cross-review

Run Until Unanimous

run_until_unanimous

Run iterative peer review with multiple AI models until all reviewers unanimously vote READY or the configured round limit is reached. Evidence is persisted durably and automatically transported to every reviewer.

Instructions

Generate or revise a draft and continue real API peer-review rounds until unanimous READY or the configured max_rounds is reached. AI evidence supplied in evidence is persisted durably and transported automatically; no separate attachment step is required. v2.11.0: when caller is set to a peer id (claude|codex|gemini|deepseek|grok|perplexity), the relator lottery activates: omit lead_peer to have the server randomly select a non-caller peer as relator (modeled on judicial panels), or supply an explicit lead_peer that is NOT the caller. An explicit lead_peer === caller is rejected at the server with caller_cannot_be_lead_peer — an agent never reviews itself (workspace HARD GATE).

Input Schema

TableJSON Schema

Name	Required	Description	Default
`mode`	No		ship
`task`	Yes
`peers`	No
`caller`	Yes
`evidence`	No	Raw evidence from the authenticated AI caller. It is persisted automatically as durable, SHA-256-addressed caller_submitted_unverified material and transported to reviewers. This is the routine path; `session_attach_evidence` attaches the same material out of band and grants it no additional provenance.
`lead_peer`	No
`max_rounds`	No	Hard review-round ceiling unless allow_auto_extension is explicitly true.
`max_cost_usd`	No
`review_focus`	No	Optional provider-neutral review scope anchor. This is not Claude Code's /focus UI command; it is injected as a front-loaded Review Focus prompt block for every selected peer, including OUT OF SCOPE handling for unrelated findings.
`initial_draft`	No
`until_stopped`	No
`response_format`	No		json
`allow_auto_extension`	No	Opt in to at most two evidence-only auto-extensions. False keeps max_rounds rigid.
`reasoning_effort_overrides`	No	Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex\|claude\|gemini\|deepseek\|grok\|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-6 Astra, GPT-5.6, Kimi K3 via Perplexity and DeepSeek, and xhigh on Grok 4.6; older GPT-5 families use their own ceilings).

Schema Changelog

Changes observed during successful MCP inspections.

Changed5 schema fields changedv4.4.18Sep 17, 2026

removedInput schema / properties / caller / default
```
Removed value: -"operator"
```

changedInput schema / properties / caller / enum

Previous value: -[
-  "codex",
-  "claude",
-  "gemini",
-  "deepseek",
-  "grok",
-  "perplexity",
-  "operator"
-]New value: +[
+  "codex",
+  "claude",
+  "gemini",
+  "deepseek",
+  "grok",
+  "perplexity"
+]

changedInput schema / properties / evidence / description

Previous value: -"Raw evidence from the authenticated AI caller. It is persisted automatically as durable, SHA-256-addressed caller_submitted_unverified material and transported to reviewers; no manual operator attachment is required. Do not call session_attach_evidence for this routine path."New value: +"Raw evidence from the authenticated AI caller. It is persisted automatically as durable, SHA-256-addressed caller_submitted_unverified material and transported to reviewers. This is the routine path; `session_attach_evidence` attaches the same material out of band and grants it no additional provenance."

changedInput schema / properties / reasoning_effort_overrides / description

Previous value: -"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-5.6, Kimi K3 via Perplexity and DeepSeek, and xhigh on Grok 4.6; older GPT-5 families use their own ceilings)."New value: +"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-6 Astra, GPT-5.6, Kimi K3 via Perplexity and DeepSeek, and xhigh on Grok 4.6; older GPT-5 families use their own ceilings)."

changedInput schema / required

Previous value: -[
-  "task"
-]New value: +[
+  "task",
+  "caller"
+]

Changed1 schema field changedv4.4.13Aug 26, 2026

changedInput schema / properties / reasoning_effort_overrides / description

Previous value: -"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-5.6 and high on Grok 4.5; older GPT-5 families use their own ceilings)."New value: +"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-5.6, Kimi K3 via Perplexity and DeepSeek, and xhigh on Grok 4.6; older GPT-5 families use their own ceilings)."

