Skillz Forge
Reusable capabilities from the OverKill Hill P³™ forge.
Skillz turns recurring AI methods into portable, inspectable, versioned SKILL.md contracts that agents can load, compose, and improve. Discover, inspect, install, and contribute, entirely through the browser and GitHub.
The prompt is not the artifact. The durable method is.
Every recurring AI task deserves a governed method. Most get a prompt buried in chat history instead.
Disposable prompts in chat history
A prompt that works today is not findable tomorrow. It lives in a chat window: no version, no review, no handoff. Every collaborator re-invents it from scratch.
Prompt protocols that define conditions
A governed prompt protocol specifies its role, context, inputs, boundaries, and failure modes. It is writable, reviewable, and testable, but it still lives in a document, not a repository.
Reusable skills that package methods
A SKILL.md packages a governed method as a delegation contract for an AI agent. The agent loads the file, activates on trigger, and follows the specified method, without re-prompting for the instructions every time.
GitHub repositories that preserve the method
GitHub stores the source, history, issues, discussions, and pull requests. The method becomes a durable artifact: versioned, forkable, improvable, and attributable. That is the gap Skillz closes.
What is a SKILL.md?
A SKILL.md is a plain-text delegation contract for an AI agent. Drop it into your agent's skill directory (Claude Code, Cursor, GitHub Copilot, VS Code, Replit Agent) and the agent gains the defined capability without requiring a re-prompt for every task.
Activation triggers
Specific phrases or task types that cause the agent to load and execute this skill rather than improvising a response.
Role and context
Who the agent is acting as, what environment it is operating in, and what governing authority the skill carries.
Inputs and method
What information the skill expects, what steps it follows, and what decisions it is authorized to make without asking.
Boundaries and failure handling
What the skill must not do, what it does when inputs are missing or ambiguous, and how it signals a handoff back to the operator.
Outputs and validation
What the skill produces, what a correct output looks like, and how the result can be verified against the contract.
Companion skills
Related skills in the same family: upstream skills that feed this one, or downstream skills that consume its output.
A discovery workbench, not a marketplace
- ◆A public Agent Skills discovery workbench
- ◆A generated catalog of repository skill content
- ◆A way to inspect and install
SKILL.mdfiles - ◆A way to compose curated or local skill stacks
- ◆A GitHub collaboration bridge: issues, PRs, discussions
- ◆Open source, browser-only, no login, free always
- •Not a replacement for GitHub
- •Not an authenticated social network
- •Not a hosted agent runtime
- •Not a guarantee that every skill is production-ready
- •Not a marketplace requiring payment or account creation
- •Not a second Prompt Forge
- •Not a claim that all skills are complete or validated
How a prompt becomes a portable capability
The same method moves through four stages: design, packaging, discovery, and collaboration. Each stage has a specific surface in the OverKill Hill ecosystem.
The same capability asset (one SKILL.md file) is designed in Prompt Forge, packaged in Found‑Rᵧ, discovered via Skillz Forge, and maintained in GitHub. No duplication of surfaces. Each stage does one job.
148 public distribution skills across 16 active families
Source-derived count from the live generated catalog (project-summary.json) as of 2026-08-20. This page fetches that file at load time and updates the numbers above automatically; if the fetch fails (e.g. offline), the last-synced values shown here remain. A prior July 29, 2026 audit reported 75 skills across 12 families; the count grew as new packages were added and audited.
Forge currently lists 119 draftable, 24 skeleton, and 5 usable contracts; no contract carries validated or published maturity. Maturity describes the contract's completeness; it is tracked separately from evidence status, which describes what proof exists for the current version (live, analytical, local-checks, historical, not-run, or none). See how maturity and evidence work below.
This page does not maintain a live family directory. The SPA is the authoritative exploration surface: family breakdowns, maturity filters, and per-skill detail are all available there. Explore the full catalog in Skillz Forge ↗ · Read the July 29, 2026 maturity audit ↗
Verified application capabilities
These are the currently shipped behaviors of the live Skillz Forge application. Future capabilities are listed separately in the roadmap and labeled as planned.
