific PR (with gh CLI)
+gh pr diff <PR#>
+
+# Staged changes only
+git diff --cached
+```
+
+### 2. Collect Context
+
+Before reviewing, read:
+- Files changed in the diff
+- Adjacent files that depend on changed code
+- Existing tests for changed functionality
+- Any CLAUDE.md conventions that apply
+
+### 3. Apply Effort-Appropriate Analysis
+
+**Correctness bugs (all levels):**
+- Logic errors, off-by-one errors, null/undefined dereferences
+- Incorrect API usage, wrong argument order
+- Race conditions, improper async/await
+- Missing error handling at system boundaries
+
+**Simplification/reuse (medium+):**
+- Duplicated logic that could use an existing utility
+- Overly complex implementations with simpler alternatives
+- Dead code or unused imports
+
+**Efficiency (high+):**
+- N+1 queries, unnecessary re-renders, wasteful allocations
+- Missing memoization or caching opportunities
+
+**Speculative (max only):**
+- Hypothetical future issues
+- Style suggestions beyond project conventions
+
+### 4. Rank Findings
+
+Order findings by:
+1. **Severity** — bugs that would break production first
+2. **Confidence** — high-confidence findings before speculative ones
+3. **Impact** — changes that affect many callsites before isolated issues
+
+### 5. Output Findings
+
+For each finding:
+- **File and line reference** — `path/to/file.ts:42`
+- **Issue description** — what is wrong and why
+- **Suggested fix** — concrete code change, not vague advice
+
+## Flags
+
+### `--comment` — Post as Inline PR Comments
+
+Posts each finding as an inline comment on the GitHub PR at the relevant line. Requires `gh` CLI authenticated. Does not modify the working tree.
+
+```bash
+gh pr review <PR#> --comment --body "..."
+# or per-line:
+gh api repos/:owner/:repo/pulls/:pr/comments ...
+```
+
+### `--fix` — Apply Findings
+
+After generating findings, applies fixes directly to the working tree. Equivalent to running `/simplify` after the review. Use only when findings are high-confidence.
+
+## What NOT to Review
+
+- Formatting — let the linter handle it
+- Naming preferences with no correctness impact
+- Architectural changes outside the PR scope
+- Hypothetical future requirements (at low/medium effort)
+
+## Related Skills
+
+- `simplify` — Apply fixes from the current diff review
+- `verify` — Confirm a change works by running the app
+- `security-review` — Security-focused review of pending changes
+- `agent-architecture-audit` — Deep audit for agent/LLM applications