Provenance

Security claims are everywhere, and almost all of them are taken on trust: a scanner says "this app is vulnerable," a report assigns a grade, and nobody can show the working. Provenance replaces trust with receipts. Provenance (the SecScan RedTeam app) is an autonomous red team that lives on one chat page. You describe the target in plain language. Before anything is tested, the system verifies that you own it — for web targets, verification is server-authoritative, checked at the moment of dispatch. Outside verified scope, the fleet sends zero packets. Then five agents, all seated on Muse Spark through the Meta Model API — Morpheus (Lead), Operator (Recon), Neo (Exploiter), Architect (attack design and toolbuilding), Reporter — plus a dedicated Scribe seat, plan the engagement, map the surface, attack it, and challenge their own findings before anything is reported. The proof bar is the product. A finding is PROVEN only when demonstrated end to end and captured in a replayable receipt with a passing negative control; a second independent observation makes it CONFIRMED; everything else is labeled DEFERRED or RULED OUT, never quietly upgraded. Grades come from one anchored scoring engine: confirmed findings deduct in full, heuristics are damped and capped, and one proven critical anchors the grade at F. A run that cannot prove anything says so — halted runs report Grade N/A rather than a hollow A. The finished report renders in the same chat thread: the grade, a severity breakdown, and every finding with its evidence and a paste-ready fix. The entry was built with Meta's models on both sides of the glass. Inside the product, Muse Spark runs the fleet. Outside it, the product itself was built by a Muse Spark agent, Cho-zen1, working through AlterEgo — our own bridge that gives the agent a tool surface over a real Windows PC (109 tools, registry-verified): running builds, firing test suites, operating the app, and verifying every result on the machine. On the final morning, Meta's own coding agent, Muse Code, audited AlterEgo in the open and raised four material findings; we verified each against the code, the auditor implemented the remediation itself in the public repo, its landing commit passed independent verification (487/487 tests), and the fixed code was deployed to the running bridge the same day. The demo video is the visible record of that collaboration — the founder's and the agent's actual working messages, start to finish. All of the dialog between myself and Muse shows as evidence word for word.