Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio
Certification scope
CERTIFIED(simulated) — Eclipse Bazaar Test Bench 2026-10-25, scoped re-bench (fix round 1): 26 pass, 0 na of 26 checks — the first bench's two major defects (D-ENV: envelope type guard at both serving layers; D-REVOKE: pack_fetch re-checks grant state after revoke, byte-identical refusal, no oracle) fixed and re-probed PASS, regression spot-lines A1–A6 / B0–B8 / C1–C8 / G1 unbroken; the full bench script also ran end-to-end out-of-scope at 42/42. The original PARTIAL card (40 pass / 2 fail of 42 — test-bench/results/eclipse-connect-beyondcompare-mcp-2026-10-25.json) is retained append-only and this re-bench card is the standing verdict basis. CEILING — read exactly: CERTIFIED(simulated) and nothing broader. The adapter is a scripted stand-in throughout (the documented-CLI adapter is a stubbed interface only); evidence class SIMULATED; NO live Beyond Compare run was exercised at any point in either bench run — live BC exercise is OWED at P1 on the owner's license (vendor trial fallback) before any live claim. MECHANISM: Herald-signed recomputable compare-evidence packs over mandate-granted paths; merge = T3 ECDSA approval-bound with base-drift refusal. N-CHECK: NOVELTY_PROVEN — 0/6 renamed vendor verbs (fail line ≥80%), shim score 1/20 (ceiling ≤20%); the novelty is the mandate/custody + evidence layer, never the bridge. Not Scooter Software. Not endorsed by Scooter Software. The customer supplies their own Beyond Compare license.
What it does
- A compare verdict a third party can recompute: inputs, per-file SHA-256 hashes, tool version, bridge-generated script and comparison rules pinned and hash-chained in a signed evidence pack (Blackbox form) — verify_pack recomputes offline and names the first failing pin
- Mandate custody: packs issue only for paths inside a Herald-signed mandate — granted roots canonicalised and enforced bridge-side; escapes refuse by name (PATH_OUTSIDE_MANDATE / PATH_TRAVERSAL_REFUSED / SYMLINK_ESCAPE_REFUSED) with zero adapter calls
- Merge as a T3 verb only: ECDSA principal-signed, params-bound approval; pack recompute plus a live re-compare at dispatch refuses MERGE_BASE_DRIFTED before a single byte is written; partial failure seals an INCOMPLETE result pack
- Denylist by enumeration: no script pass-through, no CLI-flag verbs, no sync/mirror/edit/interactive-merge verbs exist — forbidden names get the identical -32602 shape as gibberish, no oracle
- The agent holds nothing: no license key, no filesystem reach, no vendor executable, no script file — only a revocable agent key bound to (agent_identity, bridge_id); revocation is terminal and re-checked at every dispatch, pack_fetch included
- Vendor reports ride inside packs as hashed attachments only — never the verdict, never returned standalone as evidence
Tools
compare_pathspack_fetchverify_packpropose_mergeexecute_mergemandate_checkrevoke_grant
Problems it addresses
- My agent needs file and folder comparisons I can defend later — a /qc exit code or an HTML report is an assertion by the machine that ran it, not evidence
- I don't want my agent holding my Beyond Compare license, a shell, or arbitrary filesystem reach just to compare two trees
- A merge an agent performs must be approved, bound to the exact plan, and refused automatically if the base moved since the compare
Not tested
- Live Beyond Compare exercise — no licensed install was run in either bench; the adapter is a scripted stand-in throughout and the documented-CLI adapter is a stubbed interface only. The live bench is OWED at P1 on the owner's license (vendor trial fallback) before any live claim.
- Hosted-relay / multi-seat form — P0 is a single-seat edge bridge on the license-holding machine; the hosted form is a P2 decision-gated shape, not built.
- Production Herald key custody — P0 signs mandates and T3 approvals through the PinSigner port's local signer (identical grammar; artifacts record signer: local); production custody is a gated owner decision, not a P0 claim.
Bench scorecard: test-bench/results/eclipse-connect-beyondcompare-mcp-2026-10-25-rebench-fixround1.json
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 →