Operational Memory Gates
August 26, 2026 ยท View on GitHub
A reference implementation and operational specification of 21 commit-time pre-code gates and a daily model roster audit.
This is the literal reference implementation running in production at NiraNexus. It is shared as a specification for mechanical governance, not as a turnkey CLI tool. It is hardcoded to our paths, registries, and stack on purpose โ the value is the specification, not the shell.
Read the specification: SPEC.md
What is here
| File | Purpose |
|---|---|
SPEC.md | The provenance-locked specification: which gates are fires made mechanical, which are disciplines made mechanical. |
pre-code-gate.sh | The 21-check gate. 20 refuse (non-zero exit), 1 advises (pre-completion pipeline). |
model-audit.py | Daily roster audit against the live OpenRouter directory. |
validate-audit.js | Zero-dependency production-readiness validator. |
run-pipeline.js | Zero-dependency pre-completion pipeline runner. |
The thesis
A pre-code gate is a production lesson compiled into a mechanical check. It does not review. It does not recommend. It refuses.
Every gate is one of two things: a fire made mechanical, or a discipline made mechanical. The distinction is stated per gate in the spec.