AI Workflows

assertcheck ships five Claude Code skills — structured workflows that guide your AI assistant through Negative Space Programming tasks from spec to audit.

Each skill enforces a strict protocol: it interviews before producing output, runs a defined set of passes or phases in order, and validates its own output before delivering. They are designed to work together as a complete development workflow.

---

The five skills

assertcheck-spec

Define what a feature must NEVER accept before a single line of implementation. Produces a precondition/postcondition table and an implementation checklist — every invariant traceable to one assertion.

View skill

assertcheck-feature

Build a new function, class, or service with the guard block first. Maps every input, prior state, and external response to an assertion — contract table before any code.

View skill

assertcheck-audit

Scan existing code for every unguarded boundary — nil dereferences, silent exits, absorbed errors. 6-category pass, risk-ranked report, ready-to-paste assertions.

View skill

assertcheck-refactor

Modify existing code without breaking contracts. Every change maps to a guard impact. Every removed safety check is replaced by an assertion — never silently deleted.

View skill

assertcheck-selector

Instant lookup: which assert.* call for which invariant? Always picks the most specific assertion. Used internally by all other skills.

View skill

---

How they fit together

The skills follow the natural lifecycle of a feature:

assertcheck-spec       → define what must NEVER happen
      ↓
assertcheck-feature    → build with the guard block first
      ↓
assertcheck-refactor   → modify without breaking contracts
      ↓
assertcheck-audit      → find the gaps that slipped through
      ↑
assertcheck-selector   → lookup at every step

Start with spec before touching a keyboard. Use feature to scaffold. Use refactor when extending. Use audit before any release. Selector is always available as a lookup when you're not sure which assertion to reach for.

---

How to trigger a skill

Each skill listens for natural-language triggers in your AI assistant:

You say	Skill invoked
"I'm building a new service for…"	assertcheck-feature
"Review this code for unguarded inputs"	assertcheck-audit
"Write a spec for the payment module"	assertcheck-spec
"I'm adding a parameter to this method"	assertcheck-refactor
"Which assert function should I use for…"	assertcheck-selector

Each skill begins with a mandatory interview step before producing any output — and runs a self-check on its own output before delivering. The quality of the contract map depends on understanding the feature's boundaries, not just its implementation.

---

What skills produce

Skill	Output
assertcheck-spec	Feature identity, preconditions table, postconditions table, state machine (if applicable), implementation checklist
assertcheck-feature	Contract table, guarded implementation (guard block + logic block), rejected-states summary
assertcheck-audit	Guard coverage score, risk-ranked findings table, proposed assertions, mindset note
assertcheck-refactor	Impact table, guard diff with [EXISTING] / [NEW] annotations, removed-safety ⚠ flags, caller note
assertcheck-selector	Most-specific assertion, msg / note phrasing, assert.* vs check() recommendation

---

Installation

Install all five skills into your current project:

npx skills add thonymg/assertcheck --skill='*'
bunx skills add thonymg/assertcheck --skill='*'
pnpx skills add thonymg/assertcheck --skill='*'

Install globally for use across all projects:

npx skills add thonymg/assertcheck --skill='*' -g

Install a single skill:

npx skills add thonymg/assertcheck --skill='assertcheck-audit'

Learn more at vercel-labs/skills.

---

Built for Claude Code

These skills are designed for the Claude Code CLI. They work in any Claude-powered AI assistant that supports the skills protocol.