team

Child mode. Check for child mode as claudish:claudish-usage → "Inside a claudish session (child mode)" describes. In child mode, claudish allows no panel below this session. Say so in one line, then do the task yourself and report one verdict, labelled as yours and not a panel vote. If claudish has no tools at all, that is not child mode: report it as a failure.

Step 1: Setup

Step 1a — Load alias table: Follow the claudish:claudish-usage skill → "Model Alias Resolution" procedure to build ALIAS_TABLE from the live catalog (list_models) + .claude/multimodel-team.json customAliases.

Step 1b — Parse args: Parse: defaultModels, contextPreferences, agentPreferences, defaultThreshold from prefs. Parse command args: task, --models, --threshold, --no-memory. If no task: ask the user.

Step 1c — Resolve models (stop at first match, resolve each name via ALIAS_TABLE):

1. --models flag provided → resolve each via ALIAS_TABLE
2. contextPreferences keyword matching task → resolve each via ALIAS_TABLE
3. defaultModels from prefs → resolve each via ALIAS_TABLE, announce "Using saved models: {list}"
4. None matched → compose a team from list_models: take internal plus 2–4 current models from different providers, picked for the task type (see table below)
5. Still nothing → AskUserQuestion listing the models list_models currently returns

Resolve threshold: unset/"majority" → 50%, "supermajority" → 67%, "unanimous" → 100%

Resolve agent (always — it applies to the whole panel): match task keywords to the context detection table, check agentPreferences[context] first, else table default, else dev:researcher. Announce: "Agent: {RESOLVED_AGENT}"

It is passed to team as the agent argument and applies to EVERY model in the run, native and external alike; there is no per-model form. For a blind panel that is the correct shape — every voter reviews by the same method, so a vote difference reflects the model rather than the prompt.

Session directory (the team tool's session path): Bash: SESSION_DIR="$(pwd)/ai-docs/sessions/team-$(date +%Y%m%d-%H%M%S)" && mkdir -p "$SESSION_DIR" && echo "$SESSION_DIR"

Build vote prompt using the template below with {TASK} substituted, then write it to ${SESSION_DIR}/input.md. Step 2 passes that path, not the text — a vote prompt is 100+ lines, and an inline input echoes every line of it verbatim in the user's terminal, burying the model list, the agent and the shape check.

Unless --no-memory, save resolved models to defaultModels in the preferences file.