Daves Claude Code Skills

Dave's Claude Code Skills

A curated collection of Claude Code skills and hooks for software architecture, knowledge management, and engineering. Drop .md files into .claude/skills/ for slash commands, or .py/.sh scripts into your hooks directory for automated guardrails.

42 skills across 8 categories. 17 skills use agent teams — parallel sub-agents launched via the Task tool that analyse different dimensions simultaneously, process batches at scale, or triage before deep-diving. 16 hooks across 6 categories provide automated security, quality, UX, and validation guardrails. Plus boxes-cli, a YAML-to-SVG/PNG diagram renderer, and a meta-agent for creating new skills.

---

Skills Overview

Category	Skills	Agent Teams	Focus
Architecture	8	5	Decisions, impact analysis, NFRs, cost optimisation
Content Processing	10	2	PDFs, PowerPoint, YouTube, web pages, books, CSV
Diagramming	3	1	C4 diagrams, system landscapes, diagram review
Vault Health	9	6	Quality reports, broken links, orphans, auto-tagging, diagnostics
Scoring	2	1	Document scoring, executive summaries
Reporting	2	2	Weekly summaries, project status reports
Meetings	3	1	Meeting notes, voice transcripts, email capture
Knowledge	5	0	Summarisation, related content, decisions, timelines
Total	42	17
Hooks	16	—	Security, quality, UX, safety, validation, notifications

---

Architecture (8 skills)

Skills for documenting decisions, analysing change impact, comparing scenarios, and managing non-functional requirements. Five of eight skills use agent teams for multi-dimensional parallel analysis.

Skill	Command	Agents	Description
ADR	/adr	—	Create Architecture Decision Records with structured context, rationale, and consequences
Impact Analysis	/impact-analysis	4	Analyse cascading impact across technical, organisational, financial, and risk dimensions
Scenario Compare	/scenario-compare	3	Compare 2-4 architectural scenarios with cost, timeline, complexity, and risk analysis
NFR Capture	/nfr-capture	—	Capture non-functional requirements with measurable acceptance criteria (ISO 25010)
NFR Review	/nfr-review	3	Review NFRs for completeness, measurability, and feasibility
Architecture Report	/architecture-report	5	Generate comprehensive reports for governance, audit, and stakeholder communication
Cost Analysis	/cost-analysis	3	Analyse infrastructure, licensing, and operational costs; identify savings opportunities
Dependency Graph	/dependency-graph	—	Visualise system dependencies with colour-coded criticality in Mermaid

Content Processing (10 skills)

Skills for extracting and structuring content from PDFs, presentations, videos, web pages, articles, books, and CSV data. Captures external knowledge into searchable Markdown notes.

Skill	Command	Agents	Description
PDF Extract	/pdf-extract	—	Extract structured content from PDFs with optional docling support for native table recognition
PPTX Extract	/pptx-extract	—	Convert PowerPoint slides to Markdown with docling/python-pptx dual extraction and Visual Mode
YouTube Analyze	/youtube-analyze	—	Analyse videos via transcripts with timestamped summaries and key takeaways
Video Digest	/video-digest	N	Batch-triage videos by relevance (Haiku), then deeply process the best (Sonnet)
Weblink	/weblink	—	Quick web page capture with AI-generated summary
Article	/article	—	Quick article capture with summary, key quotes, and relevance scoring
Book Notes	/book-notes	3	Create book notes with parallel extraction and optional knowledge compounding via spawned Concept/Pattern/Theme notes
Document Extract	/document-extract	—	Extract from any format (PDF, DOCX, HTML, CSV) with auto-detection
CSV to Page	/csv-to-page	—	Convert CSV data to structured Markdown notes with frontmatter
De-AI-ify	/de-ai-ify	—	Remove AI-generated phrasing and rewrite in natural, human voice

Diagramming (3 skills)

Skills for generating and reviewing architecture diagrams. Built on graph drawing research (Purchase et al.) and real-world C4 modelling experience. See the blog post for the theory.

Skill	Command	Agents	Description
Diagram	/diagram	—	Generate architecture diagrams in multiple formats (C4, system landscape, data flow, AWS)
C4 Diagram	/c4-diagram	—	Specialised C4 generation: Mermaid C4, flowchart LR with C4 styling, or PlantUML
Diagram Review	/diagram-review	4	Analyse existing diagrams for readability and architecture quality

Vault Health (9 skills)

Skills for measuring and improving the quality of a Markdown knowledge vault. Six use agent teams — fan-out for multi-dimensional analysis, batch for high-throughput processing. Three additional utilities handle renaming, diagnostics, and URL validation.

