Reusable agent skills for reducing unknowns in long-horizon coding work. The core idea is to keep the agent's map, its prompt, plan, and context, aligned with the territory: the real codebase, domain, runtime, data, constraints, and review expectations.
List the skills in this repository:
npx skills add MaybeMonad/skills --listInstall the full set:
npx skills add MaybeMonad/skills --allInstall a specific skill:
npx skills add MaybeMonad/skills --skill map-territory-workflowUse one skill without installing it:
npx skills use MaybeMonad/skills --skill map-territory-workflowInstallable skill directories live under skills/<category>/<skill>/, matching
the category-first layout used by repos such as mattpocock/skills.
catalog/skills.json is the machine-readable map, and suites/ keeps
human-facing notes for each category.
Folder map:
skills/unknown-reduction/: map-vs-territory skills for reducing unknowns before, during, and after implementation.skills/codex-orchestration/: Codex planning, delegation, verification, and reporting.skills/design-workflow/: source-first design workflow, visual direction, variations, prototypes, and polish gates.skills/web-reconstruction/: source-first website cloning and web-effect reconstruction.skills/search-growth/: search-facing product surfaces, canonical URL policy, indexing inventory, and submission evidence.
Suite metadata: suites/unknown-reduction/
| Skill | Purpose |
|---|---|
map-territory-workflow |
Choose the smallest unknown-reduction workflow across before, during, and after implementation. |
blindspot |
Run a pre-implementation blind spot pass to find unknown unknowns and cheap probes. |
brainstorm-prototype |
Explore multiple options or disposable prototypes before touching durable app code. |
interview-unknowns |
Ask high-leverage questions that can change architecture, data models, APIs, UX, or scope. |
reference-alignment |
Extract requirements from source code, screenshots, docs, libraries, or prior examples. |
implementation-plan-review |
Write implementation plans that foreground decisions likely to change. |
implementation-notes |
Log decisions, deviations, edge cases, and evidence during implementation. |
pitch-explainer |
Package work into a reviewer-ready explainer for buy-in and approval. |
change-quiz |
Explain a completed change and quiz the user before merge, release, or handoff. |
Suite metadata: suites/codex-orchestration/
| Skill | Purpose |
|---|---|
orchestrate-codex-workflow |
Plan, delegate, verify, and report complex Codex implementation workflows. |
Suite metadata: suites/design-workflow/
Credit: inspired by Trystan-SA/claude-design-system-prompt (MIT), adapted into this repository's source-first skills format.
| Skill | Purpose |
|---|---|
design-workflow |
Choose the smallest useful design path across source maps, direction, variations, prototypes, and polish. |
design-source-map |
Extract tokens, components, layout grammar, states, and constraints from real design sources. |
design-direction |
Commit to a concrete greenfield visual system before hi-fi work. |
design-variations |
Produce meaningful options for layout, hierarchy, visual treatment, flow, or interaction choices. |
design-prototype |
Build and verify clickable, stateful prototypes with real feedback and interaction states. |
design-polish |
Run a final design quality gate and fix accessibility, hierarchy, state, responsiveness, and trope issues. |
Suite metadata: suites/web-reconstruction/
| Skill | Purpose |
|---|---|
web-clone |
Clone, study, and remix websites from real source/runtime evidence, including static mirrors and WebGL/Canvas effect reconstruction. |
Suite metadata: suites/search-growth/
| Skill | Purpose |
|---|---|
technical-seo |
Run a canonical technical SEO workflow across metadata, routes, redirects, sitemap, robots, structured data, crawlability, and indexing evidence. |
skills.sh does not require a separate publish command. It discovers GitHub
repositories through the skills CLI flow and ranks skills from anonymous
installation telemetry.
This repo also has an automated GitHub release publisher in
.github/workflows/publish-skills.yml. Pull requests run discovery and
gh skill publish --dry-run; pushes to main validate the same way and then
publish non-interactively with gh skill publish --tag v0.0.<run_number>.
Manual workflow runs can override the tag.
The workflow uses GITHUB_TOKEN by default. If GitHub blocks release or
repository-topic writes for the default token, add a SKILLS_PUBLISH_TOKEN
secret with the repository permissions needed by gh skill publish.
Useful checks before sharing:
npx skills add MaybeMonad/skills --list
npx skills use MaybeMonad/skills --skill map-territory-workflow
npx skills use MaybeMonad/skills --skill web-clone
npx skills use MaybeMonad/skills --skill technical-seo
npx skills use MaybeMonad/skills --skill design-workflowDo not put installable SKILL.md files under suites/; default discovery finds
skills/<category>/<skill>/SKILL.md and treats suites/ as metadata only.
Search indexing on skills.sh may lag behind a successful install. If a valid
public repo still does not appear after the normal indexing delay, open a
listing/indexing issue in vercel-labs/skills with the repository URL and an
install command.