Skip to content
Agentmart
Submit a tool

invariance-battery

Runtime assertion system that continuously verifies AI agent invariants — properties that must ALWAYS hold.

Skill by EvezArt

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/evezart-invariance-battery

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Runtime assertion system that continuously verifies AI agent invariants — properties that must ALWAYS hold. Use when building reliable autonomous agents, auditing agent behavior for drift, enforcing safety constraints, or building self-certifying AI systems. Covers invariant definition, runtime checking, drift detection, and falsification reporting.

Use cases: agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/invariance-battery

What's inside

Skill name
invariance-battery
Description
Runtime assertion system that continuously verifies AI agent invariants — properties that must ALWAYS hold. Use when building reliable autonomous agents, auditing agent behavior for drift, enforcing safety constraints, or building self-certifying AI systems. Covers invariant definition, runtime checking, drift detection, and falsification reporting.
License
MIT
Instructions
56 lines in SKILL.md · view on GitHub

Files (4)

  • origin.json · .clawhub/
  • SKILL.md
  • _meta.json
  • skill-card.md

SKILL.md (first lines)

# EVEZ Invariance Battery

Runtime assertion system for AI agents. Properties that must ALWAYS hold.

## When to Use

- Building autonomous agents that must maintain safety guarantees
- Auditing agent behavior for gradual drift from design intent
- Enforcing hard constraints that override any optimization objective
- Building self-certifying AI systems that can prove their own compliance
- Falsification testing — prove the agent VIOLATED an invariant, not just that it passed tests

## Architecture

The Invariance Battery runs as a continuous verification layer:

```
Agent Action → Invariance Check → PASS (proceed) / FAIL (halt + report)
                     ↓
              Append-Only Spine (audit trail)
```

## Invariant Types

1. **State Invariants** — Properties of the agent's internal state that must always hold
2. **Action Invariants** — Constraints on what actions the agent can take
3. **Temporal Invariants** — Properties over time (no oscillation, monotonic improvement)
4. **Boundary Invariants** — Hard limits the agent can never cross

## Key Concepts

- **Battery = Collection**: Multiple invariants checked in parallel, like electrical cells in series
- **Falsification > Verification**: A single violation PROVES the agent failed. No number of passes proves it works.
- **Drift Detection**: Gradual degradation of invariant scores over time
- **Spine Integration**: Every check is written to the append-only spine

## The Falsifier Gate

From EVEZ-OS: before any agent action is committed, it must pass through the falsifier gate.

```python
@invariant("action_cost < budget_threshold")
def check_budget(action):
    return action.estimated_cost < BUDGET_LIMIT

# If ANY invariant fails, the action is BLOCKED
# and the failure is written to the spine forever
```

## References

- EVEZ-OS: falsifier gate enforcement
- MAES: VERIFIED/PENDING/INVESTIGATING status model
- poly_c formula: τ × ω × topo / 2√N

From EvezArt/evez-os, MIT.

More from EvezArt/evez-os

Signals

Stars
1
Forks
0
Open issues
13
Last commit (this folder)
Jun 22, 2026 (4 months ago)
Last release
Jun 22, 2026 (4 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from EvezArt

Cite this page

invariance-battery skill. Agent Mart. https://agentmart.dev/skills/evezart-invariance-battery (last updated 2026-10-24)