← Catalog

Cyan Eclipse Promptseal

Prompt change-control with approver receipts over fixture prompts/deployments/incidents — production system prompts are sealed and versioned; every runtime incident can be traced to the exact prompt version and its approver. PASS (local/fixture) — local stdio MCP, fixture prompts/deployments/incidents only, value_class "test"; the chain proves which version was sealed and who approved it — it does not prove the sealed prompt is what actually ran in any live process.

In developmentMCP serverPASS (local/fixture)Trust & Receipts Developer Tools

Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio

Certification scope

PASS (local/fixture) — Verity bench of v0.1.0 (2026-10-27, argus101-program-20261027/wave2/bench-promptseal.md) PASS-CONDITIONAL (local/fixture) with condition C1 (minor, stored-read: incident_link idempotent path echoed stored link before chain check); delta re-bench of v0.1.1 (rebench-promptseal.md): C1 CLOSED, new C2 (lineage binding) + C3 (incident_link idempotent pairing); delta re-bench round 2 of v0.1.2 (rebench2-promptseal.md): C2+C3 CLOSED, new C4 (serve canonicalization) + C5 (pack duplicates unbound) + C6 (whole-store validation); delta re-bench round 3 of v0.1.3 (rebench3-promptseal.md): C4+C5+C6 CLOSED, C1–C6 all CLOSED, no new conditions — CONDITIONS CLOSED, Promptseal — PASS (local/fixture). Builder suite re-run by bencher at every bench: 27/27 at v0.1.0, 30/30 at v0.1.1, 36/36 at v0.1.2, 49/49 at v0.1.3; R3 independent probes: 102/102 checks + 31/31 C6 probes PASS. Decisive fixture F-PS13 independently reproduced: pack written to file, trace_verify in fresh subprocess with empty data dir verified:true, prompt_hash + approver_ref + version_record_hash equal receipt face values, prompt_hash=sha256(body) recomputed. PASS (local/fixture) is not CERTIFIED — certified is false, and nothing beyond the PASS verdict at its exact scope is claimed. Scope is exactly: seal chain + Herald approval binding + incident traceability over fixture prompts/deployments/incidents ONLY; NOT production prompts, real deployments, or runtime attestation (SPEC §11.2 — the chain proves which prompt version was sealed and who approved it, and that a fixture deployment's binding named it; it does not prove the sealed prompt is what actually ran in any live process); dispute use is SPEC §10.1 counsel-gated and UNBUILT — no production ingest, deployment connector, or runtime hook exists. Novel mechanism (SPEC §2.1, verbatim): production system prompts are sealed and versioned; every runtime incident can be traced to the exact prompt version and its approver. Stored serves in v0.1.3 are canonical (exact per-type field sets, injected counter_signature/unknown fields are chain defects), pack duplicates are bound exactly to chain copies, and the whole store is validated at load on every path. Non-blocking observations O1 (lineage_status lists state:APPROVED under chain_ok:false, disclosed in-band, not an export/receipt) and O2/R3 observations remain observations, no conditions. Fixture Herald key is a test double (pinned herald-fixture, keys/ gitignored mode 600); Promptseal consumes Herald-signed approval records, issues none, redefines none; value_class:"test" throughout.

What it does

Tools

prompt_sealapproval_attachdeployment_bindincident_linklineage_statustrace_gettrace_verifypack_get

Problems it addresses

Not tested

Evaluation-intake listing

This listing records the tool's place in the Bazaar's evaluation and monitoring pipeline, with the evidence gathered so far printed above — verdict, scope, and not-tested items included. It is not sold on this site, and its presence in the catalog is not a claim beyond that evidence. The service behind every verdict is the bench: what a bench run buys →