@su-record/vibe 2.9.3 → 2.9.5

package/skills/tool-fallback/SKILL.md

@@ -1,104 +1,104 @@
----
-name: tool-fallback
-tier: optional
-description: "Tool failure fallback strategies with circuit breaker. Auto-activates on API errors, search failures, timeouts, 429, 5xx, overloaded errors."
-triggers: [API error, search failure, timeout, 429, 5xx, overloaded, fallback, circuit breaker]
-priority: 80
----
-
-# Tool Fallback Strategies
-
-## Pre-check (K1)
-
-> Did a tool just fail? If the error is a simple typo or wrong path, fix the input first. This skill is for persistent failures (429, 5xx, timeouts).
-
-## Circuit Breaker State Machine
-
-```
-┌─────────┐   3 failures   ┌─────────┐   30s cooldown   ┌─────────────┐
-│ CLOSED  │ ─────────────→ │  OPEN   │ ───────────────→ │ HALF-OPEN   │
-│ (normal)│                │ (block) │                  │ (test 1 req)│
-└─────────┘                └─────────┘                  └─────────────┘
-     ↑                                                        │
-     │                      success                           │
-     └────────────────────────────────────────────────────────┘
-```
-
-| State | Behavior |
-|-------|----------|
-| CLOSED | Normal operation, count failures |
-| OPEN | Skip tool immediately, use alternative |
-| HALF-OPEN | Allow 1 test request after cooldown |
-
-## Decision Trees
-
-### Web Search Fails
-
-```
-Web Search fails (429, 529, timeout)
-  → Check circuit state
-  → OPEN? → Skip to alternative immediately
-  → CLOSED? → Try context7 for library docs
-  → Still fails? → Claude's built-in knowledge (last resort)
-```
-
-### External LLM Fails
-
-```
-VibeOrchestrator.smartRoute({ type, prompt })
-  → Primary LLM fails (429, 401, 5xx)
-  → Skip to secondary LLM (no retry on rate limit)
-  → Secondary fails → Claude handles directly
-```
-
-### File/Code Not Found
-
-```
-Glob fails → Expand pattern: *.ts → **/*.ts → **/*
-  → Use Grep for content-based search
-  → Check git log for file history
-```
-
-## Error Response Actions
-
-| Error | Action | Circuit Impact |
-|-------|--------|---------------|
-| 429 Rate Limit | Skip to next alternative (don't retry) | +1 failure |
-| 5xx Server Error | Retry with backoff, then switch | +1 failure |
-| 529 Overloaded | Wait and retry once | +1 failure |
-| Timeout | Split request or retry | +1 failure |
-| 401/403 Auth | Re-auth or switch alternative | Don't count |
-
-## Retry Strategy
-
-```
-Request → Check circuit
-  ├─ OPEN → Use alternative immediately
-  └─ CLOSED/HALF-OPEN → Try request
-       ├─ Success → Reset failure count
-       └─ Fail → Backoff (2s → 4s → 8s)
-            └─ All retries failed → +1 failure
-                 └─ failures ≥ 3 → OPEN circuit (30s)
-                      └─ Use alternative
-```
-
-## LLM Priority by Task Type
-
-| Task Type | Primary → Secondary → Fallback |
-|-----------|-------------------------------|
-| architecture, debugging | GPT → Gemini → Claude |
-| uiux, code-analysis | Gemini → GPT → Claude |
-| code-gen, general | Claude only |
-
-## Principles
-
-1. **Never stop** — always find an alternative
-2. **Try before asking** — exhaust alternatives before asking user
-3. **Fail fast** — skip OPEN-circuit tools immediately
-4. **Auto-recover** — test after 30s cooldown
-
-## Done Criteria (K4)
-
-- [ ] Work continued despite tool failure
-- [ ] Alternative tool/method used successfully
-- [ ] No unnecessary retries on rate-limited tools
+---
+name: tool-fallback
+tier: optional
+description: "Tool failure fallback strategies with circuit breaker. Auto-activates on API errors, search failures, timeouts, 429, 5xx, overloaded errors."
+triggers: [API error, search failure, timeout, 429, 5xx, overloaded, fallback, circuit breaker]
+priority: 80
+---
+
+# Tool Fallback Strategies
+
+## Pre-check (K1)
+
+> Did a tool just fail? If the error is a simple typo or wrong path, fix the input first. This skill is for persistent failures (429, 5xx, timeouts).
