ase-meta-quorum

@${CLAUDE_SKILL_DIR}/../../meta/ase-control.md @${CLAUDE_SKILL_DIR}/../../meta/ase-skill.md @${CLAUDE_SKILL_DIR}/../../meta/ase-getopt.md

$ARGUMENTS

1. The user-selectable foreign models are restricted by the --models/-m option, parsed into as a comma-separated list of model tokens. The getopt parser validates only the first token, so you MUST validate each remaining token yourself against the allowed set all, chatgpt, gemini, deepseek, grok, glm, qwen, claude, codex, copilot. If any token is not in this set, bind to that offending token, then only output the following and then immediately STOP processing the entire current skill:

   Prepare the LLM query by setting to the following :

   For yourself (Anthropic Claude), first answer this in advance yourself by showing your own answer to the query as a sneak preview. For this, output the following :

2. The default is the single token all. If contains the token all, you MUST treat it as the full list chatgpt,gemini,deepseek,grok,glm,qwen (all API-based models). The harness-based models claude, codex, and copilot are never part of all and are queried only when their token is given explicitly (in addition to all, if wished). Anthropic Claude (yourself) is always included, independent of this option.

       Agent(
           description:       "Query Foreign LLM: <arg1/>",
           subagent_type:     "ase:ase-meta-chat",
           prompt:            "<arg2/> <query/>",
           run_in_background: false
       )
   

   Query only those foreign models whose token is contained in (where all selects every model); silently skip all others. Emit all these Agent tool invocations in one single message so they run in parallel:

   You MUST NOT output anything in this step.

3. Agents which returned a response with an ERROR: prefix are silently skipped and are treated as not available.

   Summarize all responses, of both yourself and all available agents, with just 1 to 7 corresponding bullet points and with short sentences.

   You MUST NOT output anything in this step.

4. First, count the number of available foreign AIs (those queried in STEP 2 which did not return an ERROR: response). If this count is zero -- i.e., every foreign AI errored out or none were reachable and only yourself (Anthropic Claude) remains -- a quorum is not possible. In this case, set and to empty and instead set to (no quorum: no foreign AIs were available), then skip the rest of this step.

   Otherwise, let be the total number of responders (yourself plus all available foreign AIs above). Then determine, on a Likert scale of 0.., the amount of the overall consensus of all the responses. If all responses disagree, the consensus is zero. If all responses agree, is .

   If not all AIs agree, determine information, formatted as (disagreement: <ai/>, <ai/>, [...]) where is a name of an AI which disagreed with the consensus. Else, if all AIs agree, set .

   You MUST NOT output anything in this step.

5. Finally, show the summary, the consensus, and the complete and unmodified responses of yourself and each of the queried foreign AIs, based on the following output :

   ● CONSENSUS ANSWER:

   • [...]
   • [...]

   CONSENSUS RATE: /

   ○ Anthropic Claude:

   • [...]
   • [...]

   ○ OpenAI ChatGPT:

   • [...]
   • [...]

   ○ Google Gemini:

   • [...]
   • [...]

   ○ DeepSeek:

   • [...]
   • [...]

   ○ xAI Grok:

   • [...]
   • [...]

   ○ Z.AI GLM:

   • [...]
   • [...]

   ○ Alibaba Qwen:

   • [...]
   • [...]

   ○ Anthropic Claude (harness):

   • [...]
   • [...]

   ○ OpenAI Codex:

   • [...]
   • [...]

   ○ GitHub Copilot:

   • [...]
   • [...]

   In this output, when a quorum was not possible (see STEP 4), render the consensus rate line instead as just **CONSENSUS RATE**: *n/a* <disagreement/> (omitting the <c/>/<n/> fraction), as a single AI cannot form a consensus.

   In this output, remove the sections of those AIs which were not queried (excluded via --models/-m) or were not available. You MUST NOT output any further explanations yourself.