Soju06 · Soju06 · Mar 10, 2026 · Mar 10, 2026 · Mar 10, 2026 · Mar 10, 2026
diff --git a/skills/knowpatch/corrections/cli-tools.md b/skills/knowpatch/corrections/cli-tools.md
@@ -3,12 +3,12 @@ ecosystem: cli-tools
 description: CLI tool renames, command changes, major versions
 tags: [shadcn, tailwind, eslint, create-react-app, vite, webpack, prettier, cli]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-24"
+last_updated: "2026-03-10"
 ---
 
 # CLI Tools — Version Corrections
 
-> Last updated: 2026-02-24
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [shadcn (formerly shadcn-ui)](#shadcn)
@@ -19,14 +19,15 @@ last_updated: "2026-02-24"
 
 ---
 
-### shadcn — 2025-10
-- **Outdated**: Package name `shadcn-ui`, install command `npx shadcn-ui@latest init`
+### shadcn — 2026-03
+- **Outdated**: Package name `shadcn-ui`, install command `npx shadcn-ui@latest init`, shadcn 2.x/3.x
 - **Current**:
+  - **shadcn 4.0** (2026-03-06) — Major version bump
   - Package name: `shadcn`
   - Install: `npx shadcn@latest init`
   - The `shadcn-ui` package is discontinued — use `shadcn` package instead
   - shadcn dropped Tailwind CSS 3 support — **Tailwind 4 is required**
-- **Impact**: `npx shadcn-ui@latest` fails or installs a broken version. Using with Tailwind 3 causes incompatibility.
+- **Impact**: `npx shadcn-ui@latest` fails or installs a broken version. Using with Tailwind 3 causes incompatibility. shadcn 3.x is now outdated.
 - **Lookup**: `npm view shadcn version`
 
 ### Tailwind CSS v4 — 2025-03

diff --git a/skills/knowpatch/corrections/frontier-models.md b/skills/knowpatch/corrections/frontier-models.md
@@ -3,12 +3,12 @@ ecosystem: frontier-models
 description: Proprietary frontier AI model names, IDs, SDK versions, multimodal support
 tags: [claude, gpt, gemini, openai, anthropic, model, llm, sdk, ai]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-27"
+last_updated: "2026-03-10"
 ---
 
 # Frontier (Proprietary) Models — Version Corrections
 
-> Last updated: 2026-02-27
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [Anthropic Claude](#anthropic-claude)
@@ -31,27 +31,28 @@ last_updated: "2026-02-27"
 - **Impact**: Using legacy IDs like `claude-3-5-sonnet`, `claude-3-opus` results in deprecated or significantly degraded performance
 - **Lookup**: `npm view @anthropic-ai/sdk version`
 
-### OpenAI GPT-5 Family — 2026-02
+### OpenAI GPT-5 Family — 2026-03
 - **Outdated**: GPT-4 Turbo, GPT-4o are the latest/best models
 - **Current**:
+  - `gpt-5.4` — New flagship (2026-03-05), combines coding + reasoning, native computer-use, 1M ctx
+    - ChatGPT: GPT-5.4 Thinking / Pro
+    - API: `gpt-5.4`, `gpt-5.4-pro`
   - `gpt-5.3-codex` — Best agentic coding model (2026-02-05)
   - `gpt-5.3-codex-spark` — Ultra-fast real-time coding, Cerebras-powered (2026-02-12), text-only 128k
-  - `gpt-5.2` — Professional work (2025-12-11)
-    - ChatGPT: GPT-5.2 Instant / Thinking / Pro
-    - API: `gpt-5.2`, `gpt-5.2-chat-latest`, `gpt-5.2-pro`
-  - `gpt-5.2-codex` — Coding optimized
-  - `gpt-5.1` — Legacy (sunset planned in ChatGPT)
-  - `gpt-5` — Previous generation (still available via API, $1.25/$10 per 1M tokens)
-  - `gpt-5-mini`, `gpt-5-nano` — Small variants
+  - `gpt-5.3-chat-latest` — Everyday conversational (2026-03-03), replaces GPT-5.2 Instant
+  - `gpt-5.2` — Previous flagship (Instant deprecated 2026-06-03)
+    - API: `gpt-5.2`, `gpt-5.2-pro`
+  - `gpt-5.1` and below — Legacy
   - GPT-4 family (4o, 4-turbo, etc.) — Fully legacy
 - **Input**: Text, Image / **Output**: Text (Codex variants are text-only)
-- **Impact**: Using GPT-4 model IDs invokes legacy models with significantly degraded performance
+- **Impact**: Using GPT-4 or GPT-5.2 Instant model IDs invokes deprecated models
 - **Lookup**: `npm view openai version`
 