+
+## Circuit Breaker State Machine
+
+```
+┌─────────┐   3 failures   ┌─────────┐   30s cooldown   ┌─────────────┐
+│ CLOSED  │ ─────────────→ │  OPEN   │ ───────────────→ │ HALF-OPEN   │
+│ (normal)│                │ (block) │                  │ (test 1 req)│
+└─────────┘                └─────────┘                  └─────────────┘
+     ↑                                                        │
+     │                      success                           │
+     └────────────────────────────────────────────────────────┘
+```
+
+| State | Behavior |
+|-------|----------|
+| CLOSED | Normal operation, count failures |
+| OPEN | Skip tool immediately, use alternative |
+| HALF-OPEN | Allow 1 test request after cooldown |
+
+## Decision Trees
+
+### Web Search Fails
+
+```
+Web Search fails (429, 529, timeout)
+  → Check circuit state
+  → OPEN? → Skip to alternative immediately
+  → CLOSED? → Try context7 for library docs
+  → Still fails? → Claude's built-in knowledge (last resort)
+```
+
+### External LLM Fails
+
+```
+VibeOrchestrator.smartRoute({ type, prompt })
+  → Primary LLM fails (429, 401, 5xx)
+  → Skip to secondary LLM (no retry on rate limit)
+  → Secondary fails → Claude handles directly
+```
+
+### File/Code Not Found
+
+```
+Glob fails → Expand pattern: *.ts → **/*.ts → **/*
+  → Use Grep for content-based search
+  → Check git log for file history
+```
+
+## Error Response Actions
+
+| Error | Action | Circuit Impact |
+|-------|--------|---------------|
+| 429 Rate Limit | Skip to next alternative (don't retry) | +1 failure |
+| 5xx Server Error | Retry with backoff, then switch | +1 failure |
+| 529 Overloaded | Wait and retry once | +1 failure |
+| Timeout | Split request or retry | +1 failure |
+| 401/403 Auth | Re-auth or switch alternative | Don't count |
+
+## Retry Strategy
+
+```
+Request → Check circuit
+  ├─ OPEN → Use alternative immediately
+  └─ CLOSED/HALF-OPEN → Try request
+       ├─ Success → Reset failure count
+       └─ Fail → Backoff (2s → 4s → 8s)
+            └─ All retries failed → +1 failure
+                 └─ failures ≥ 3 → OPEN circuit (30s)
+                      └─ Use alternative
+```
+
+## LLM Priority by Task Type
+
+| Task Type | Primary → Secondary → Fallback |
+|-----------|-------------------------------|
+| architecture, debugging | GPT → Gemini → Claude |
+| uiux, code-analysis | Gemini → GPT → Claude |
+| code-gen, general | Claude only |
+
+## Principles
+
+1. **Never stop** — always find an alternative
+2. **Try before asking** — exhaust alternatives before asking user
+3. **Fail fast** — skip OPEN-circuit tools immediately
+4. **Auto-recover** — test after 30s cooldown
+
+## Done Criteria (K4)
+
+- [ ] Work continued despite tool failure
+- [ ] Alternative tool/method used successfully
+- [ ] No unnecessary retries on rate-limited tools

package/skills/tool-fallback/rubrics/fallback-chain.md

@@ -1,58 +1,58 @@
-# Tool Fallback Priority Chain
-
-## Web / Documentation Search
-
-```
-1. Web Search (Brave/Tavily)
-2. context7 plugin          ← library/framework docs
-3. Claude built-in knowledge ← last resort, may be stale
-```
-
-**Skip to step 2 immediately** for library/API questions — context7 is faster and more accurate than web search for docs.
-
-## Code/File Discovery
-
-```
-1. Glob (pattern match by name)
-2. Grep (content search)
-3. git log --all -- <path>  ← file was deleted or renamed
-4. Ask user for path        ← exhausted all options
-```
-
-## LLM Routing
-
-| Task Type | Primary | Secondary | Fallback |
-|-----------|---------|-----------|----------|
-| Architecture, debugging | GPT | Gemini | Claude direct |
-| UI/UX, code analysis | Gemini | GPT | Claude direct |
-| Code generation, general | Claude | — | — |
-
-**On 429 / rate limit**: skip to secondary immediately — no retry on rate-limited primary.