Search by task or phrase
Phrase search across the generated catalog: find skills by the task they perform, the triggers they activate on, or the problem they address.
Browse skill families
Skills are organized into functional families; browse by category to find related capabilities and understand how skills compose into workflows.
Filter by maturity
Each skill carries a maturity level: from placeholder through published. Filter to see only validated skills, or explore skills under active development.
Inspect individual skills
Open any skill to read its full contract: triggers, role, method, boundaries, outputs, and failure handling, before deciding to install it.
Copy a raw installation URL
Each skill detail page surfaces the raw GitHub URL for its SKILL.md file, ready to paste into a curl command or reference in your agent's configuration.
Save favorites locally
Mark skills you use frequently. Favorites are stored in browser local storage, with no account needed and no server involved.
Explore curated stacks
Pre-composed skill combinations for common workflows: browse curated stacks to understand how skills work together before assembling your own.
Start with the work
Answer guided discovery prompts to turn a goal into a focused route through the catalog without guessing the right skill name first.
Compose a local stack
Select and order skills in the browser, then export a personal stack as an installation manifest or curl script.
Open native GitHub paths
Every skill links directly to its GitHub source, issue tracker, and discussion thread. Contribute, report, or improve, directly in the repository.
View static generated activity context
Build-time activity signals from the repository: recent commit context and family provenance embedded into the generated catalog without live GitHub API polling.
Skills in action: four recurring scenarios
Each pathway shows a recurring problem, the relevant skill family in the repository, the resulting artifact, and where to find it. These are real repository examples, not invented scenarios.
Governed Mermaid Workflow
The problem: AI-generated Mermaid diagrams drift in style, break in target renderers, and require repeated follow-up prompts to fix. The governance rules are re-stated from scratch every session.
The skill family: Mermaid: palette definitions, renderer profiles, prompt scaffold generation, and diagram-family capability rules.
The artifact: A SKILL.md that an agent loads once. Every subsequent diagram generation inherits the brand palette, renderer target, and constraint rules, without re-prompting.
Process Documentation Pipeline
The problem: Documenting a business process requires consistent vocabulary, BPMN-aligned structure, and reproducible output across team members and AI tools, none of which a one-off prompt delivers.
The skill family: Process Capture: BPMN for Mermaid structural patterns, swim-lane conventions, gateway semantics, and output format rules.
The artifact: A composable SKILL.md that maintains structural consistency across every diagram in the pipeline, from first draft to version-controlled deliverable.
New Skill Creation Pipeline
The problem: Writing a well-formed SKILL.md from scratch requires remembering the full contract schema (triggers, inputs, method, boundaries, failure handling, outputs, and validation) every time.
The skill family: Agent Foundry: schema validation, field completeness checking, trigger description quality, and companion skill mapping.
The artifact: A SKILL.md for creating SKILL.md files. The agent follows the authoring contract and produces a well-formed, schema-valid skill file on the first pass.
AI Conversation Capture and Repository Reconciliation
The problem: Valuable insight from an AI conversation (a working method, a reusable protocol, a verified approach) lives in a chat window and disappears. Reconciling it against existing repository artifacts is manual and error-prone.
The skill family: Context Extraction: thread routing, deduplication against existing skills, classification as new, complementary, or duplicate, and repository integration paths.
The artifact: A governed capture workflow that converts a chat export into a reconciled repository contribution, preserving the method, not just the conversation.
Browse Context Extraction family in Skillz Forge ↗How skill maturity and evidence work
Maturity and evidence answer two different questions, and the catalog tracks them separately rather than merging them into one badge. Maturity describes the state of the contract itself: is the trigger, method, and boundary language complete? Evidence status describes what proof exists that the current version actually works: a live run, an analytical design, local checks, a historical benchmark from an older version, an unexecuted evaluation design, or no indexed record at all. A skill can be draftable with strong local-checks evidence, or draftable with no evidence record; the maturity label alone does not tell you which.
