Codifide Direttore Quickstart
The free harness is the installable part of Codifide Direttore: steering files, hook templates, and four enforcement scripts you add to a repository that an AI coding agent works in. This page condenses the setup from the repository README. The repository is the authority; if the two differ, follow the repository.
One naming note: the GitHub repository is still named agentic-stage-gate-governance and the harness keeps some earlier identifiers (.sgl/, SGL_* variables). A rename to match the product name is planned.
1. Get the steering files
Clone the repository, then copy the steering folder into your project’s AI context directory.
# Kiro
cp -r agentic-stage-gate-governance/steering /path/to/your/project/.kiro/steering
# Cursor
cp -r agentic-stage-gate-governance/steering /path/to/your/project/.cursor/rules
# Windsurf
cp -r agentic-stage-gate-governance/steering /path/to/your/project/.windsurf/rules
GitHub Copilot reads a single file, .github/copilot-instructions.md, so combine 00-welcome.md with the files you use most. For any other tool, paste 00-welcome.md into its custom instructions. The repository README has the full table. Hooks and on-demand loading depend on the host tool; the harness marks whether a host honors blocking hooks as unverified for some tools, so outside the hosts it documents, treat the rules as plain context.
One small core file (00-welcome.md, about 2,200 words) loads every session. The other files load on demand.
2. Install the enforcement scripts
Run this from the root of your project (not inside the harness clone). HARNESS is the path to the clone from step 1; the four scripts and their two config files go into a .sgl/ directory in your project.
HARNESS=../agentic-stage-gate-governance # adjust to where you cloned it
mkdir -p .sgl
cp "$HARNESS"/templates/enforcement/immutable-paths.example .sgl/immutable-paths
cp "$HARNESS"/templates/enforcement/allowed-licenses.example .sgl/allowed-licenses
cp "$HARNESS"/templates/enforcement/{guard-immutable-paths,check-immutable-paths,check-review-level,check-licenses}.sh .sgl/
chmod +x .sgl/*.sh
# Edit .sgl/immutable-paths and .sgl/allowed-licenses, then commit them on main.
# Then run the harness's own test suite once, from the harness clone:
(cd "$HARNESS" && bash templates/enforcement/test-enforcement.sh) # expect: RESULT ... ALL PASS
The immutable-path list is yours to write: it names the files an agent loop may read but never modify, such as verifiers and golden fixtures.
3. Register the guard and add CI
- Register the guard as a PreToolUse hook: see
templates/HOOKS-SESSION-STATE.md, Hook 4. It blocks agent file-tool writes to protected paths. - Of the four scripts, three run in CI:
check-immutable-paths.sh,check-review-level.sh, andcheck-licenses.sh. Run them from the base branch, not from the pull request, so a PR cannot edit the check that judges it. The GitHub Actions example intemplates/enforcement/README.mdwires up the first two;check-licenses.shreads an SBOM or licence report on standard input, so you add a step that produces one. - Pair the CI approval label with CODEOWNERS and branch protection so that only a human can approve changes to protected paths.
4. Start working
Open the project in your AI IDE. Say “Let’s start a new initiative” to begin, or “Assess this codebase against the gates” on an existing project.
What to expect
Only the four scripts are mechanical checks, and the guard depends on a host that honors blocking hooks, which the harness itself marks as unverified for some tools. Gate approvals, reviewer independence, and personas are written rules that agents follow. The full list is in what is mechanically enforced, and what agents follow. The README’s “Scope and limits” and “Validation status” sections say plainly what has not been done, including a controlled study of outcomes with and without the harness.
Questions or a larger rollout? See Work with Codifide.
Informational only. Not legal, medical, regulatory, security, or compliance advice. Provided as is, without warranty. See Notices.