Loading ToolBench
Trust tools for agent-to-agent dealings — prove, vet, and bind with verifiable evidence (MCP server)
20 tools with mostly complete schemas and descriptions, but significant gaps in parameter documentation and error handling guidance. Tool names follow verb_noun convention well (orient, learn, start_conversation, make_claim, verify_claim). Descriptions range 50-200 chars, meeting baseline. However, many parameters lack type constraints (enums, ranges), and error recovery guidance is absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Output schemas not documented. Composition is sound, tools are single-responsibility and chainable (e.g., start_conversation → add_message → end_conversation). Security-sensitive tools (make_claim, prove_independence) lack confirmation/dry-run patterns.
Add a local evaluation of a counterparty to the journal
Add a message or event to an active conversation
End a conversation, persist it, and return summary
Query reputation and identity information for an agent across multiple providers
Export a conversation as verifiable JSON
Retrieve an agent's published profile from the directory
Deep dive on any topic: interaction forms, trust networks, verification, claims, reasoning, and area guides
List all active conversations
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite explicit risk classifications. Tools marked WRITE (start_conversation, make_claim, record_interaction, etc.) should declare destructiveHint=true to guide agent planning.
Output schemas not documented. Tools like orient, export_conversation, list_conversations, my_identity lack visible return type definitions. LLMs cannot plan downstream calls without knowing response structure.
No error recovery guidance. Tools lack descriptions of failure modes and recovery steps. E.g., verify_claim should document what happens if signature is invalid, commitment hash mismatches, or witness seal verification fails.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Create a verifiable signed claim
Get the agent's own identity (DID and public key)
Situational guide: what applies to your situation, your identity, and anything in flight
Create a sealed commitment for independent assessment
Publish agent identity to the Synpareia directory
Retrieve local journal record of a counterparty
Record an interaction outcome with a counterparty on the network
Reveal a previous commitment and verify it matches
Start a new verified conversation with optional counterparty and description
Verify a claim signature, identity, commitment, or witness seal
Request a witness seal for a signed claim (second step of witness attestation)
Verify a witness seal offline against the witness's published key
Parameter constraints missing. Many parameters accept free-form strings without enums or format constraints. E.g., event_type in add_message lists 'message, thought, observation, or decision' in description but no enum schema. claim_type in verify_claim similarly lacks enum.
Irreversible operations lack confirmation/dry-run patterns. make_claim, prove_independence, record_interaction, and publish_profile modify state but offer no confirmation step or preview. Agents can accidentally publish profiles or record false interactions.