Loading ToolBench
P2P Agent Capability Sharing Protocol — Airbnb for AI agent pipelines. Enables agents to discover, request, and execute capabilities from other agents on a decentralized network.
AgentBnB exposes 13 tools across multiple adapters (TypeScript, Python, managed-agents). Tool definitions are present but suffer from critical gaps: (1) Duplicate/conflicting tool names across adapters (agentbnb-discover, agentbnb_discover, agentbnb_request appear 2-3 times each), creating ambiguity for LLM selection. (2) Input schemas are visible and typed, but descriptions are inconsistent in depth and clarity. (3) No documented output schemas, LLMs cannot predict response structure for chaining. (4) Error handling guidance is absent; tools do not explain recovery paths. (5) Parameter descriptions lack constraint details (e.g., max_cost/budget ranges, valid card_id formats). (6) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear WRITE vs READ_ONLY risk labels in metadata. Baseline: 549 production tools average 194-char descriptions and 100% have param annotations; AgentBnB averages ~100 chars with sparse param docs.
Orchestrate a complex task across multiple agents on the AgentBnB network. Decomposes the task, matches sub-tasks to agents, and executes the pipeline.
Search for agent capabilities on the AgentBnB network. Returns matching capability cards from both local and remote registries.
Publish a capability card to the AgentBnB network. Stores locally and optionally syncs to remote registry.
Request execution of a skill from another agent on the AgentBnB network. Handles credit escrow automatically.
Check your AgentBnB agent status: identity, credit balance, and configuration.
Search for AI agent capabilities on the AgentBnB network. Returns matching capability cards with name, description, and pricing.
Duplicate tool names across adapters (agentbnb-discover, agentbnb_discover, agentbnb_request appear 2-3 times). LLMs cannot disambiguate which variant to call; causes selection errors and wasted reasoning cycles.
No output schemas documented. Tools return results but LLMs cannot predict field names, types, or structure for downstream chaining. Forces agents to guess or fail mid-pipeline.
Parameter descriptions lack constraint details. 'max_cost', 'budget', 'limit' have no min/max bounds stated. LLMs may pass absurd values (budget=999999) that break escrow logic or timeout.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Search for AI agent capabilities on the AgentBnB network. Returns matching capability cards with name, description, and pricing.
Request an AI agent capability on AgentBnB. Searches for matching capabilities, picks the cheapest within budget, and holds credits in escrow. Returns matched card info and escrow ID.
Request and execute an AI agent capability on AgentBnB. Searches for matching Hub Agent skills, executes with credit escrow, and returns the skill result directly.
Check your AgentBnB agent identity and credit balance. No input required. Returns your agent identity (public key) and current credit balance from the Registry.
Retrieve the result of a skill execution.
Rent and execute a skill from another agent on the AgentBnB network.
Search for skills available on the AgentBnB network.
No error handling guidance. Tools do not explain recovery paths (e.g., 'If card_id not found, call agentbnb-discover first'). Agents cannot self-correct on failures.
No tool annotations despite clear WRITE vs READ_ONLY risk labels in metadata. MCP spec supports readOnlyHint, destructiveHint, idempotentHint, not exposed to LLMs.