stoa export --assurance
Enterprise assurance frameworks — insurance underwriting, AIUC-1 (the AI agent trust standard), vendor security reviews — ask for data across 18 areas, split across three source layers: scanned (derivable from code), declared (human-supplied business facts, see Declarations), and ingested (external artifacts — eval results, incident logs — Stoa reserves structured slots for these but doesn't generate them).
The 18 areas are grouped under AIUC-1's six standard categories (Data & Privacy, Security, Safety, Reliability, Accountability, Society), plus a seventh Stoa-only group — insurance-specific exposure (business exposure, economic authority, claims evidence) — that AIUC-1 doesn't cover, because AIUC-1 is a trust standard, not an insurance standard. This grouping is a display header, not an AIUC-1 certification claim; certification requires their accredited-auditor process.
stoa export --assurance walks all 18 areas and emits a packet where every row is tagged with exactly one status — never silently omitted:
| Status | Meaning |
|---|---|
scanned | Derived from the scan, with a rule id and file:line. |
declared | From stoa-declared.toml, with the exact key path. |
ingested | A pointer from the evidence.* block (kind, ref, date). |
not_provided | Explicitly listed as a gap — an assurance reviewer needs to see what's missing as much as what's covered. |
Usage
stoa export --assurance [REGISTRY] [--format json|md] [--out PATH]REGISTRYoptional: read an existingstoa-registry.json; omitted → scans
the current directory fresh.
--format md(default): human/reviewer-readable, one table per area.--format json: the raw packet,assurance-packet/1.2.--out PATH: write to a file; omitted → stdout.
stoa scan .
stoa export --assurance stoa-registry.json --format md --out assurance.mdThe 18 areas
| # | Group | Area | Layer(s) |
|---|---|---|---|
| 1 | Index | AI inventory | scanned + declared |
| 2 | Index | Historical evidence | ingested |
| 3 | A — Data & Privacy | Data access | scanned + declared |
| 4 | B — Security | Permissions | scanned |
| 5 | B — Security | Dependencies | scanned |
| 6 | B — Security | Technical controls | scanned |
| 7 | B — Security | Security testing | ingested |
| 8 | C — Safety | Autonomy | scanned (inferred) + declared (intent) |
| 9 | C — Safety | Safety evaluation | declared + ingested |
| 10 | D — Reliability | Reliability scores | scanned |
| 11 | E — Accountability | Accountability | declared + ingested |
| 12 | E — Accountability | Monitoring | ingested (+ CTRL004 scanned) |
| 13 | E — Accountability | Contracts | declared + ingested |
| 14 | E — Accountability | Vendor due diligence | ingested |
| 15 | F — Society | Societal impact | declared (attestation only — never scored) |
| 16 | G — Insurance-specific | Business exposure | declared |
| 17 | G — Insurance-specific | Economic authority | declared + scanned (enforcement check) |
| 18 | G — Insurance-specific | Claims evidence | ingested (reserved observed provenance — see below) |
Groups A–F mirror AIUC-1's own six categories, split further where AIUC-1 itself draws a distinction Stoa's data already supports — e.g. AIUC-1 separates third-party adversarial testing (Security) from third-party harmful-output testing (Safety), so those are two areas (7 and 9), not one. Group G has no AIUC-1 equivalent: it's the loss-exposure data an insurance submission needs that a general agent-trust standard was never built to ask for.
Most per-agent areas are self-explanatory tables (one row set per scanned agent); area 15 (Societal impact) and area 2 (Historical evidence) are repository-level, as are most of Group G.
Area 18 (Claims evidence) is deliberately all not_provided today: Stoa is a static scanner and has no runtime traces to report. The observed edge-provenance value (already reserved in the architecture graph and SCHEMA.md) is where a future runtime-trace overlay would surface here. Area 15 (Societal impact) is also deliberately never scored — see dimensions.md for why Stoa treats AIUC-1's Society category as attestation-only.
Contradictions section
Every packet leads with a dedicated Contradictions section — every DECL001-DECL007 finding from the current scan, with both evidence sides (code + declared) inline. This is the headline for a reviewer: it's the one thing a self-attested questionnaire can't produce on its own.
Determinism
The packet body is deterministic — identical registry input produces an identical areas/contradictions output. The only place wall-clock or caller-supplied values appear is the header block (git_sha, scan_timestamp), matching the rest of Stoa's no-timestamps-in-content invariant.
Sample packet (Markdown, trimmed)
## Stoa · Assurance Packet — support-desk
`a1b2c3d` · scanned 2026-07-25T12:00:00Z · Stoa `0.3.0` · registry schema `1.2` · 9 agent(s) · 1 contradiction(s)
### ⚠️ Contradictions (1)
| Rule | Severity | Title | Code | Declared |
|---|---|---|---|---|
| DECL001 | critical | Declared autonomy contradicts inferred autonomy | `agents/billing_executor.py:14` | `agents."509c9ce8c11d".autonomy_intent` |
## Index
### Area 1 — AI inventory (scanned + declared)
...
## G — Insurance-Specific Exposure (beyond AIUC-1)
### Area 16 — Business exposure (declared)
| Field | Status | Evidence |
|---|---|---|
| industries | 📝 declared | `business.industries` |
| revenue | ⬜ not_provided | — |
...--sign
No signing mechanism exists in Stoa today. This is a documented extension point (# TODO(assurance-sign) in src/stoa/cli.py), not a built feature — deliberately deferred rather than half-implemented.