discipline · tamper alarm · enforcement
Instructions can be rationalized past. So a deterministic diff check catches tampering — and CI on a machine no agent controls makes that check unskippable. rigor-suite installs all three layers with one command, pinned to known-good versions, each to its own home. Never mixed.
== layer 1: dev-rigor-stack-lite (v0.7.0) == Installed 20 hook-free skills == layer 3: deterministic-detector (0fb0c4b) == == layer 2: tampercheck (0.1.1 from PyPI) == rigor-suite install report: layer 1 discipline dev-rigor-stack-lite PASS layer 2 tamper alarm tampercheck PASS layer 3 enforcement deterministic-detector PASS next: new agent session; then /install-detector inside each target repo.
the design
These are not three features of one product. They are three independent projects, each useful alone, arranged so that the whole set fails safe.
Hook-free, evidence-first skills that tell the agent how to work: lanes, gates, receipts, honest verification reports. Portable across Claude Code, Codex, and Agent Skills hosts.
A deterministic diff check that catches the cheat layer 1 can't self-report: deleted, skipped, and focused tests, forced-success expressions, hollowed assertions, swallowed errors. No LLM, offline, provider-agnostic.
Wires the deterministic checks into places no agent controls:
randomized-order test runs, diff-scoped mutation reports, and a pinned tampercheck
lane in CI — installed per repo with /install-detector.
quick start
git clone https://github.com/scottconverse/rigor-suite
.\rigor-suite\install.ps1 # -Force to replace
git clone https://github.com/scottconverse/rigor-suite
./rigor-suite/install.sh # --force to replace
/install-detector, run inside each target repo