This is Keel's public, release-only bundle for agent-assisted onboarding. It contains the two skills a coding agent needs to inspect an application, prepare a bounded Keel integration, and help a human author a friendly policy without taking authority from that human.
keel-setupinventories execution paths, prepares state-D integration, and verifies one intended allow and one intended deny throughPOST /v1/execute.keel-policytranslates human intent into Keel's canonical policy document and reports where the requested rule is unsupported or unenforceable.
Read shared/CONSTITUTION.md, then read the selected SKILL.md in full.
Use an immutable 40-character commit SHA. Verify SHA256SUMS before running a bundled helper. Do not
use curl | sh or reconstruct a missing release from memory.
The copied setup prompt and this repository contain no Keel credential or customer identifier. A coding agent must never ask a human to paste a credential into a conversation. When runtime verification is reached, the human creates a Runtime key in the Keel dashboard and enters it only at the release-pinned verifier's password-style prompt in a terminal they control. The key is never a command-line argument, never appears in shell history, and is not installed into the application by the verification step. The clipboard remains an exposure surface.
Only keel-setup/scripts/verify_execute.py performs network I/O. In hidden-input mode it verifies this
release before prompting, obtains a project-bound read-only verification profile, makes the two bounded
verification requests described by the setup skill, and confirms the profile did not change during the
proof. The remaining helpers are local, read-only analysis or schema validation.
SOURCE.json identifies this public bundle version and the files maintained only by the publication
layer. This repository intentionally excludes internal specifications, production workflows, private
repository history, and unrelated tools.
Apache-2.0. See LICENSE.