← ClaudeAtlas

bug-councillisted

Last resort for ONE stubborn bug the ordinary attempts already failed on — a blind, evidence-driven council of Codex, Claude, Kimi, or available native subagents that falsifies competing hypotheses, implements the smallest causal fix, and verifies it independently. EXPLICIT INVOCATION ONLY - deliberately slow and token-expensive, so never auto-trigger it, never chain into it from another skill, and never run it on bugs a hunt report merely listed. Use only when the user names it (/bug-council, "convene the council", "bring in the bug council") or explicitly asks for the heaviest possible investigation of one specific bug. Otherwise, at most SUGGEST it - when a fix attempt has already failed, a bug is intermittent or unreproducible, the cause is genuinely disputed, or a regression has no obvious culprit - and wait for the user to say yes. For ordinary bugs use tdd --debug; to find unknown bugs use hunt.
supermodo/skills · ★ 0 · AI & Automation · score 70
Install: claude install-skill supermodo/skills
# Bug Council > **Requires:** the sibling `protocols` skill (shared protocol masters); uses `skills.config.json` when present. Missing protocols → tell the user to install the full supermodo package. > **Project rules.** Read `.supermodo/rules/bug-council.md` if present, plus any > `.supermodo/rules/INDEX.md` rows naming `bug-council` — that file IS this > project's council process (seats, models, mode, rounds) and replaces the > defaults below wherever they overlap. Contract: > `../protocols/references/rules.md`. Never in that file, so never switchable > off: investigations stay blind and are never shown to each other, one > designated implementer, a fresh verifier that did not implement, passing > tests never mean solved, explicit invocation only. Run a structured, evidence-driven bug hunt. ## Invocation policy — read before starting This skill is the toolkit's most expensive act: several independent agent seats, blind investigation, falsification rounds, experiments, and a separate verification pass. It buys certainty about ONE bug at a price no routine bug is worth. For the toughest stains only. - **Explicit invocation only.** Run it when the user asked for it by name. Never auto-trigger, never chain into it from `hunt`, `flow`, `tdd`, or `work`, and never fan it out over a list of findings — a hunt report with twelve findings is twelve `tdd --debug` jobs, not twelve councils. - **Suggest, don't start.** When the signals below appear, offer it in one line an