A collection of AI skills for research, strategy, writing, and communication. Each skill is a structured prompt file that instructs any AI assistant how to perform a specific task to a high standard. Skills are format-compatible with the Agent Skills open standard.
Skills
| Skill | Description |
|---|---|
| adr | Creates and maintains Architecture Decision Records; captures context, decision, consequences, and alternatives for significant architectural choices |
| backlog-manager | Command-driven outcome-focused backlog in a single file; add, refine, and track work items by observable result |
| citation-discipline | Binds every factual claim to a verifiable source at the point of assertion; enforces citation placement, source quality, and precision |
| code-review | Systematic multi-dimensional code review covering correctness, security, performance, and maintainability; produces prioritised, actionable findings |
| feedback | Structured, evidence-grounded critique of written work, arguments, decisions, or plans; findings are specific, prioritised, and paired with concrete recommendations |
| peer-reviewer | Checks a completed research item for logical coherence, alternative explanations, and cross-item integration; audit-only, never edits |
| plain-language | Rewrites complex or technical text so a non-expert reader can understand it without losing accuracy or completeness |
| research | Rigorous, evidence-driven research using recursive decomposition and verification loops |
| research-question | Validates and scopes a research question before investigation begins; checks specificity, answerability, scope, motivation, and decomposability |
| research-reviewer | Audits a completed research item for citation discipline, speculation control, writing quality, and logical coherence; audit-only, never edits |
| speculation-control | Enforces strict separation between evidence-based statements and speculative, interpretive, or subjective content |
| strategy-author | High-rigour strategy documents grounded in Rumelt and Porter frameworks |
| remove-ai-slop | Eliminates AI detection signals from text — statistical, structural, and alignment artifacts |
| strategic-persuasion | Cognitive rhetoric and audience-targeted persuasive content construction |
| technical-writer | Clear, accurate technical documentation for developers, operators, or end users; covers READMEs, API references, guides, runbooks, and architecture docs |
| swe | Software engineering grounded in SOLID principles, Fielding's REST constraints, Gang of Four design patterns, and Enterprise Integration Patterns; emphasises planning, design, and iterative improvement |
| tdd | Test-driven development using Red-Green-Refactor; enforces test-first discipline, testing pyramid balance, and systematic test design principles for any feature, fix, or behaviour change |
Using a skill
Each skill is a plain markdown file. There are two ways to use them.
Any AI assistant (generic)
Copy the contents of a SKILL.md into your conversation as a system prompt or
prepend it to your message:
[paste contents of research/SKILL.md]
Now research: What are the second-order effects of LLM proliferation on knowledge work?
Or reference it via file attachment, context window upload, or however your tool accepts external context.
Tools with native SKILL.md support
Some AI coding tools discover and load skills automatically. Copy the skill directory into your tool's configured skills location:
| Tool | Install path |
|---|---|
| Claude Code | ~/.claude/skills/<name>/ or .claude/skills/<name>/ |
| OpenAI Codex CLI | ~/.codex/skills/<name>/ or .codex/skills/<name>/ |
Example:
cp -r research ~/.claude/skills/researchOnce installed, invoke by name or let the tool trigger the skill automatically
based on the description field in the frontmatter.
Repository structure
Skills/
├── README.md # This file
├── PRD.md # Project requirements and goals
├── CHANGELOG.md # What changed and when
├── PROGRESS.md # Session-by-session build journal
├── decisions/ # Architecture Decision Records
│ └── 0001-adopt-skill-md-standard.md
├── adr/
│ └── SKILL.md
├── backlog-manager/
│ └── SKILL.md
├── citation-discipline/
│ └── SKILL.md
├── code-review/
│ └── SKILL.md
├── feedback/
│ └── SKILL.md
├── peer-reviewer/
│ └── SKILL.md
├── plain-language/
│ └── SKILL.md
├── remove-ai-slop/
│ └── SKILL.md
├── research/
│ └── SKILL.md
├── research-question/
│ └── SKILL.md
├── research-reviewer/
│ └── SKILL.md
├── speculation-control/
│ └── SKILL.md
├── strategic-persuasion/
│ └── SKILL.md
├── strategy-author/
│ └── SKILL.md
├── technical-writer/
│ └── SKILL.md
├── swe/
│ └── SKILL.md
└── tdd/
└── SKILL.md
Creating a new skill
- Create a directory:
mkdir my-skill - Create
SKILL.mdwith YAML frontmatter and a markdown body:
--- name: my-skill description: What this skill does and when to use it. Tools that support automatic skill selection use this field to decide when to load the skill. --- # Skill instructions here...
- Test by pasting the content into any AI assistant with a relevant prompt.
See the Agent Skills open standard for the full specification.
Skill quality checklist
Use this to evaluate a skill before adding it or after making changes.
Frontmatter
-
namematches the directory name -
versionis present -
descriptionis a single, precise sentence naming the task and when to trigger the skill — vague descriptions cause spurious auto-invocation in tools that support automatic skill selection
Sections (all must be present)
-
## When Not to Use— excludes clearly inappropriate contexts; at least two concrete exclusions -
## Interaction Protocol— lists clarifying questions and specifies output style -
## Inputs and Outputs— states input type, output type, and composability with other skills in this repo
Instruction quality
- Instructions are written as imperatives, not suggestions ("List all items" not "You should list all items")
- No vague verbs: "handle", "manage", "deal with", "consider" each replaced with a precise action
- Failure modes or stopping conditions are defined
- No AI slop patterns: no formulaic transitions ("Furthermore", "In conclusion"), no symmetrical filler paragraphs, no safety-prefacing language
- No implementation-specific references: no CLI commands, repo paths, file paths, or tool-specific syntax — skills must work with any agent and any repo
Verification
- Paste into an AI assistant with a realistic prompt and confirm the output matches the skill's stated intent
Related resources
Broader collections of skills and prompts that may contain further inspiration:
| Resource | Description |
|---|---|
| agentskills.io | The Agent Skills open standard this repo follows |
| anthropics/skills | Official skills from Anthropic: document processing, design, development, and communication |
| ComposioHQ/awesome-claude-skills | Curated list of community-built skills with categories for dev tools, data analysis, business, and more |
| travisvn/awesome-claude-skills | Curated index of official and community skills with installation instructions |
| BehiSecc/awesome-claude-skills | Large community skill list searchable by domain; dev tools, automation, data, security, and project management |
| obra/superpowers | Battle-tested skills library for Claude Code including TDD, debugging, brainstorm, and collaboration patterns |
Contributing
Add new skills to the root of the repo as a named directory containing SKILL.md. Follow the format in "Creating a new skill" above. Update this README's index table and repository structure tree. Add a CHANGELOG entry.
Skills must be implementation-agnostic. Do not reference specific CLI commands, file paths, repo names, or tool-specific syntax in a SKILL.md. This repo is consumed as a git submodule by other repositories. A skill that references one repo's CLI will break silently when used in any other context. If you find yourself writing a path like Research/in-progress/ or a command like python -m src.main in a SKILL.md, stop — that content belongs in the consuming repo's prompt files, not in the skill.