Skill	Command	Agents	Description
Quality Report	/quality-report	5	Comprehensive quality metrics with Flesch readability formulas, link density scoring, and type-aware freshness thresholds
Broken Links	/broken-links	3	Find broken wiki-links, heading anchors, and missing attachment references
Orphan Finder	/orphan-finder	4	Detect disconnected notes and suggest meaningful connections
Auto-Tag	/auto-tag	N	Batch auto-tag notes using type-based rules and customisable keyword-to-tag mapping tables
Auto-Summary	/auto-summary	N	Batch-generate one-line summary fields with type-specific patterns and quality validation rules
Link Checker	/link-checker	N	Validate external URLs with curl-based checking, frontmatter status tracking, and cross-reference verification
Rename	/rename	—	Safely rename notes with automatic backlink updates across the vault
Startup Diagnostic	/startup-diagnostic	—	Run vault health checks on session start — stale notes, broken links, orphans
Check Weblinks	/check-weblinks	—	Validate external URLs in frontmatter and note body with status reporting

Scoring (2 skills)

Skills for evaluating documents against rubrics and generating stakeholder-ready summaries.

Skill	Command	Agents	Description
Score Document	/score-document	4	Score documents against customisable rubrics with optional SQLite persistence for querying and multi-scorer comparison
Exec Summary	/exec-summary	—	Generate executive summaries tailored to CEO, CTO, board, or PM audiences

Reporting (2 skills)

Skills for generating periodic status reports by analysing vault content across multiple dimensions in parallel.

Skill	Command	Agents	Description
Weekly Summary	/weekly-summary	5	Generate weekly activity reports from daily notes, tasks, meetings, and projects
Project Report	/project-report	4	Generate RAG project status reports with tasks, risks, and timeline assessment

Meetings (3 skills)

Skills for capturing and structuring meeting content from transcripts, voice recordings, and emails.

Skill	Command	Agents	Description
Meeting Notes	/meeting-notes	3	Create structured meeting notes with decision and action item extraction
Voice Meeting	/voice-meeting	—	Process voice transcripts with speech-to-text correction into structured notes
Email Capture	/email-capture	—	Capture important emails as structured vault notes with action items

Knowledge (5 skills)

Skills for discovering, connecting, and visualising knowledge across a vault.

Skill	Command	Agents	Description
Summarize	/summarize	—	Summarise notes with configurable depth (one-liner, paragraph, page) and audience
Find Related	/find-related	—	Discover related content via tag overlap, backlinks, keywords, and temporal proximity
Find Decisions	/find-decisions	—	Extract and catalogue formal and informal decisions across a date range
Timeline	/timeline	—	Generate visual timelines (Mermaid Gantt, table, or list) from vault events
Skill Creator	/skill-creator	—	Generate new Claude Code skill files with agent team boilerplate

---

How Agent Teams Work

17 of the 37 skills use agent teams — parallel sub-agents launched via Claude Code's Task tool. Each agent runs independently in its own context window, analyses one dimension of the problem, and returns structured results. The coordinator synthesises everything into a unified output.

flowchart TD
    User["/skill invoked"] --> Coord1["COORDINATOR\nGathers input, prepares agent prompts"]

    Coord1 --> A1["Agent 1 · Sonnet\nDimension A\ne.g. Technical"]
    Coord1 --> A2["Agent 2 · Sonnet\nDimension B\ne.g. Cost"]
    Coord1 --> AN["Agent N · Sonnet\nDimension N\ne.g. Risk"]

    A1 --> Coord2["COORDINATOR\nSynthesises all agent results\ninto unified output"]
    A2 --> Coord2
    AN --> Coord2

Three Orchestration Patterns

Pattern	How It Works	Example Skill	Typical Speedup
Fan-Out/Fan-In	3-5 agents analyse different dimensions of the same input in parallel; coordinator synthesises	/impact-analysis (technical, org, financial, risk)	3-4×
Batch Processing	Same operation on many items, divided into parallel batches of 15-20	/auto-tag (N Haiku agents tag notes in parallel)	4-9×
Triage + Selective	Fast agents score everything; only high-scoring items get deep processing	/video-digest (Haiku triages, Sonnet deep-analyses)	2-3× time, 60-80% cost saving

For the full explanation with worked examples, anti-patterns, and model selection guidance, see the Agent Teams Guide.

