A collection of reusable skills for Claude Code that enhance writing quality, research, media processing, and investigation workflows.
Skills are markdown instruction files that give Claude Code specialized capabilities. Drop them into ~/.claude/skills/ and invoke them with slash commands (e.g., /humanizer, /arese).
git clone https://github.com/kbbrazel/claude-code-skills.git
cp -r claude-code-skills/skills/* ~/.claude/skills/Copy any skill folder into your ~/.claude/skills/ directory:
cp -r claude-code-skills/skills/humanizer ~/.claude/skills/| Skill | Description | Use When |
|---|---|---|
| humanizer | Removes AI writing patterns from text | Editing drafts to sound natural |
| arese | Adversarial multi-agent document rewriting | "Make this better" for any document |
| ffmpeg | Audio/video processing orchestration | Trimming, converting, compressing media |
| docx | Word document creation and editing | Building .docx files programmatically |
| doc-coauthoring | Structured collaborative writing workflow | Writing docs, proposals, specs |
| deep-investigation | Evidence-based investigation protocol | Bugs, outages, complex technical questions |
| investigation-synthesis | Transforms findings into deliverables | 1-pagers, exec summaries, engineering briefs |
| reddit-research | Comprehensive Reddit research | Subreddit analysis, engagement metrics, topic discovery |
Identifies and removes 29 documented AI writing patterns based on Wikipedia's "Signs of AI writing" guide. Includes voice calibration (match a user's existing writing style) and a two-pass audit process. Works on any text.
Adversarial Reader-Empathy Synthesis Engine. Spawns isolated agent roles (Synthesizer, Critic, Editor, Judges) that cannot see each other's reasoning, eliminating sycophancy. Iterates until scores converge. Includes a deterministic lint pass for AI-tell removal.
Orchestrates ffmpeg/ffprobe for media processing. Covers trimming, format conversion, compression, resizing, merging, stream extraction, audio normalization, GIF creation, speed changes, filters, and batch processing. Always probes first, never overwrites inputs.
Full .docx manipulation using docx-js (creation) and pandoc (reading). Handles tables, images, headers/footers, TOC, multi-column layouts, tracked changes, and more. Includes critical gotchas for the docx-js library.
A three-stage workflow for collaborative document creation: Context Gathering (closes knowledge gaps), Refinement & Structure (section-by-section brainstorming/curation/drafting), and Reader Testing (validates with a fresh Claude instance). Works in Claude Code with sub-agents or in claude.ai with manual testing.
Structured investigation protocol using parallel sub-agents for evidence gathering. Enforces source citations on every claim, confidence classification (HIGH/MEDIUM/LOW), and produces structured reports with executive summaries, timelines, root cause analysis, and recommendations.
Takes raw investigation findings and transforms them through four layers (evidence, analysis, narrative, deliverable). Includes templates for 1-pagers and engineering briefs, an expert panel pattern (Skeptic/Operator/Executive), and the "reframe principle" for audience-appropriate delivery.
Routes through Arctic Shift API (sees deleted/removed posts), Reddit JSON API (live data), and Playwright (fallback). Covers engagement metrics, tone assessment, user activity, subreddit discovery, shadowban detection, and batch analysis. Read-only research only.
- Claude Code CLI or IDE extension
- For
ffmpegskill: ffmpeg/ffprobe installed - For
docxskill: pandoc (reading), Node.js (creation via docx-js) - For
reddit-research: WebFetch tool access, curl
PRs welcome. Each skill should:
- Have a
SKILL.mdwith proper frontmatter (name, description) - Be self-contained (no external dependencies beyond standard tools)
- Include examples where helpful
- Follow the existing format
MIT