RAGBAZ / grove / Sylvae
The same evidence log, met through the CLI, a loopback page, or an agent handoff — each with its own authority.
published · Updated 2026-10-05
01 Sylvae
sylvae runs filters the log by skill, backend and status; sylvae show prints one run's metadata, input, output and error, refusing unknown or ambiguous identifiers. JSON output stays byte-clean under pipes, NO_COLOR or TERM=dumb, so scripts and humans read the same bytes.
Shell completion derives from the live command list. Nothing here needs a server, a token or a browser.
02 Sylvae
sylvae review serves the same records at loopback, with filtering, full-run expansion and a form that triggers real runs at real cost. Without a token it answers only to loopback hostnames; binding beyond loopback without a bearer token is refused.
Treat the page as a window onto the log, not a second authority. Triggered runs land in the same evidence log as CLI runs, and a page refresh never signs anything. Audit wrappers stay terminal commands by design.
03 Sylvae
Through MCP, an agent lists skills and runs one mid-task instead of doing the work inline — typically to spend a cheaper model rather than its own inference. Results return the run ID and namespaced runtime reference for correlation.
The recursion guards are the point: cheap by default, the subscription-sharing backend refused, auto refused. A coordinator may still preallocate a canonical run ID so the delegated run and external evidence share one identity. The receipt, not the transcript, is what travels onward.
04 Sylvae
Nostoi wrappers verify the compatible ledger, attest a head with an explicit principal and key, and check attestations against pinned fingerprints. Daily run records and the audit chain are distinct; the browser excludes the chain from run lists.
Sidecar existence is not verified signing, and a signature never proves a test passed. The human step — signing, reviewing, approving — is never performed by a page load or a background agent. That separation is what makes the remaining evidence trustworthy.
A conversation, not a sales funnel
Ask a question, tell us what you need, offer a contribution or simply show your support. RAGBAZ is a software studio working toward kindness, mutual cooperation and the good of all beings.
Prefer email? ragbaz@proton.me