Blog
Repo readiness, execution governance, and AI agent notes.
Browse product notes, engineering notes, field notes, and release essays about contract-first onboarding, CI alignment, and agent-safe repository operations.
Pressure-testing Ota on Dagger: generated SDK lineage and bounded engine truth
A focused Dagger pressure slice models generated TypeScript SDK lineage and replay authority, verifies release-asset fulfillment and admission, and records where execution stops at Dagger's managed-engine boundary.
A Protected Runner Is Not an Approval System
Four VPS runs show Ota binding exact-scope authority to one governed repository action while expired, revoked, and out-of-scope grants refuse before work.
Ota v1.6.27 Now Available: Effect-Bound Refusal Assurance
Ota v1.6.27 closes V12 with typed effect declarations, pre-execution refusal, private evidence reconciliation, and bounded Linux/macOS pressure.
Pressure-testing Ota on EventCatalog: generated artifact lineage across sibling consumers
A generated Langium artifact is useful only when its real sibling consumer builds through the complete declared closure. EventCatalog pressure-tested that boundary with released Ota across three operating systems.
Pressure-testing Ota on cloud.devenv.sh with Nix and devenv
A hosted Linux and macOS matrix tested how Ota models Nix as launcher and devenv as orchestrator, while keeping execution and mutation claims bounded.
Ota v1.6.26 Now Available: Protected Execution Foundations and Receipt History
Ota v1.6.26 closes the bounded V11.7 OSS audited-crossing slice through protected systemd-launcher execution, immutable receipt history, recovery evidence, and refined service-listener admission.
Pressure-testing Ota on Flagr: native and container Go verification with honest boundaries
Flagr pressured released Ota 1.6.25 across native and container Go verification, managed Compose lifecycle proof, refusal canaries, and contract-to-CI drift without turning local evidence into a deployment claim.
Test What Your AI Agents Must Not Do
AI agent guardrails need negative tests. Ota refusal canaries verify that the real execution boundary still rejects selected unsafe tasks and workflows before any work starts.