Risk OS Red Try Stoa

The report

stoa scan writes a self-contained stoa-report.html — no network, no accounts, opens offline. It is verdict-first: the answer to "how bad is it, and what do I fix" is at the top; the exhaustive detail is one collapse away.

Information architecture

The report reads top to bottom as a decision, not a database:

SectionThe question it answers
Verdictone generated sentence — agent count, how many at the worst tier, and the theme of the risk
Scoreboardhow much risk, and where it sits — agents by exposure tier + a one-line findings ribbon
Fix firstwhat to repair, in order — the shortest path that clears every critical finding
Contradictionswhere declarations don't match the code — what a self-attested questionnaire can't catch
Agentswhich agents carry the risk — the severe tier up front, the rest one click down
Risk dimensionswhat kinds of risk, and which rules they map to
Standardswhich OWASP LLM classes were assessed — gaps kept visible
Appendixfull agent detail, every finding, NIST alignment, and the architecture graph — collapsed

Every number reconciles, and nothing is deleted — only demoted. The full finding list, the per-agent breakdowns, and the graph all live in the Appendix; the body leads with the verdict and the fixes.

Built to be read by a first-timer

An inline explainability layer decodes Stoa's vocabulary in place, without a glossary section:

code makes possible, not what has happened*);

exposure*, gate-eligible, the exposure levels, suppressed;

and framework tags.

Fix first, in four lines

Each remediation item is exactly: a title, a one-line impact, the fix, and a rule · file:line · OWASP · Article reference. Findings that share a rule and a fix are merged into one item; a critical contradiction points down to the Contradictions section rather than repeating its detail. Flow analysis and methodology caveats sit behind a Why this fired expander.

Properties

makes no network request. Its content-security policy hash-pins every inline script and forbids remote loads.

stays collapsed.

sets the fix-first cap, the agent-collapse tier, and the contradiction grouping threshold. See Configuration.

There is also a Download report button, and a Generate underwriting evidence action — see Underwriting evidence.

Open a live one: the demo scan report is a real scan of the Meridian reference app.