Loading ToolBench
MCP adapter for ScopeBlind Agent Vault: signed context capsules, memory proposals, and offline verification for Hermes/GBrain-style agents.
Server has 8 tools with explicit schemas and descriptions. Naming follows verb_noun pattern (vault_*) consistently. Descriptions are present (avg ~120 chars) but lack actionable context for LLM selection. Schemas use proper JSON Schema with types, but many parameters lack descriptions or constraints. No output schemas documented. Error handling is minimal. Security model relies on environment variables for secrets, which is correct, but no audit logging or permission gates visible.
Alias for vault_context_request, provided for runtimes that distinguish context requests from export operations.
Verify a Context Capsule signature and included signed page signatures offline.
Issue a signed Context Capsule: only allowed pages are disclosed; sealed pages remain committed but hidden.
Explain which pages would be disclosed, withheld, or sealed for a requested set of kinds and auth proof.
Read an Agent Vault manifest and return stable identity/state hashes for a runtime like Hermes or GBrain.
Create a signed memory proposal. It does not mutate the Vault; the owner must approve before the State Sigil should change.
No output schemas documented. LLMs cannot predict response structure or plan downstream calls. Tools return objects but structure is inferred from code, not declared.
Parameter descriptions are sparse or missing. 'auth_proof' is described as 'BRASS auth proof or demo proof' but lacks format/structure guidance. 'requester' is '{kind, id, label}' but not validated. LLMs cannot construct valid payloads.
No error handling guidance. Code does not show try-catch blocks, validation, or recovery hints. If auth_proof is invalid or manifest_path does not exist, LLM receives raw error with no actionable next step.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
Sign a page payload with an explicit agent secret. This is for owner-approved writes only, not automatic memory writes.
List signed Agent Vault pages with sensitivity labels; content is redacted unless requested.
vault_capsule_export is an alias for vault_context_request. Duplicate tools with overlapping functionality force LLM to reason about which to call. Should consolidate or clearly document when each is preferred.
Descriptions lack WHEN-to-use context. 'vault_disclosure_policy_check' vs 'vault_context_request', when should an LLM call one vs the other? Descriptions do not explain the workflow or decision tree.