---

Examples

Worked examples with prompts, outputs, and explanations of what makes each diagram good (or bad).

Diagramming Examples

Example	What It Shows
Declaration Order	Why element declaration order is the single most important factor for diagram readability. Side-by-side comparison of the same 9 elements in random vs. structured order.
C4 Context Diagram	How to create a clean Level 1 C4 diagram showing actors, system boundary, and external dependencies.
C4 Container Diagram	How to create a Level 2 C4 diagram with system boundaries, technology labels, and database shapes. Includes PlantUML alternative.
C4 Component Diagram	How to diagram internal components (controllers, services, repositories) at C4 Level 3.
Real-World Example	Complete Context + Container diagrams for a real AI-powered incident management platform.

Example Screenshots

All examples include rendered diagram screenshots:

Bad declaration order	Good declaration order	C4 Context
C4 Container	C4 Component	Real-world Context

---

Installation

All Skills

mkdir -p .claude/skills
cp skills/**/*.md .claude/skills/

By Category

# Architecture — ADRs, impact analysis, NFRs, cost analysis
cp skills/architecture/*.md .claude/skills/

# Content Processing — PDF, PPTX, YouTube, web, books
cp skills/content-processing/*.md .claude/skills/

# Diagramming — C4 diagrams, system landscapes, diagram review
cp skills/diagramming/*.md .claude/skills/

# Vault Health — quality reports, broken links, orphans, auto-tag
cp skills/vault-health/*.md .claude/skills/

# Scoring — document scoring, executive summaries
cp skills/scoring/*.md .claude/skills/

# Reporting — weekly summaries, project status reports
cp skills/reporting/*.md .claude/skills/

# Meetings — meeting notes, voice transcripts, email capture
cp skills/meetings/*.md .claude/skills/

# Knowledge — summarise, find related, decisions, timelines
cp skills/knowledge/*.md .claude/skills/

Individual Skills

# Pick only the skills you need
cp skills/architecture/adr.md .claude/skills/
cp skills/vault-health/quality-report.md .claude/skills/
cp skills/content-processing/youtube-analyze.md .claude/skills/

Verify Installation

After copying, invoke any skill in Claude Code:

/adr Use Event-Driven Integration for Orders
/impact-analysis Migrate database from Oracle to PostgreSQL
/quality-report --type ADR
/weekly-summary
/youtube-analyze https://www.youtube.com/watch?v=example

---

Documentation

Document	Description
Agent Teams Guide	How agent teams work: patterns, model selection, best practices, anti-patterns, worked example
Skills Reference	Quick-reference card: all 42 skills by category, installation commands, model cost guide
Blog Post	Why Your AI-Generated Diagrams Look Terrible — the graph drawing research behind the diagramming skills
Hooks Guide	Complete hooks documentation: lifecycle events, patterns, configuration, installation
boxes-cli README	YAML-to-SVG/PNG diagram renderer — format reference and usage

---

Key Concepts

Diagramming

Built on three insights from graph drawing research:

• Declaration order controls layout. Dagre (Mermaid) and Sugiyama (PlantUML) position elements based on where they appear in the source. Declare elements in reading order — actors left, data stores right.
• Edge crossings are the strongest predictor of comprehension difficulty. Research by Helen Purchase showed reducing crossings improves reader accuracy by 30-40%. Always set an explicit crossing target.
• Gestalt proximity overrides colour. Elements placed close together are perceived as related, regardless of styling. Use subgraphs and boundaries to group related components.

Agent Teams

Built on three orchestration principles:

• Fan-Out/Fan-In for multi-dimensional analysis. When evaluating something from multiple angles (technical, financial, risk), run agents in parallel and synthesise. Used by 13 skills.
• Batch Processing for scale. When the same operation applies to many items, divide into batches and process in parallel with cost-effective Haiku agents. Used by 3 skills.
• Triage before deep processing. When only some items are worth deep analysis, use fast agents to score first, then invest in expensive processing only for high-value items. Used by 1 skill.

---

Hooks (16 hooks)

Production-tested Claude Code hooks that run automatically during your workflow — blocking secrets, validating content, formatting code, and providing contextual hints. No manual invocation required. Includes 4 standalone validators for CI/CD or pre-commit integration.

