Pyshackle: A hard pre-execution gate for AI agent tool calls (open source)

Pyshackle is listed as an open source package on PyPI at https://pypi.org/project/pyshackle/. A Hacker News item under the AI Agents track (https://news.ycombinator.com/item?id=49182815) presented it as a hard pre-execution gate for AI agent tool calls; at capture it had 2 points and 0 comments.

The product mechanic in the title is the control: a hard gate that runs before an agent’s tool call executes, not after. That places the check on the path from “agent decides to call a tool” to “tool actually runs,” so blocked or rejected calls never reach side-effecting code. Distribution via PyPI targets Python stacks that already host agent loops and tool adapters.

For engineers shipping agents that can shell out, hit internal APIs, or mutate state, pre-execution enforcement is a concrete safety boundary. Logs and post-mortems do not undo a bad write, a destructive command, or an unintended external call; a hard gate is meant to stop those before they fire. Builders who treat tool use as a privilege boundary can treat this as a policy choke point rather than optional middleware.

The listing sits in a crowded agent-tooling market where guardrails, sandboxes, and policy layers compete for the same “agent wants to act” moment. Open source on PyPI is a low-friction entry for teams already on Python agent frameworks, without a commercial SDK. Early HN signal is weak—2 points, no comments—so public review and adoption evidence are still thin.

Practical next step is to open the PyPI project page, confirm install and API surface, and map whether a hard pre-execution gate can sit on your existing tool-call path without inventing a second control plane. Watch the HN thread for technical pushback and whether maintainers publish integration patterns for common agent runtimes.