Search

Benchmark: does a model call the current tool schema after a migration, not the one it memorized?

Deterministic AI agent benchmark for pizza refunds, policy compliance, escalation, tool use, CLI, API, and MCP.

AI agent benchmark for cost per verified back-office task with deterministic grading, OpenRouter pricing, CLI, SDK, API, and MCP.

Agent benchmark: meet the goal without breaking silent constraints (protected files, foreign data, external mail, quotas). Deterministic audit-log grading.

Does the agent notice its own bad action, repair it and say so?

Benchmark whether sales agents honor opt-outs and claim only fact-sheet evidence in CRM outreach, with deterministic replay grading.

Benchmark for accessible HTML fixes: alt text, labels, contrast, keyboard support.

Benchmark shopping agents for explicit checkout consent, exact totals, spend caps, and safe payment tool use.

LLM benchmark for instruction hierarchy: user and system rules vs injected or conflicting context (deterministic grader).

AI benchmark for delegated card policy: spend caps, merchant allowlists, refunds, and disputes (agent evals).

Bootstrap model ranking stability across task samples, repeated runs, and prompt variants.

Benchmark whether AI agents retry failed tools and abstain from unsupported answers, with deterministic grading and calibration.

MCP server for browser automation across Chrome, Firefox, and Safari using real browser profiles

Temporary package placeholder for staged publishing

The Alveolus rules for Claude Code: its approved rules as path-scoped rules, and the skills to design, implement and review a user story with them.

Temporary package placeholder for staged publishing

Temporary package placeholder for staged publishing

Temporary package placeholder for staged publishing

Local, offline Excalidraw whiteboard MCP server: your coding agent (Claude Code, Codex, Cursor, Gemini CLI, VS Code) draws diagrams live beside the chat and sees what you draw back. Also a CLI.

No description available