Maturity is set in the repository by the skill author or a maintainer. It is not automatically inferred. If a skill's maturity does not match your observations, open an issue ↗.
Evidence status, separately
draftable does not mean validated, and usable does not mean production-safe in every environment. validated is reserved for a fresh, version-matched benchmark with a measurable result and a protected holdout; no skill in the current catalog has reached it yet. An evidence note may be historical (a benchmark from an older version of the same skill), not-run (a written evaluation design that has not been executed), or none (no indexed evidence record, not a claim that the skill is unsafe).
Per-skill evidence state and evidence notes are shown on each skill's detail page and in compare view in the live Forge; this page states the model, not every skill's individual record.
How to contribute, improve, or report
Skillz Forge surfaces GitHub paths; it does not imitate or replace GitHub's collaboration interface. All contribution happens natively in the repository.
Open an issue for a bug
Skill behaves differently than its contract states? The trigger fires incorrectly? File an issue ↗
Suggest a new skill
Describe the recurring task, the gap the skill would fill, and what a correct output looks like. Suggest via Issues ↗
Improve trigger descriptions
Better trigger language improves match rates across agents and coding assistants. Open a pull request with the improved wording. Open a PR ↗
Use Discussions for design questions
Scope, contract design, and family organization questions belong in Discussions, not issues. Open a Discussion ↗
Report missing metadata
Maturity level wrong? Description missing? Family tag incorrect? These are low-barrier contributions with high catalog impact.
Security advisories
For sensitive findings (skills that produce harmful outputs, expose credentials, or violate stated boundaries), use GitHub's private security advisory channel.
How Skillz Forge is built
The Skillz Forge application is a static, client-side React SPA. There is no server, no database, and no authenticated runtime. The catalog is generated from repository content at build time.
React + TypeScript
Component architecture with full type safety. No framework dependencies beyond React and Vite.
Vite
Fast build toolchain. Static output deployed to GitHub Pages without a server or CDN dependency.
Fuse.js phrase search
Client-side fuzzy search across the generated catalog. No search backend, no query API. All search runs in the browser.
Static generated catalog
Repository skill content is parsed at build time and emitted as a static JSON catalog. The application reads this catalog; it does not poll the GitHub API at runtime.
GitHub Pages
The live application is hosted at okhp3.github.io/skillz/ via GitHub Pages. No custom server. Hash-based routing where applicable.
GitHub as source of truth
The repository is the canonical artifact store. The application is a discovery and installation layer on top of it, not a separate data system.
Where the build is going
-
✓
Public Skillz Forge Application Shipped
Live at okhp3.github.io/skillz/. Generated repository catalog. Phrase search and guided discovery. Family and maturity browsing. Skill detail pages. Curated stacks. Local stack composition and export. FAQ. GitHub deep links.
-
▶
Brand Alignment and Catalog Hardening Active
OverKill Hill brand alignment. Catalog provenance. Stable share URLs. Structured metadata across all skill entries. Analytics consistency. Activity freshness signals.
-
✓
Compare Mode Shipped
Side-by-side skill comparison: review trigger language, method steps, and boundary conditions across two to four skills before selecting one for a workflow.
-
✓
Custom Local Stack Composition Shipped
Build a personal skill stack in the browser: select skills, order them, and export the stack as a single installation manifest or curl script.
-
✓
Live GitHub Activity Shipped
Build-time snapshot of recent commits for individual skills, without requiring authentication or live API polling from every user session.
-
○
Issue and Pull-Request Context Panels Planned
Embedded summary of open issues and recent pull requests for a skill, visible in the detail view, linking to the native GitHub thread rather than reimplementing the interface.
-
○
Authenticated Collaboration Planned
Secure authenticated paths: submit issues, rate skills, and contribute from within the application. Not implemented; no write access to the repository exists in the current application.
User Guide: Installing a Skill
Get from browsing to an installed skill in under three minutes.
-
01
Open Skillz Forge
Navigate to okhp3.github.io/skillz/ in your browser, or use the embedded tool above. No login required.
