Agent Gate

A local stdio gate that checks an AI agent's consequential actions against a signed mandate and records a proof trail before writes proceed.

What you can do

• Gate agent actions against a signed mandate policy loaded from a local file, blocking writes that fall outside it.
• Record each gated action into a Forge proof trail stamped with a tenant id, failing closed or best-effort depending on record mode.
• Halt all gated writes instantly when a designated kill-switch file exists.
• Optionally route actions through a live Kalshi venue adapter using a local API key id and RSA private key.
• Initialize a mandate file through the package's own init command.

When to use it

Use it when an assistant or agent performs consequential, state-changing actions and you want a local policy check plus an auditable record before those actions go through. It suits teams that already use Forge for decision records or want a hard stop switch for agent writes.

Requirements

Runs locally over stdio via npx from the npm package forge-agent-gate, on desktop only. Needs a signed mandate JSON at an absolute path (created with the init command) and a Forge tenant id. Requires either FORGE_API_KEY or FORGE_BEARER_TOKEN for the Forge decision API, plus network access to it unless FORGE_API_BASE_URL points elsewhere. The optional Kalshi adapter needs KALSHI_API_KEY_ID, KALSHI_ENV and KALSHI_PRIVATE_KEY_PATH.