How Hooks Work

User submits prompt
  |
  v
[UserPromptSubmit hooks]        -- secret-detection.py scans for API keys
  |                              -- context-loader.sh injects skill context
  v
Claude generates tool call
  |
  v
[PreToolUse hooks]              -- file-protection.py blocks .env edits
  |                              -- secret-file-scanner.py blocks secret writes
  v                              -- search-hint.sh suggests faster tools
Tool executes (if not blocked)   -- bash-safety.py auto-allows safe commands
  |
  v
[PostToolUse hooks]             -- frontmatter-validator.py checks YAML
  |                              -- tag-taxonomy-enforcer.py validates tags
  v                              -- wiki-link-checker.py finds broken links
Response shown to user           -- filename-convention-checker.py checks names
  |                              -- code-formatter.py auto-formats code
  v
[Stop hook]                     -- desktop-notify.sh sends notification

Exit codes: 0 = allow | 1 = warn (show message, continue) | 2 = block (prevent tool execution)

Hook Categories

Category	Hooks	Event	Description
Security	3	PreToolUse / UserPromptSubmit	Block secrets in prompts and file content, protect sensitive files
Quality	4	PostToolUse	Validate frontmatter, enforce tag taxonomy, check wiki-links, verify filenames
UX	3	PostToolUse / PreToolUse / UserPromptSubmit	Auto-format code, load context for skills, suggest faster search tools
Safety	1	PermissionRequest	Auto-allow safe bash commands to reduce permission prompts
Notification	1	Notification (Stop)	Desktop notifications when long tasks complete (macOS + Linux)
Validators	4	Standalone	Reusable validation scripts for frontmatter, links, file contents, and new file checks

Individual Hooks

Hook	File	Event	What It Does
Secret Detection	secret-detection.py	UserPromptSubmit	Scans prompts for API keys, tokens, passwords, and connection strings (25 regex patterns). Blocks before Claude sees the secret.
Secret File Scanner	secret-file-scanner.py	PreToolUse (Edit|Write)	Scans file content being written for embedded secrets. Catches hardcoded credentials that slip into code.
File Protection	file-protection.py	PreToolUse (Edit|Write)	Blocks edits to .env, .key, credentials.json, CI/CD configs, and lockfiles. Configurable allow-list for safe directories.
Frontmatter Validator	frontmatter-validator.py	PostToolUse (Edit|Write)	Validates YAML frontmatter against configurable note type schemas. Checks required fields, valid values, and date formats.
Tag Taxonomy Enforcer	tag-taxonomy-enforcer.py	PostToolUse (Edit|Write)	Enforces hierarchical tag naming (area/engineering not engineering). Configurable taxonomy with approved flat tags.
Wiki-Link Checker	wiki-link-checker.py	PostToolUse (Edit|Write)	Warns about broken [[wiki-links]] by scanning the vault for matching files. Supports aliases and type prefixes.
Filename Convention	filename-convention-checker.py	PostToolUse (Edit|Write)	Validates filenames match note type conventions (e.g. Meeting - 2026-01-15 Title.md). Checks prefix, casing, and location.
Code Formatter	code-formatter.py	PostToolUse (Edit|Write)	Auto-formats files after edit using the right tool: Prettier (JS/TS/CSS), Black (Python), gofmt, rustfmt, or shfmt.
Context Loader	context-loader.sh	UserPromptSubmit	Detects skill commands (e.g. /meeting, /adr) and auto-loads relevant .claude/context/ files so Claude has domain knowledge.
Search Hint	search-hint.sh	PreToolUse (Grep)	When Claude uses Grep for simple keyword searches, suggests faster alternatives like SQLite FTS or dedicated search indexes.
Bash Safety	bash-safety.py	PermissionRequest (Bash)	Auto-allows safe read-only commands (ls, git status, npm test, cat) so you don't get prompted for every harmless command.
Desktop Notify	desktop-notify.sh	Notification (Stop)	Sends macOS/Linux desktop notification with sound when Claude finishes a long task. Never miss a completed response again.
Validate File Contains	validate_file_contains.py	Standalone	Validate that files contain required content patterns (e.g. frontmatter, headers). Usable as a PreToolUse or PostToolUse hook.
Validate Frontmatter	validate_frontmatter.py	Standalone	Validate YAML frontmatter structure against configurable schemas per note type. More thorough than the PostToolUse version.
Validate Links	validate_links.py	Standalone	Validate internal wiki-links and external URLs within Markdown files. Checks for broken references and dead links.
Validate New File	validate_new_file.py	Standalone	Validate newly created files meet naming conventions, have required frontmatter, and are in the correct directory.