-### Google Gemini 3 Family — 2026-02
+### Google Gemini 3 Family — 2026-03
 - **Outdated**: Gemini 1.5 Pro is the latest
 - **Current**:
   - Gemini 3.1 Pro (Preview) — Highest capability, 1M input tokens, 64k output tokens
+  - Gemini 3.1 Flash-Lite (Preview) — Fastest/cheapest, $0.25/$1.50 per 1M tokens, 1M ctx, 64k output (2026-03-03)
   - Gemini 3 Pro — Stable version
   - Gemini 3 Flash — Fast variant
   - Gemini 2 family — Previous generation (still available)
@@ -61,18 +62,19 @@ last_updated: "2026-02-27"
 - **Impact**: Using Gemini 1.5 models results in legacy performance; failing to inform about video/audio input support
 - **Lookup**: `npm view @google/genai version`
 
-## Multimodal Input Comparison — 2026-02
+## Multimodal Input Comparison — 2026-03
 
 | Model | Text | Image | Video | Audio | PDF |
 |-------|------|-------|-------|-------|-----|
 | Claude Opus 4.6 | O | O | X | X | O |
 | Claude Sonnet 4.6 | O | O | X | X | O |
 | Claude Haiku 4.5 | O | O | X | X | O |
-| GPT-5.2 (Thinking/Pro) | O | O | X | X | - |
-| GPT-5.2 Instant | O | O | X | X | - |
+| GPT-5.4 (Thinking/Pro) | O | O | X | X | - |
 | GPT-5.3-Codex | O | X | X | X | X |
 | GPT-5.3-Codex-Spark | O | X | X | X | X |
+| GPT-5.3 Instant | O | O | X | X | - |
 | Gemini 3.1 Pro | O | O | O | O | O |
+| Gemini 3.1 Flash-Lite | O | O | O | O | O |
 | Gemini 3 Pro/Flash | O | O | O | O | O |
 | Kimi K2.5 | O | O | X | X | X |
 | MiniMax M2.5 | O | X | X | X | X |

diff --git a/skills/knowpatch/corrections/javascript.md b/skills/knowpatch/corrections/javascript.md
@@ -3,17 +3,17 @@ ecosystem: javascript
 description: JS/TS library API changes
 tags: [zod, react, typescript, npm, bun, deno, pnpm, esm, types, javascript, js, ts]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-24"
+last_updated: "2026-03-10"
 ---
 
 # JavaScript/TypeScript — Version Corrections
 
-> Last updated: 2026-02-24
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [Zod 4](#zod-4)
 - [React 19](#react-19)
-- [TypeScript 5.9](#typescript-59)
+- [TypeScript 5.9 / 6.0](#typescript-59--60)
 
 ---
 
@@ -104,12 +104,13 @@ last_updated: "2026-02-24"
 
 ---
 
-### TypeScript 5.9 — 2025
+### TypeScript 5.9 / 6.0 — 2026-03
 - **Outdated**: TypeScript 5.3-5.5 is the latest
 - **Current**:
-  - TypeScript 5.9
+  - TypeScript 5.9.3 — Current stable
+  - **TypeScript 6.0 RC** (`6.0.1-rc`, 2026-03-06) — Stable release imminent
   - `moduleResolution: "bundler"` now the default
   - `satisfies` operator stabilized
   - `import type` auto-separation improved
-- **Impact**: Low (good backwards compatibility)
-- **Lookup**: `npm view typescript version`
+- **Impact**: TypeScript 6.0 is a major version — verify migration notes when stable. 5.9 code is expected to be mostly compatible.
+- **Lookup**: `npm view typescript version`, `npm view typescript@rc version`
diff --git a/skills/knowpatch/corrections/macos.md b/skills/knowpatch/corrections/macos.md
@@ -3,12 +3,12 @@ ecosystem: macos
 description: macOS 26 version naming, Liquid Glass, Swift 6.2, system toolchain, Apple framework changes
 tags: [macos, tahoe, xcode, swift, swiftui, liquid-glass, metal, rosetta, intel, apple-silicon, foundation-models]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-27"
+last_updated: "2026-03-10"
 ---
 