Changed10 schema fields changedv4.4.9Jul 14, 2026

addedInput schema / properties / allow_auto_extension

Added value: +{
+  "default": false,
+  "description": "Opt in to at most two evidence-only auto-extensions. False keeps max_rounds rigid.",
+  "type": "boolean"
+}

addedInput schema / properties / evidence / description

Added value: +"Raw evidence from the authenticated AI caller. It is persisted automatically as durable, SHA-256-addressed caller_submitted_unverified material and transported to reviewers; no manual operator attachment is required. Do not call session_attach_evidence for this routine path."

addedInput schema / properties / max_rounds / description

Added value: +"Hard review-round ceiling unless allow_auto_extension is explicitly true."

changedInput schema / properties / reasoning_effort_overrides / description

Previous value: -"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. Useful to dial down expensive peers (e.g. Grok grok-4.20-multi-agent xhigh = 16 agents, or Perplexity sonar-deep-research that bills citation + reasoning + search queries separately) for routine reviews without editing the host MCP configs."New value: +"Optional per-peer reasoning_effort overrides for this call. Keys are peer ids (codex|claude|gemini|deepseek|grok|perplexity); missing keys fall back to global config. This is a shared scale: adapters normalize unsupported literals to the selected model's documented enum (`ultra` becomes max on GPT-5.6 and high on Grok 4.5; older GPT-5 families use their own ceilings)."

changedInput schema / properties / reasoning_effort_overrides / properties / claude / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

changedInput schema / properties / reasoning_effort_overrides / properties / codex / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

changedInput schema / properties / reasoning_effort_overrides / properties / deepseek / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

changedInput schema / properties / reasoning_effort_overrides / properties / gemini / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

changedInput schema / properties / reasoning_effort_overrides / properties / grok / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

changedInput schema / properties / reasoning_effort_overrides / properties / perplexity / enum

Previous value: -[
-  "none",
-  "minimal",
-  "low",
-  "medium",
-  "high",
-  "xhigh",
-  "max"
-]New value: +[
+  "none",
+  "minimal",
+  "low",
+  "medium",
+  "high",
+  "xhigh",
+  "max",
+  "ultra"
+]

First observedv4.4.8Jul 4, 2026

TDQS

A4/5.0

Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations only flag readOnly=false, idempotent=false, and openWorld=true. The description adds valuable behavioral context: evidence is 'persisted durably and transported automatically,' review rounds hit 'real API' peers, and an explicit lead_peer === caller is rejected server-side via 'caller_cannot_be_lead_peer' as a 'workspace HARD GATE.' This goes beyond the annotations, though it does not cover side effects like session creation or response behavior.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is dense but not bloated. The core purpose is front-loaded in the first sentence, followed by evidence-transport behavior and the relator-lottery rules. The 'judicial panels' metaphor is mild color but supports understanding. The version marker and hard-gate detail are useful, though the description is longer than strictly necessary for the tool's most common invocation.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness3/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

This is a complex, open-world, mutating tool with 14 parameters and no output schema. The description explains the main loop, evidence flow, and relator constraint, but it does not describe return values, output format, session-related side effects, cost behavior, or how to set up peers and initial drafts. An agent could call it correctly with task and caller, but many behavioral expectations remain underspecified.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is only 36%, so the description must compensate. It does add meaningful semantics for key parameters: evidence (durable, SHA-256-addressed, no attachment needed), caller and lead_peer (relator lottery, non-caller constraint, server rejection), and max_rounds (termination ceiling). However, most other parameters — mode, peers, initial_draft, response_format, until_stopped, and reasoning_effort_overrides — are left to the schema or omitted entirely, leaving a partial compensation gap.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a specific verb and resource: 'Generate or revise a draft and continue real API peer-review rounds until unanimous READY or the configured max_rounds is reached.' This clearly states the tool's core behavior and terminal condition, making it distinguishable from sibling session-control tools like session_start_round or session_poll.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides clear operational guidance: evidence is automatically persisted and transported, so 'no separate attachment step is required' — a direct hint about when not to use the sibling session_attach_evidence. It also explains how to use the relator lottery via caller and lead_peer, including the rejection rule. It stops short of explicitly naming when to prefer run_until_unanimous over session_start_unanimous or ask_peers, so no exclusions are given.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Run Until Unanimous

Instructions

Input Schema

Schema Changelog

TDQS

Other Tools