Claims
Every claim comes with a receipt. Each claim below is judged at main@db3c2343 from the checks that ran on that commit. A claim whose check is red, or did not run there, is struck out until it passes again. Claims we do not make are listed too.
0 hold · 7 retracted · 0 not checked · 3 designed, not built · 2 not claimed
Developer preview. Not production-ready, not audited, not for sensitive code.
| Claim | Proof you can rerun | Verdict |
|---|---|---|
| It tells you it is not ready for production: its readiness check answers harnessReady false | ci not-run · tests/public/cli.test.mjs | retracted |
| Calling harness with no subcommand is refused with a typed error, no_harness_subcommand, that lists every available verb | ci not-run · tests/public/cli.test.mjs | retracted |
| A run executes once: reopening it never starts a second execution | ci not-run · tests/prevention/run-executes-once.test.ts | retracted |
| Commands the model starts cannot read outside the project (Linux, Landlock) | ci not-run · tests/prevention/subprocess-confinement-owner.test.ts | retracted |
| Secrets the harness detects are redacted before the prompt reaches the provider | ci not-run · tests/prevention/redaction-reaches-provider.test.ts | retracted |
| Every change passes a secret scan before it lands in the public tree | secret not-run · scripts/check-secrets.mjs | retracted |
| CodeQL analyses every change | codeql not-run · .github/workflows/codeql.yml | retracted |
| A malicious module cannot bypass the core | decision UD-CORE-NO-BYPASS: pending | designed, not built |
| The model never sees a credential | decision CRED-DIR: pending | designed, not built |
| Layered memory with provenance and forgetting | decision MEM-DIR: pending | designed, not built |
| An independent security audit | None has been done: no third-party review, penetration test or certification exists. Evidence | not claimed |
| Signed releases | Releases carry checksums and a promotion receipt, but no signature yet. | not claimed |