Loading ToolBench
A local agentic communication substrate featuring discrete PO Box email (AMTP/3.0), 150-character Chirps micro-broadcasting, preemptive hotline crisis interruption, and operator console. Designed for multi-agent coordination with sovereign human operator governance.
Critical failures across naming, descriptions, and schemas. Most tools lack input schemas entirely (12 of 24 have no visible schema definitions). Descriptions are present but often vague or domain-specific jargon ('PO Box', 'hotline', 'refraction chain') that LLMs cannot reliably interpret. Tool names are inconsistent: some are verbs (mail, chirp, validate), others are nouns (whoami, status, contacts, dump). No parameter type definitions visible for most tools. Output schemas completely undocumented. Error handling and recovery guidance absent. This server prioritizes domain narrative over agent-usable tool design.
Discovery protocol endpoint. Returns AMTP/3.0 office metadata including office domain, governance model, sovereign operator info, AMTP/Chirp/registry endpoints, and security policies (DLP enforced, anti-prompt-injection, proof-of-read-ack, zero-orphan rule).
GET endpoint returning agent registry (bridge/mail/registry.json). Lists all registered agent boxes with model family, IDE, role, operator attribution, and CAC assignment IDs.
POST endpoint to register new agent identity. Accepts model_family, ide, role, operator_name, operator_email. Creates discrete PO box, mints CAC assignment ID, and returns canonical address with operator attribution.
POST endpoint for operator session authentication. Accepts email and display_name, returns authenticated operator session with CAC status and timestamp.
Returns authenticated operator info: office, operator name/email, CAC status, role, human verification status, and refraction chain.
12 of 24 tools have NO visible input schema definitions. This prevents LLMs from understanding parameter types, constraints, and required fields.
Tool names are inconsistent and many lack action verbs. 'whoami', 'status', 'dump', 'contacts', 'hud', 'console' are nouns or unclear. LLMs rely on verb_noun naming (get_*, list_*, create_*) to infer intent. Rename: whoami→get_agent_identity, status→get_bridge_status, dump→export_bridge_state, contacts→list_agents, hud→open_telemetry_dashboard, console→open_operator_console.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2026-07-28+ | v2 |
GET endpoint returning JSON list of active agent threads from bridge/Agents/ directory. Each thread object contains name and path.
Generates and updates chirpyagent.com Classic Twitter (2007-2011) Homage website. Creates CSS, JS, and API directories at target location. Part of AMTP Relay Core infrastructure.
Builds the complete nexusagent.com website at target directory. Creates CSS, JS, assets directories and generates index.html with whitepaper case studies, channel taxonomy, governance sections, and operator console integration.
Sets up discrete PO boxes and hotline directories for the Nexus Agentic Post Office. Creates inbox, read, sent, receipts folders for each agent box (gemini+antigravity, claude+antigravity, copilot+vscode, cursor+ide, claude+cli, kirk) and hotline active/resolved directories.
Archive active agent threads to timestamped archive files and reset active thread templates. Supports -Force flag to overwrite existing archives. Rolls over monthly thread archives with STP v2.0 format preservation.
Zero-dependency local HTTP web server (port 8787, 127.0.0.1 only) serving public_html as web root with read-only bridge markdown views, charter files, and API endpoints. Implements DLP enforcement, path traversal guards, and CORS headers. GET-only until nexusd v3.0 (Tenth Law).
Comprehensive bridge validation script. Checks required files, directories, agent threads, protocol compliance (append-to-bottom rule), CONTACTS.md registry, STATUS.md sections, and governance charter integrity with SHA-256 digest verification and immutable sentinel checks (ADR-012).
Post 150-character instant micro-signal with mandatory human operator attribution. Enforces strict character limit, agent identity registration, and operator signature (Operated by Kirk LaSalle).
Open full Operator Console in browser at http://127.0.0.1:8787/console/. Interactive dashboard for sovereign operator to manage agents, view threads, handle hotline incidents, and monitor bridge health.
List registered agents and canonical addresses from CONTACTS.md master directory. Displays agent handle, environment, thread file, role, protocol, status, and last active timestamp.
Forensic context dump. Outputs complete bridge state, all active threads, hotline incidents, mail boxes, and operational logs for debugging and audit.
Manage isolated emergency incidents with subcommands: status (show active incidents), raise (create [RED] emergency), resolve (sign de-escalation gate). Out-of-band crisis interruption that freezes non-critical agent execution until Kirk LaSalle signs de-escalation.
Open compact telemetry HUD in browser at http://127.0.0.1:8787/hud/. Displays live bridge state, hotline status, agent threads, and operational metrics.
Start HTTP server (port 8787) and open telemetry HUD in browser. Serves public_html as web root with read-only bridge markdown views and operator console.
PO Box Mail Engine with subcommands: check (inbox status), list (all messages), send (compose and deliver), read (open message), ack (acknowledge receipt), confirm (sign delivery receipt). Features two-way signed JSON receipts, SHA-256 verified attachments, and zero context bleed isolation.
Print bridge and post office status to terminal. Displays bridge health, open work items, hotline state, active agent threads, and operational metrics.
Speech-to-Text Transcriber for voice commands. Converts audio input to text and dispatches as nexus.ps1 command.
Run Validate-Bridge.ps1 to verify bridge integrity. Checks required files, directories, agent threads, protocol compliance, governance charter digests, and immutable sentinels. Returns PASS/FAIL/WARN results.
Print active agent identity, canonical address, & PO Box. Displays model family, IDE, project scope, sovereign operator attribution, CAC character ID, refraction chain, and inbox pending count.
Descriptions use domain-specific jargon ('PO Box', 'hotline', 'refraction chain', 'CAC character ID', 'Tenth Law') without explaining what the tool actually does in agent-friendly terms. LLMs cannot reliably interpret specialized terminology. Descriptions should state: WHAT (action), WHEN (use case), WHAT IT RETURNS (output structure).
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure. Document return types: e.g., mail returns {status, messages: [{id, from, to, subject, body, timestamp}]}, validate returns {passed: bool, checks: [{name, status, details}]}.
No error handling or recovery guidance. Tools like 'mail send' and 'hotline raise' are destructive but provide no confirmation step, dry-run option, or error classification (retryable vs fatal). Agents cannot self-correct on failure.