 # macOS & Apple Platforms — Version Corrections
 
-> Last updated: 2026-02-27
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [macOS Version Naming](#macos-version-naming)
@@ -31,7 +31,7 @@ last_updated: "2026-02-27"
   - macOS 15 Sequoia → **macOS 26 Tahoe** (not 16)
   - Same scheme for iOS 26, iPadOS 26, watchOS 26, tvOS 26, visionOS 26
   - Internal Darwin version: 25.x
-  - Current release: macOS Tahoe 26.3 (as of 2026-02)
+  - Current release: macOS Tahoe 26.3.1 (2026-03-04), 26.4 in beta
 - **Impact**: Referencing "macOS 16" confuses users and produces incorrect deployment target values
 - **Lookup**: `sw_vers` (local), apple.com/macos
 
@@ -136,14 +136,15 @@ last_updated: "2026-02-27"
 - **Impact**: Using system Python/Ruby leads to missing security patches, incompatible packages, and broken tooling
 - **Lookup**: `python3 --version`, `ruby --version` (local). Use Homebrew/pyenv/rbenv for current versions
 
-### Xcode 26 & CLI Tools — 2025-06
+### Xcode 26 & CLI Tools — 2026-03
 - **Outdated**: Xcode 16 with Clang 16, older Git versions
 - **Current**:
   - Xcode 26 requires **macOS Sequoia 15.6+** to run
+  - **Xcode 26.3** (2026-02-26): Built-in AI coding agents — **Claude Agent + OpenAI Codex** integrated via MCP
   - Bundled: **Clang 17.0.0**, **Git 2.50.1**
-  - Compilation caching (opt-in) for faster iterative builds
+  - Compilation caching (opt-in) for faster iterative builds, 40% faster workspace loading
   - `#Playground` macro for interactive code exploration
   - Known: Clang auto-corrects deployment version `16.0` → `26.0` (harmless warning, GCC 15.2+ fixes upstream)
   - Enhanced async debugging in LLDB (async stepping, task context, named tasks)
-- **Impact**: GCC/Fortran formulas in Homebrew may show deployment version warnings. C++ code compiled with Xcode 26 may have minor regressions
+- **Impact**: Xcode 26.3 enables AI-assisted coding natively. GCC/Fortran formulas in Homebrew may show deployment version warnings.
 - **Lookup**: `xcodebuild -version`, `clang --version`, `git --version` (local)
diff --git a/skills/knowpatch/corrections/open-source-models.md b/skills/knowpatch/corrections/open-source-models.md
@@ -3,12 +3,12 @@ ecosystem: open-source-models
 description: Open-source frontier LLMs for coding, agentic tasks, and self-hosting (2026)
 tags: [open-source, self-hosted, coding-model, llama, deepseek, mistral, kimi, minimax, glm, qwen, vllm, sglang, swe-bench, moe, model, llm, ai]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-27"
+last_updated: "2026-03-10"
 ---
 