-
02
Find a skill by search or family
Use the search bar to find a skill by task description, or browse by skill family if you know the category you need.
-
03
Inspect the skill contract
Open the skill detail page and read the full contract before installing. Check the maturity level, activation triggers, and stated boundaries.
-
04
Copy the raw installation URL
The detail page surfaces the raw GitHub URL for the
SKILL.mdfile. Copy it. -
05
Install in your agent
Drop the raw URL into a
curlcommand to save the file, or reference it directly in your agent's skill configuration. For example:# Example: replace with the actual raw URL from Skillz Forge curl -o .agents/skills/my-skill.md \ https://raw.githubusercontent.com/OKHP3/skillz/main/skills/family/SKILL.md
-
06
Verify activation
Test the skill by using one of its documented trigger phrases with your agent. If it does not activate or produces incorrect output, file an issue ↗.
FAQ
Is Skillz Forge free?
Yes. Free now, free always. No subscription, no freemium gate, no account required. MIT licensed. The application and all skills in the public catalog are open source.
Do I need a GitHub account to use it?
No. You can browse, search, and copy installation URLs without any account. A GitHub account is only needed if you want to contribute: open issues, submit pull requests, or join discussions.
Which agents support SKILL.md files?
Any agent that accepts file-based context: Claude Code, Cursor, GitHub Copilot, VS Code with agent extensions, Replit Agent, and similar tools. The skill file is plain text; the agent reads it as context before responding to a trigger.
Are all skills production-ready?
No. Skills have maturity levels from placeholder through published, and a separate evidence status (live, analytical, local-checks, historical, not-run, or none) that describes what proof exists for the current version. Draftable does not mean validated, and usable does not mean production-safe in every environment. Validated requires a fresh, version-matched benchmark. Check both the maturity label and the evidence note before using a skill in a production workflow.
What is the difference between Skillz Forge and Prompt Forge?
Prompt Forge is the design workshop, where prompts, protocols, and contracts are written and hardened. Skillz Forge is the discovery and installation layer, where finished skills are browsed, installed, and improved. They are adjacent stages in the same pipeline, not competing surfaces.
What is Found‑Rᵧ and how does it relate?
Found‑Rᵧ is the OverKill Hill meta-framework for recursive AI agent architecture: the packaging stage that turns a governed prompt protocol into a formal repository artifact. A skill goes through Found‑Rᵧ on its way to appearing in the Skillz catalog. See the Found‑Rᵧ page for details.
Can I contribute my own skills to the catalog?
Yes. Open a pull request against the OKHP3/skillz repository ↗. Include a well-formed SKILL.md following the schema, a proposed family assignment, and an initial maturity level. The contribution guide is in the repository.
Does the application send my data anywhere?
Search and browsing run locally in the browser against the static generated catalog, and raw search text is never transmitted. The app does make a few outbound network requests: it loads the catalog JSON from GitHub Pages (public, read-only content), it loads its typeface files from Google Fonts, and it sends aggregate usage events to Google Analytics 4. Those events include page views and bucketed interaction counts, never raw search text or personal information. See the Privacy page ↗ for the full disclosure.
How do I report a skill that produces harmful or incorrect output?
For public, non-sensitive issues: open a GitHub issue with the skill name and a description of the incorrect behavior. For sensitive findings: use GitHub's private security advisory feature on the OKHP3/skillz repository.
The reason Skillz exists is straightforward: I kept losing the methods. A prompt would work. A protocol would get refined. A workflow would become reliable. And then it would disappear into chat history or a private document, invisible to the next project and unreachable by anyone else.
The question became: what would it look like if every reusable AI method was a first-class repository artifact? Version-controlled, inspectable, improvable, attributable. Not buried in a prompt library or locked behind a subscription.
That is Skillz. A public capability library where the method is the artifact and GitHub is the preservation layer.
Not a marketplace. Not a SaaS. A forge product: designed, maintained, and improved in the open.
OverKill Hill P³
Ready to start?
Browse the catalog, find a skill, install it in your agent.