-
-## External API Calls
-
-```
-1. Try primary endpoint
-2. On 429: skip to alternative, mark circuit OPEN
-3. On 5xx: retry once with backoff (2s), then skip
-4. On timeout: retry once with smaller payload, then skip
-5. On 401/403: re-auth, do NOT count as circuit failure
-```
-
-## Circuit States
-
-| State | Trigger | Recovery |
-|-------|---------|----------|
-| CLOSED | Normal | — |
-| OPEN | 3 failures | Auto after 30s cooldown |
-| HALF-OPEN | After cooldown | 1 test request; success → CLOSED |
-
-## Never-Stop Rule
-
-Every chain must have a terminal fallback that produces **some** output. Acceptable terminal fallbacks:
-
-- Partial result with `TODO` annotation
-- Claude's own knowledge with explicit uncertainty note
-- Ask the user for the missing piece (with specific question)
-
-**Never** return empty-handed without exhausting the chain first.
+# Tool Fallback Priority Chain
+
+## Web / Documentation Search
+
+```
+1. Web Search (Brave/Tavily)
+2. context7 plugin          ← library/framework docs
+3. Claude built-in knowledge ← last resort, may be stale
+```
+
+**Skip to step 2 immediately** for library/API questions — context7 is faster and more accurate than web search for docs.
+
+## Code/File Discovery
+
+```
+1. Glob (pattern match by name)
+2. Grep (content search)
+3. git log --all -- <path>  ← file was deleted or renamed
+4. Ask user for path        ← exhausted all options
+```
+
+## LLM Routing
+
+| Task Type | Primary | Secondary | Fallback |
+|-----------|---------|-----------|----------|
+| Architecture, debugging | GPT | Gemini | Claude direct |
+| UI/UX, code analysis | Gemini | GPT | Claude direct |
+| Code generation, general | Claude | — | — |
+
+**On 429 / rate limit**: skip to secondary immediately — no retry on rate-limited primary.
+
+## External API Calls
+
+```
+1. Try primary endpoint
+2. On 429: skip to alternative, mark circuit OPEN
+3. On 5xx: retry once with backoff (2s), then skip
+4. On timeout: retry once with smaller payload, then skip
+5. On 401/403: re-auth, do NOT count as circuit failure
+```
+
+## Circuit States
+
+| State | Trigger | Recovery |
+|-------|---------|----------|
+| CLOSED | Normal | — |
+| OPEN | 3 failures | Auto after 30s cooldown |
+| HALF-OPEN | After cooldown | 1 test request; success → CLOSED |
+
+## Never-Stop Rule
+
+Every chain must have a terminal fallback that produces **some** output. Acceptable terminal fallbacks:
+
+- Partial result with `TODO` annotation
+- Claude's own knowledge with explicit uncertainty note
+- Ask the user for the missing piece (with specific question)
+
+**Never** return empty-handed without exhausting the chain first.

package/skills/typescript-advanced-types/SKILL.md

@@ -1,67 +1,67 @@
----
-name: typescript-advanced-types
-tier: core
-description: "TypeScript advanced type system master guide. Covers Generics, Conditional Types, Mapped Types, Template Literals, Utility Types. Activates for complex type logic, reusable type utilities, and compile-time type safety."
-sections:
-  - name: "Core Concepts"
-    triggers: [generic, generics, constraint, conditional, infer, mapped, utility, Pick, Omit, Partial, Required, Record, Exclude, Extract, readonly]
-  - name: "Advanced Patterns"
-    triggers: [template literal, branded, opaque, phantom, discriminated union, exhaustive, builder pattern, recursive type, variadic tuple]
-  - name: "Type Inference"
-    triggers: [inference, infer, type narrowing, type guard, satisfies, const assertion]
-  - name: "Best Practices"
-    triggers: [best practice, type safety, convention]
-  - name: "Type Testing"
-    triggers: [type test, type assertion, expectType, tsd]
-  - name: "Common Mistakes"
-    triggers: [mistake, pitfall, error, wrong]
-  - name: "Performance"
-    triggers: [performance, compile time, type complexity, depth limit]
----
-
-# TypeScript Advanced Types
-
-## Pre-check (K1)
-
-> Is this a VIBE-specific type safety issue? Standard TypeScript generics/conditionals/mapped types are well-known — read the code instead. This skill exists for VIBE's strict rules and non-obvious gotchas only.