 # Open-Source Frontier Models — Version Corrections
 
-> Last updated: 2026-02-27
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [Outdated Recommendations](#outdated-recommendations)
@@ -26,7 +26,6 @@ last_updated: "2026-02-27"
 ### Outdated Recommendations — 2026-02
 - **Outdated**: When asked to recommend open-source LLMs, agents suggest:
   - **Llama 3.3 70B** (Meta, 2024-12) — Was frontier, now surpassed on coding/agentic benchmarks
-  - **Mistral Large 2** (Mistral, 2024-07) — Surpassed; no competitive open-weight MoE since
   - **DeepSeek R1** (DeepSeek, 2025-01) — Strong reasoning but predates the 2026 MoE generation
   - Agents also claim open-source models are "significantly behind" proprietary frontier models
 - **Current**: Five MoE models released 2025–2026 match or exceed proprietary models on SWE-bench and agentic benchmarks. See selection guide below.
@@ -44,7 +43,7 @@ last_updated: "2026-02-27"
 | Rank | Model | Active | SWE-bench | Differentiator |
 |------|-------|--------|-----------|----------------|
 | 1 | GLM-5 | 40B | 77.8% | Lowest hallucination, agentic engineering, non-NVIDIA support |
-| 2 | DeepSeek V3.2 | 37B | 73.1% | Cheapest API ($0.028 input), proven DSA architecture |
+| 2 | DeepSeek V3.2 | 37B | 73.1% | Cheapest API, DSA long-context, proven architecture |
 | 3 | Kimi K2.5 | 32B | 76.8% | Native vision, Agent Swarm (100 parallel sub-agents) |
 | 4 | Qwen 3.5 | 17B | 76.4% | Native multimodal, 201 languages, lightest active params |
 
@@ -92,8 +91,9 @@ Note: MiniMax M2.5 (10B active) outperforms larger models on coding via RL speci
 - 201 languages/dialects supported (broadest multilingual coverage)
 - SWE-bench Verified 76.4%, BrowseComp 78.6 (w/ context management)
 - Qwen3.5-Plus: hosted version with 1M context window
-- Best for: multilingual tasks, native multimodal, lightweight inference
-- HuggingFace: `Qwen/Qwen3.5-397B-A17B`
+- **Small series** (2026-03-02): 0.8B/2B/4B/9B open-source (Apache 2.0) — 9B scores 81.7 GPQA Diamond
+- Best for: multilingual tasks, native multimodal, lightweight inference (small series for on-device)
+- HuggingFace: `Qwen/Qwen3.5-397B-A17B`, `Qwen/Qwen3.5-9B`, etc.
 
 ---
 

diff --git a/skills/knowpatch/corrections/platforms.md b/skills/knowpatch/corrections/platforms.md
@@ -3,16 +3,17 @@ ecosystem: platforms
 description: BaaS/platform API key changes, auth patterns
 tags: [supabase, anon, service_role, publishable, secret, jwks, baas, firebase, platform]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-25"
+last_updated: "2026-03-10"
 ---
 
 # Platforms & BaaS — Version Corrections
 
-> Last updated: 2026-02-25
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [Supabase API Keys](#supabase-api-keys)
 - [Supabase JWKS JWT Verification](#supabase-jwks-jwt-verification)
+- [Supabase OpenAPI Schema Access](#supabase-openapi-schema-access)
 