Quick Start — Copy-Paste Configuration

1. Copy hooks to your project:

cp -r hooks/ your-project/

2. Add to .claude/settings.json (or settings.local.json):

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": ".*",
        "hooks": [
          {"type": "command", "command": "python3 hooks/security/secret-detection.py", "timeout": 10},
          {"type": "command", "command": "hooks/ux/context-loader.sh", "timeout": 5}
        ]
      }
    ],
    "PreToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {"type": "command", "command": "python3 hooks/security/file-protection.py", "timeout": 5},
          {"type": "command", "command": "python3 hooks/security/secret-file-scanner.py", "timeout": 10}
        ]
      },
      {
        "matcher": "Grep",
        "hooks": [
          {"type": "command", "command": "hooks/ux/search-hint.sh", "timeout": 5}
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {"type": "command", "command": "python3 hooks/quality/frontmatter-validator.py", "timeout": 10},
          {"type": "command", "command": "python3 hooks/quality/tag-taxonomy-enforcer.py", "timeout": 10},
          {"type": "command", "command": "python3 hooks/quality/wiki-link-checker.py", "timeout": 15},
          {"type": "command", "command": "python3 hooks/quality/filename-convention-checker.py", "timeout": 10},
          {"type": "command", "command": "python3 hooks/ux/code-formatter.py", "timeout": 10}
        ]
      }
    ],
    "PermissionRequest": [
      {
        "matcher": "Bash",
        "hooks": [
          {"type": "command", "command": "python3 hooks/safety/bash-safety.py", "timeout": 5}
        ]
      }
    ],
    "Notification": [
      {
        "matcher": "Stop",
        "hooks": [
          {"type": "command", "command": "hooks/notification/desktop-notify.sh"}
        ]
      }
    ]
  }
}

3. Pick only what you need:

# Security only (recommended minimum)
cp hooks/security/ your-project/hooks/ -r

# Security + quality (Obsidian vaults)
cp hooks/security/ hooks/quality/ your-project/hooks/ -r

# Everything
cp hooks/ your-project/ -r

Customisation

All hooks include # Customise: comments marking configurable sections:

# Customise: Add your protected file patterns
PROTECTED_PATTERNS = [r"\.env$", r".*\.key$", r"credentials\.json$"]

# Customise: Add directories that are always safe to edit
ALLOWED_DIRECTORIES = ["docs/", "tests/"]

Hook	What to Customise
file-protection.py	PROTECTED_PATTERNS, ALLOWED_DIRECTORIES
frontmatter-validator.py	NOTE_SCHEMAS — add your note types and required fields
tag-taxonomy-enforcer.py	TAG_HIERARCHIES — define your tag categories and values
filename-convention-checker.py	CONVENTIONS — define filename patterns per note type
secret-file-scanner.py	SECRET_PATTERNS, SKIP_PATTERNS
context-loader.sh	Skill-to-context-file mapping
bash-safety.py	SAFE_COMMANDS — commands to auto-allow

For detailed documentation: Hooks Guide | Lifecycle | Patterns | Configuration | Installation

---

Tools

boxes-cli

A YAML-to-SVG/PNG diagram renderer using ELK.js for hierarchical layout. Define system diagrams in a simple YAML notation and render to SVG or PNG with automatic layout.

cd tools/boxes-cli
npm install
node bin/boxes.mjs render input.yaml -o output.svg
node bin/boxes.mjs render input.yaml -o output.png --format png

Dependencies: elkjs, js-yaml, sharp, commander

See tools/boxes-cli/README.md for full documentation and YAML format reference.

---

Agents

Meta-Agent

A specialised Claude agent definition for creating new skills and agents from documentation. Lives in agents/meta-agent.md — copy to .claude/agents/ to use.

The meta-agent analyses existing documentation, code, and patterns to generate well-structured skill files with proper frontmatter, agent team configuration, and output format definitions.

---

Contributing

Contributions welcome. To add a new skill:

1. Create a .md file in the appropriate skills/<category>/ directory
2. Include YAML frontmatter with description and model fields
3. Follow the skill template: When to Use, Usage, Instructions, Output Format, Examples
4. For agent team skills, define each agent with name, model, task, steps, and return format
5. Use the /skill-creator skill to generate boilerplate
6. Add a worked example in examples/<category>/
7. Update this README

See the Agent Teams Guide for patterns and best practices when building agent team skills.

---

Licence

MIT