-
-## VIBE Forbidden Patterns
-
-| Forbidden | Why | Use Instead |
-|-----------|-----|-------------|
-| `any` | Disables all type checking | `unknown` + type guard |
-| `as any` | Bypasses type system | Define proper interface |
-| `@ts-ignore` | Hides real errors | Fix the type issue at root |
-| Raw generic `T` without constraint | Too permissive | `T extends SomeInterface` |
-
-## `unknown` + Type Guard (VIBE's `any` Replacement)
-
-```typescript
-// Instead of: function process(data: any)
-function process(data: unknown): string {
-  if (typeof data === 'string') return data.toUpperCase();
-  if (isUser(data)) return data.name;
-  throw new Error('Unexpected data type');
-}
-
-function isUser(value: unknown): value is User {
-  return typeof value === 'object' && value !== null && 'name' in value;
-}
-```
-
-## Gotchas
-
-| Gotcha | Why Non-obvious | Fix |
-|--------|----------------|-----|
-| Distributive conditionals | `ToArray<string \| number>` → `string[] \| number[]`, not `(string \| number)[]` | Wrap in tuple: `[T] extends [U]` |
-| Recursive type depth | TS has ~50 level depth limit | Use tail-recursive patterns or simplify |
-| `satisfies` vs `as const` | `satisfies` validates + keeps literal types; `as` skips validation | Prefer `satisfies` |
-| Type assertions (`as X`) | Skips checking entirely | Use type guards instead |
-| `interface` vs `type` | `interface` gives better error messages for objects | Use `type` for unions/complex, `interface` for shapes |
-
-## Done Criteria (K4)
-
-- [ ] No `any`, `as any`, or `@ts-ignore` in code
-- [ ] All `unknown` usages have corresponding type guards
-- [ ] Complex types have `AssertEqual` type tests
-- [ ] Recursive types are bounded (no infinite depth)
+---
+name: typescript-advanced-types
+tier: core
+description: "TypeScript advanced type system master guide. Covers Generics, Conditional Types, Mapped Types, Template Literals, Utility Types. Activates for complex type logic, reusable type utilities, and compile-time type safety."
+sections:
+  - name: "Core Concepts"
+    triggers: [generic, generics, constraint, conditional, infer, mapped, utility, Pick, Omit, Partial, Required, Record, Exclude, Extract, readonly]
+  - name: "Advanced Patterns"
+    triggers: [template literal, branded, opaque, phantom, discriminated union, exhaustive, builder pattern, recursive type, variadic tuple]
+  - name: "Type Inference"
+    triggers: [inference, infer, type narrowing, type guard, satisfies, const assertion]
+  - name: "Best Practices"
+    triggers: [best practice, type safety, convention]
+  - name: "Type Testing"
+    triggers: [type test, type assertion, expectType, tsd]
+  - name: "Common Mistakes"
+    triggers: [mistake, pitfall, error, wrong]
+  - name: "Performance"
+    triggers: [performance, compile time, type complexity, depth limit]
+---
+
+# TypeScript Advanced Types
+
+## Pre-check (K1)
+
+> Is this a VIBE-specific type safety issue? Standard TypeScript generics/conditionals/mapped types are well-known — read the code instead. This skill exists for VIBE's strict rules and non-obvious gotchas only.
+
+## VIBE Forbidden Patterns
+
+| Forbidden | Why | Use Instead |
+|-----------|-----|-------------|
+| `any` | Disables all type checking | `unknown` + type guard |
+| `as any` | Bypasses type system | Define proper interface |
+| `@ts-ignore` | Hides real errors | Fix the type issue at root |
+| Raw generic `T` without constraint | Too permissive | `T extends SomeInterface` |
+
+## `unknown` + Type Guard (VIBE's `any` Replacement)
+
+```typescript
+// Instead of: function process(data: any)
+function process(data: unknown): string {
+  if (typeof data === 'string') return data.toUpperCase();
+  if (isUser(data)) return data.name;
+  throw new Error('Unexpected data type');
+}
+
+function isUser(value: unknown): value is User {
+  return typeof value === 'object' && value !== null && 'name' in value;
+}
+```
+
+## Gotchas
+
+| Gotcha | Why Non-obvious | Fix |
+|--------|----------------|-----|
+| Distributive conditionals | `ToArray<string \| number>` → `string[] \| number[]`, not `(string \| number)[]` | Wrap in tuple: `[T] extends [U]` |
+| Recursive type depth | TS has ~50 level depth limit | Use tail-recursive patterns or simplify |
+| `satisfies` vs `as const` | `satisfies` validates + keeps literal types; `as` skips validation | Prefer `satisfies` |
+| Type assertions (`as X`) | Skips checking entirely | Use type guards instead |
+| `interface` vs `type` | `interface` gives better error messages for objects | Use `type` for unions/complex, `interface` for shapes |
+
+## Done Criteria (K4)
+
+- [ ] No `any`, `as any`, or `@ts-ignore` in code
+- [ ] All `unknown` usages have corresponding type guards
+- [ ] Complex types have `AssertEqual` type tests
+- [ ] Recursive types are bounded (no infinite depth)