 ---
 
@@ -89,3 +90,15 @@ const { payload } = await jwtVerify(token, jwks, {
 ```
 
 **Lookup**: `npm view jose version`, Supabase Dashboard → Settings → Auth → JWT Signing Keys
+
+---
+
+### Supabase OpenAPI Schema Access — 2026-03-11
+- **Outdated**: Use anon key to fetch OpenAPI schema at `/rest/v1/` for auto-generating types or API clients
+- **Current**:
+  - As of **2026-03-11**, the OpenAPI spec endpoint (`/rest/v1/`) **requires service role or secret API key**
+  - Anon key / publishable key can no longer access the schema endpoint
+  - Affects: PostgREST schema introspection, Supabase type generation, API client auto-gen
+  - Workaround: use `SUPABASE_SECRET_KEY` or service role key for schema fetching
+- **Impact**: Tools and agents that auto-fetch Supabase schema via anon key will silently fail or return 403
+- **Lookup**: Supabase Dashboard → Settings → API Keys
diff --git a/skills/knowpatch/corrections/python.md b/skills/knowpatch/corrections/python.md
@@ -3,12 +3,12 @@ ecosystem: python
 description: Python ecosystem tool/library changes
 tags: [pip, uv, poetry, pydantic, fastapi, django, ruff, flask, sqlalchemy, python]
 version: "0.4.1" # x-release-please-version
-last_updated: "2026-02-24"
+last_updated: "2026-03-10"
 ---
 
 # Python — Version Corrections
 
-> Last updated: 2026-02-24
+> Last updated: 2026-03-10
 
 ## Table of Contents
 - [uv](#uv)
@@ -19,25 +19,32 @@ last_updated: "2026-02-24"
 
 ---
 
-### uv — 2025
+### uv — 2026-03
 - **Outdated**: `pip install` + `python -m venv` or `poetry` is the standard for Python package management
 - **Current**:
-  - uv — Mainstream replacement for pip/poetry
+  - uv 0.10.x — Mainstream replacement for pip/poetry
   - Project init: `uv init`
   - Add dependency: `uv add fastapi`
   - Run: `uv run python app.py`
   - Virtual env: `uv venv` (auto-created)
   - Lock file: `uv.lock` (auto-managed)
   - pip compatible: `uv pip install`, `uv pip compile`
   - Rust-based, 10-100x faster than pip
-- **Impact**: pip/poetry still work but are slow; modern Python projects default to uv
+  - **Breaking in 0.10.0** (2026-02-05):
+    - `uv venv` requires `--clear` to remove existing venvs
+    - Multiple indexes with `default = true` now errors
+    - `uv python upgrade` now stable (was preview)
+- **Impact**: pip/poetry still work but are slow; modern Python projects default to uv. `uv venv` behavior change is a common footgun when upgrading from 0.9.x.
 - **Lookup**: `pip index versions uv | head -1`
 
-### Django 6 — 2025
+### Django 6 — 2026-03
 - **Outdated**: Django 4.x or 5.x is the latest
 - **Current**:
-  - Django 6.0
-- **Impact**: Django 4/5 code is mostly compatible, but installing legacy versions for new projects is inappropriate
+  - Django 6.0.3 (2026-03-03, security release)
+  - CVE-2026-25673: DoS via `URLField` Unicode normalization on Windows
+  - CVE-2026-25674: Incorrect permissions on filesystem objects in multi-threaded environments
+  - Key features: template partials, background tasks (no Celery needed), native CSP middleware
+- **Impact**: Django 4/5 code is mostly compatible. Security patches required — use 6.0.3+
 - **Lookup**: `pip index versions django | head -1`
 
 ### Pydantic v2 — 2023-07
@@ -56,23 +63,26 @@ last_updated: "2026-02-24"
 - **Impact**: v1-style code produces deprecation warnings; full removal is planned
 - **Lookup**: `pip index versions pydantic | head -1`
 
-### ruff — 2024
+### ruff — 2026-03
 - **Outdated**: `flake8` + `black` + `isort` combo is the Python linting standard
 - **Current**:
-  - ruff — All-in-one replacement for flake8 + black + isort
+  - ruff 0.15.x — All-in-one replacement for flake8 + black + isort
   - Linting: `ruff check .`
   - Formatting: `ruff format .`
   - Auto-fix: `ruff check --fix .`
   - Rust-based, 10-100x faster than flake8
   - Config: `[tool.ruff]` section in `pyproject.toml`
+  - Markdown formatting support in LSP (0.15.1+, preview)
 - **Impact**: flake8/black/isort combo works but is slow with fragmented config. Modern projects default to ruff.
 - **Lookup**: `pip index versions ruff | head -1`
 
-### FastAPI — 2024
-- **Outdated**: FastAPI 0.90-0.100 range is the latest
+### FastAPI — 2026-03
+- **Outdated**: FastAPI 0.90-0.100 range is the latest; Pydantic v1 is supported
 - **Current**:
-  - FastAPI
-  - Full Pydantic v2 support
+  - FastAPI 0.129.x
+  - **Pydantic v1 dropped** in 0.126+ — minimum `pydantic >= 2.7.0`
+  - `pydantic.v1` namespace triggers deprecation warning (0.127+)
   - `lifespan` event handler (on_event is deprecated)
-- **Impact**: Low (good backwards compatibility); just use accurate version numbers
+  - 2x+ JSON response performance improvement (0.129+)
+- **Impact**: Code using `from pydantic.v1 import ...` or Pydantic v1 patterns will break on FastAPI 0.126+
 - **Lookup**: `pip index versions fastapi | head -1`