Integrations

Keel works with any agent that can read a file or run a shell command. The SKILL.md provides instructions-mode safety with zero dependencies. The CLI adds cryptographic integrity for agents that can execute commands.

Surface compatibility

Surface	SKILL.md	CLI	Approval pause	Status
OpenClaw	Yes	Yes	Strong	LIVE
Claude Code	N/A	Yes	Strong	Pre-release
Codex	N/A	Yes	Varies	Pre-release
Gemini CLI	N/A	Yes	Varies	Pre-release
Custom agents	Yes	Yes	Depends	Available

Strong = surface can pause before tool execution and present a typed approval prompt. Varies = depends on client configuration. Keel always enforces policies and logs actions regardless of approval support.

Setup guides

OpenClaw

Install the SKILL.md as a skill. The agent loads it on every context refresh, so safety rules survive compaction. If the CLI is installed, the agent uses it automatically for cryptographic hashing and deterministic policy checks.

# When published:
clawhub install threshold-keel

# Manual install (available now):
# Copy SKILL.md to your OpenClaw skills directory

Claude Code / Codex / Gemini CLI

These surfaces can run shell commands. Install the CLI and the agent calls keel check-policy and keel wal-append directly. Configure the agent's system prompt or instructions to reference Keel.

# When published:
pip install threshold-keel
keel init
keel --human status

Custom agents

If your agent can read files, it can use the SKILL.md in instructions mode. If it can execute shell commands, it can use the CLI. If it speaks HTTP, it can call the Keel Cloud API directly.

# CLI (local or cloud)
keel check-policy --action-json '{"action_type":"send_email","target_ids":["[email protected]"],"surface":"email","reversibility":"irreversible"}'
# Exit 0 = allowed, exit 1 = blocked, exit 2 = requires approval

# Cloud API (direct)
curl -H "Authorization: Bearer $KEEL_CLOUD_API_KEY" \
     https://api.thresholdsignalworks.com/api/v1/policies/check \
     -d '{"action_type":"send_email","target_ids":["[email protected]"]}'

MCP server integration is on the roadmap. This will allow Keel to sit directly at the MCP transport boundary, intercepting tool calls before execution without requiring shell access. For now, the CLI and SKILL.md provide equivalent safety guarantees.