A growing collection of AI agent skills for software development workflows. Built for technical leads, TPMs, and developers who want Claude Code to help with planning, specification, and engineering tasks.
Contributions welcome! Found a way to improve a skill or have a new one to add? Open a PR.
Transform vision documents into traceable, testable specifications. See the TPM Planning Guide for the full methodology.
| Skill | Description | Triggers |
|---|---|---|
| tpm-spec-trace-ids | Annotate vision docs with Feature IDs (F-nnn) | "annotate my vision doc," "add feature IDs," "create coverage index" |
| tpm-roadmap-slice | Extract features into Phase PRDs with requirements | "create phase PRD," "plan next quarter," "extract phase from vision" |
| tpm-spec-verify | Add QA requirements and acceptance criteria | "add QA plan," "add acceptance criteria," "enrich with test criteria" |
- TPM Planning Guide — Full methodology for the planning pipeline
Planning Pipeline:
flowchart TD
A[Vision PRD] --> B[tpm-spec-trace-ids]
B --> C[Annotated PRD + Coverage Index]
C --> D[tpm-roadmap-slice]
D --> E[Phase I PRD with Requirements]
E --> F[tpm-spec-verify]
F --> G[Phase I PRD with Acceptance criteria]
G -->|Iterate| D
style B fill:#4a9eff,color:#fff
style D fill:#4a9eff,color:#fff
style F fill:#4a9eff,color:#fff
Go from a messy product idea to a structured go-to-market plan, then develop brand creative direction — all driven by structured workflows with phase gates and opinionated defaults.
| Skill | Description | Use When |
|---|---|---|
| idea-to-icp | Take a messy business idea and structure it into GTM primitives: Product, ICP segments, value prop, beachhead market, and channel strategy | "decompose my idea", "help me find my ICP", "who should I sell to", "idea to ICP", "structure my go-to-market" |
| creative-direction | Develop 3-5 distinct brand creative treatments from GTM data — tone, visual mood, messaging, and taglines | "creative direction", "brand treatments", "brand tone", "creative brief", "brand identity direction" |
GTM Pipeline:
flowchart TD
A[Messy Product Idea] --> B[idea-to-icp]
B --> C[GTM Decomposition: Product + ICP + Value Prop + Channels]
C --> D[creative-direction]
D --> E[Brand Creative Direction: Tone + Visual Mood + Messaging]
style B fill:#4a9eff,color:#fff
style D fill:#4a9eff,color:#fff
Track and optimize autonomous Claude Code session performance with metrics, trend analysis, and improvement tracking.
| Skill | Description | Use When |
|---|---|---|
| self-improvement | Analyze session efficiency, track improvements, view trends | "check loop metrics", "how are sessions doing", "analyze iterations", "loop performance", "session efficiency" |
Prerequisites: This skill is built for projects using an autonomous loop setup:
- ralph-wiggum loop — A bash script that runs Claude Code headlessly in batches, saving each session transcript as
claude-iteration-N.jsonl(via--output-format stream-json). Handles rate-limit backoff, zombie task cleanup, and performance feedback injection between iterations. - Beads (
bd) — A git-native CLI issue tracker (.beads/issues.jsonl). The parser detectsbd update,bd-finish.sh, andbd closecommands to determine task attribution and commit success.
Without these, the improvement tracker (improvement add/list/fix/search) still works standalone.
Key Features:
- Dashboard with recent sessions, trends, and target comparison
- Deep analysis with automatic trend comparison to previous runs
- Improvement tracking (add, fix, search, list) with severity levels
- Parses
claude-iteration-N.jsonlsession transcripts produced by the ralph-wiggum loop - Efficiency targets: completion rate, narration-only turns, parallel tool calls, turns per session
- SQLite-backed persistence with auto-created schema
Build accessible, production-ready autocomplete, token inputs, and filter query builders using proven patterns from Downshift, Headless UI, and tools like Datadog and Linear.
