Every tool call your coding agents make, decided and signed.
Coding agents run their own shell, file edits and web fetches, outside any MCP server. One command plugs Decision Horizon into each agent's own hook system, so every one of those calls gets a policy decision, an Ostiary score and a signed record before it runs.
The gap it closes
Claude Code, Cursor, Copilot and the rest can read your secrets, run any shell command and edit any file, using tools that are built in. A proxy in front of MCP never sees them. The agents' hook systems do, and that's where Decision Horizon now sits.
Capabilities
One command per agent
dh hooks install claude-code writes the agent's own hook configuration. Your existing hooks are kept, and uninstall removes only ours.
Built-in tools and MCP
Shell commands, file reads, writes and edits, web fetches and MCP tool calls all go through the same policy and record.
Answers in the agent's language
Allow, deny or ask, in each agent's format. Claude Code and Copilot show their own approval prompt for a paused call.
Can't be switched off from inside
The agent can't delete the records, read the signing key or edit its own hook configuration. Those calls are denied and recorded.
Secrets followed by value
A value read from .env and then sent by curl is caught by its content, whatever the command looks like.
Claude Agent SDK too
The same engine as in-process hooks for the Python and TypeScript Agent SDKs, with identical records.
pip install decision-horizon
dh hooks install claude-code # .claude/settings.json
dh hooks install cursor copilot --user
dh hooks list # Claude Code, Cursor, Copilot, Codex, Gemini CLI,
# Windsurf, Cline, Kiro, Factory, Continue, Augment,
# OpenCode, AmpWhat it doesn't do
- A hook runs beside the agent, from configuration its user controls: whoever controls the machine can switch it off, and calls made while it is off aren't recorded. Every record names its insertion point.
- Arguments and results are as the agent reported them, and a shell command's targets are the paths it spells out. Codex runs hooks only for its shell, apply_patch and MCP tools.
See it on one of your own workflows.
We'll put Decision Horizon in front of a real agent workflow of yours and replay the record with you.