Sheaf
Seven questions about how your agents run today. It takes two minutes and needs no account. You get a score, where the weaknesses are, and what Sheaf would do about each one.
Already running agents through Sheaf? Paste your key and the report is built from what they actually did.
The AI Agent Safety Check · your answers
Seven questions, one per dimension, two minutes. Each answer scores that dimension until Sheaf can measure it from what your agents do. Saved to the account; edit any time.
An MCP config (Claude Desktop, Cursor, Claude Code), a tool schema, a Sheaf proxy config, or the Python file that defines the agent. Four of the seven answers can be read from it, each with the evidence. Credentials are stripped in your browser before anything is sent, and nothing is stored.
Context
Which tools can your agents reach? (optional)
Composite
Measuring…
Footprint · what Sheaf sees on this account
Do next · weakest first