Capture an architecture decision once. edikt compiles it into a directive the model follows automatically — every session, every engineer, without you saying it twice.
$ curl -fsSL https://github.com/diktahq/edikt/releases/download/v0.8.0/install.sh | bash$ brew install diktahq/tap/ediktClaude Code today — Codex and other harnesses next.
Watch it happen
The model has memory. It doesn't have governance.
Every session starts from zero — and a decision your team made yesterday gets quietly contradicted today.
Write an architecture decision as plain prose. edikt compiles it into a directive the model reads automatically, before it writes a line of code — no re-explaining, no pinned Slack message, no wiki nobody opens.
/edikt:gov:compile → reads accepted decisions, checks for contradictions between them, renders enforcement. Update one, recompile. One source of truth, not a doc that drifts from the code.
Critical findings block the agent automatically — before the code lands, not after review. If someone overrides a gate, that's logged with their identity, so trade-offs are visible, not silent.
Security, DBA, and architecture review run inline, on every change. No silent failures.
Ask "what's our status?" and get the governance dashboard. Say "save this decision" and it's captured, compiled, and enforced from the next session on. No command syntax to memorize.
No slash commands to learn — describe what you want in plain English. Same vocabulary you already use.
Every file is markdown or YAML, committed to your repo like any other source file. No daemon, no build step, no vendor lock-in — stop using edikt and the files stay exactly as readable as they were.
git diff shows exactly what changed, same as any other commit. Nothing hidden, nothing binary.
One engineer installs edikt and commits it. Every teammate's session inherits the same standards from that point on — code review stops catching formatting and starts catching design decisions, whether it's your best engineer's session or the one who joined this week.
The rules live as markdown and YAML in the repo, committed like any other source file — the model reads them automatically for every teammate, at the start of every session, no separate setup step.
Working solo? Same mechanism, just you and the model staying in sync from one session to the next — no re-explaining your own architecture back to yourself. Running a consultancy? Install once, and every client engagement inherits the same standards from day one.