← ClaudeAtlas

advisorlisted

Pull a decorrelated second opinion from a DIFFERENT model on a LOAD-BEARING judgment: a design fork, a pre-commit plan, an irreversible or outward-facing call. Reversible calls just decide. `--panel` = blind critics on a prepared plan. NOT code-diff review (review), NOT mechanical work for Codex (handover), NOT sharpening a fuzzy idea (grill).
panda850819/verbs · ★ 5 · Code & Development · score 77
Install: claude install-skill panda850819/verbs
# Advisor ## Routing Boundary advisor pulls JUDGMENT in from a different model. It is the inbound half of the cross-runtime pair; `handover` is the outbound half (it sends mechanical build work OUT to Codex). Keep them distinct: - Second opinion on a judgment / design fork / decision → advisor (default). - Adversarial critique of a PREPARED plan → advisor `--panel`. - Review a code diff → `review`. - Hand unfinished mechanical build work to Codex to DO → `handover`. - Sharpen a fuzzy idea by self-interview → `grill`. advisor never writes code, never commits, never touches git. It asks another model and returns the outside view; the caller decides. Never auto-incorporate a finding. ## Step 0: Locate the seat (zero-config) Detect the current runtime only to guarantee that the outside opinion comes from a different model family: - `CLAUDECODE=1` in env → **Claude seat**; reach out through the verified Codex transport. - otherwise → **Codex seat**; reach out through the verified Claude transport. This seat check chooses direction only. Authentication, budgets, and general model-routing policy remain host concerns. Reach for the MOST INDEPENDENT model, not the nominally highest tier. From an Opus seat a different provider (Codex/GPT) usually yields more new signal than a same-family model does. ## Step 1: Select roles, then probe (fail loud, never silently self-review) Select the complete role set before probing: default uses the one seat-derived role from Step 2. A Cla