| Skill | Description | Use When |
|---|---|---|
| webdev-combobox-autocomplete | Foundational autocomplete/combobox patterns with ARIA, keyboard nav, async suggestions | Building autocomplete, command palettes, search inputs, select replacements |
| webdev-token-input | Multi-value token/chip inputs with key:value parsing | Building filter bars, tag inputs, email "To" fields, multi-select chips |
| webdev-filter-query-builder | Advanced filter query construction with AST, operators, serialization | Building observability tools, data analytics, search interfaces with boolean logic |
Key Features:
- State model patterns (highlightedIndex, virtual focus, token management)
- ARIA combobox pattern with
aria-activedescendant - Keyboard navigation (arrows, Enter, Escape, Tab, Backspace)
- Prop-getter pattern for framework-agnostic implementation
- Async suggestions with race condition prevention
- Focus management solutions (blur vs click-outside, cursor jumping)
- Context-dependent suggestions and caching
- Filter AST representation and query serialization
Go from a script or story to a prepared script ready for nano-banana or another image generator.
/plugin marketplace add https://github.com/ozten/skills
/plugin install comic-strip-pipeline@skillsDuring a claude code session:
/comic-strip-pipeline:create-comic Please take `outputs/2026-01-29_07-05-38_building-in-public-is.script.md` and identify the best comic strip lurking in their. Characters
to choose from are `character_sheets/character_sheets.md` and ideally the setting is `character_sheets/setting.md`, but choose another setting if needed.Use add-skill to install skills directly:
# Install all skills
npx skills add https://github.com/ozten/skills
# Install specific skills
npx skills add https://github.com/ozten/skills --skill tpm-spec-trace-ids
# List available skills
npx skills add https://github.com/ozten/skills --listClone the repo and copy the skills you need:
git clone https://github.com/ozten/skills
cp -r skills/* ~/.claude/skills/Download individual SKILL.md files and place them in ~/.claude/skills/skill-name/.
- Fork this repository
- Customize skills for your team's workflows
- Clone your fork into your projects
Once installed, just ask Claude Code to help with tasks:
Planning tasks:
"Annotate this vision doc with feature IDs"
→ Uses tpm-spec-trace-ids skill
"Create a phase PRD for Q2"
→ Uses tpm-roadmap-slice skill
"Add acceptance criteria to this spec"
→ Uses tpm-spec-verify skill
Web development tasks:
"Build an autocomplete search input with keyboard navigation"
→ Uses webdev-combobox-autocomplete skill
"Create a tag input with token chips like Linear's filters"
→ Uses webdev-token-input skill
"Build a filter query builder with operators and date ranges"
→ Uses webdev-filter-query-builder skill
GTM & strategy tasks:
"I have an idea for a property management tool for small landlords"
→ Uses idea-to-icp skill
"Decompose my idea into ICP segments and channels"
→ Uses idea-to-icp skill
"Create brand creative direction from my GTM doc"
→ Uses creative-direction skill
"Give me brand treatments for my product"
→ Uses creative-direction skill
Loop analytics tasks:
"Check my loop metrics"
→ Uses self-improvement skill
"Analyze the last 20 iterations"
→ Uses self-improvement skill
"Add an improvement for high narration rate"
→ Uses self-improvement skill
Or reference skills directly when starting a task:
"Using tpm-roadmap-slice, extract features F-012 through F-018 into a phase PRD"
"Using webdev-combobox-autocomplete, build a command palette with async suggestions"
Found a way to improve a skill? Have a new skill to suggest? PRs and issues welcome!
Ideas for contributions:
- Improve existing skill instructions or workflows
- Add new skills for other domains (DevOps, security, testing)
- Fix typos or clarify confusing sections
- Add examples or templates
How to contribute:
- Fork the repo
- Create or edit skill files
- Submit a PR with a clear description
Skills are markdown files that give AI agents specialized knowledge and workflows for specific tasks. When you add these to your project, Claude Code can recognize what you're working on and apply the right frameworks and best practices.
Each skill is a directory containing a SKILL.md file:
skills/
skill-name/
SKILL.md
assets/ # Optional templates
references/ # Optional detailed docs
See Claude Skills for details.
[Full instructions for the AI agent]
## License
MIT — Use these however you want.