Skip to main content
Web Summit Lisbon, 9 to 12 Nov. Meet the founder in Lisbon.Book time

What's new in Slipmark.

Every release, newest first. The full technical changelog is in the repository.

v0.5.0

Verdicts are bound to the agent's instructions as well as the files it read, and a pull request can no longer loosen the rules it is judged by.

Security

  • The policy is read from the branch you merge into, not from the pull request. A pull request that drops a required file or loosens the policy is still judged by the rules on the base branch.
  • Every report records which policy made the decision, with its SHA-256 hash.

New

  • Agent context. Verdicts record the hashes of AGENTS.md, CLAUDE.md, skills, hooks, Cursor rules and MCP configs. A verdict given under instructions that later changed stops counting.
  • A change to an agent's instructions needs an approver's verdict before it can merge. Pair it with CODEOWNERS, or require signed approvals.
  • Changed agent instructions must be plain Markdown that GitHub shows in full. Text hidden from reviewers (invisible characters, HTML comments, blank padding, images) is refused, and in GitHub Actions the file is checked against GitHub's own renderer.
  • Agent instructions that link to files outside the watched set are refused.
  • Diff coverage: every file changed since the base needs a current, passing verdict.
  • Signed verdicts through Sigstore, tied to the workflow that gave them.
  • Verdicts can record the spec they were checked against, the checks Slipmark ran, and the model that gave them.
  • A policy file, .slipmark/policy.json, for coverage, allowed models, signatures, required files and agent context.

Thanks to Kent Bodrov, for reporting that a pull request could loosen its own policy.

v0.4.0

A Claude Code plugin, Slipmark Cloud uploads, and Slipmark gating its own code.

Security

  • Files are resolved through symlinks and re-checked against the real repository root. A link inside the repository pointing outside it is no longer followed.

New

  • Claude Code plugin: a skill that records verdicts, /slipmark:verify, /slipmark:setup, and a check that asks Claude to re-attest when a file it signed off has changed.
  • Runs can be recorded in Slipmark Cloud from GitHub Actions.
  • Slipmark now gates its own code with a real review verdict, required by its CI.

Fixed

  • File names that start with two dots, such as ..notes.md, are no longer rejected.

v0.3.0

One command to set up a repository, and instructions for coding agents.

New

  • npx slipmark init adds the attestation folder, a report-only GitHub workflow and agent instructions in one step.
  • Agent instructions in AGENTS.md, and in CLAUDE.md, Cursor rules or Copilot instructions where the repository already has them.
  • Setup notes for Claude Code, Cursor, Copilot and Codex.
  • Published to npm as slipmark.

Changed

  • A repository with no verdicts yet now explains how to record the first one.

v0.2.0

DriftGuard is renamed Slipmark.

Changed

  • New name, command, attestation folder (.slipmark/attestations) and schema. A slip mark is the paint line engineers draw across a tightened bolt: if the bolt moves, the line breaks.

v0.1.1

A report-only mode for rolling the gate out gently.

New

  • The Action reports passed or refused, and can warn without failing the build while you roll it out.

v0.1.0

The first release of the verdict attestation gate.

New

  • Agents record each verdict with the SHA-256 of every file they read. The gate re-hashes them before anything ships and refuses on any change.
  • A GitHub Action, and a worked example with one deliberately stale verdict.
  • Drift diagnosis for AI features: behavioural locks, git bisection and repair prompts for coding agents.

Changed

  • Licence changed from MIT to Apache-2.0 for its explicit patent grant.