tmb_roundtable trigger ergonomics — Human shouldn't have to say the literal word "roundtable"

[ZaxShen]

Symptom

Observed in PR #64 Mode A dogfood (Zax, 2026-04-25): roundtable works end-to-end when you literally type "roundtable" in the ask, but it's the only reliable trigger phrase. Natural ways of asking for multi-consultant deliberation (`"get cto and architect's read on this"`, `"convene cto + ceo"`, `"what would pm and designer say about this"`) don't fire the skill.

That's a discoverability problem. Most users won't reach for the word "roundtable" first. They'll ask in the way they naturally would and get a single-consultant response or a bro-only response, missing the value of the skill.

Current behavior

`skills/tmb_roundtable/SKILL.md` "When invoked" presumably (re-check) names "roundtable" as the trigger keyword. Bro's prompt routing in CLAUDE.md does NOT have a routing rule for "multi-consultant" intent.

Proposed triggers

Bro should invoke `tmb_roundtable` when ANY of:

1. Explicit keyword — Human says "roundtable", "convene", "deliberation", "multi-agent discussion", "cross-functional take".
2. Multi-consultant ask — Human names ≥2 consultants in one ask: "get cto AND architect's read", "discuss with X and Y", "ceo plus pm view".
3. Cross-domain detection (auto) — Bro analyzes the ask and notices it spans ≥2 domain experts (technical strategy → cto, product scope → ceo/pm, security → security-reviewer, etc.). Bro proposes: "This question spans . Want me to convene and roundtable, or just spawn one consultant?"

(3) is the highest-leverage but trickiest — it requires bro to model "what consultant covers what domain". The simplest implementation: bro keeps a domain → consultant map in CLAUDE.md or a tiny skill, scans the ask for keywords, suggests roundtable on overlap.

Required changes

• `skills/tmb_roundtable/SKILL.md` — broaden "When invoked" with the three trigger classes above; include 2-3 verbatim trigger examples for each.
• CLAUDE.md routing table — add rows for the new trigger phrases. The "Get architect's / cto's / pm's opinion on X" row should also have a sibling: "Multi-consultant ask (≥2 named) → roundtable".
• (Optional) — author a domain → consultant map either in CLAUDE.md (small) or a new tiny skill `tmb_agent-creator-routing`. Bro consults this when classifying an ask.

Acceptance

• Asking `@bro get cto and architect's read on whether to use OAuth` triggers roundtable, not single-consultant
• Asking `@bro convene a discussion on the data retention policy` triggers roundtable
• Asking `@bro should we adopt OAuth?` (cross-domain — security + cto + maybe legal) prompts bro to propose roundtable, with the option to decline and pick one consultant
• Pure single-consultant asks (`@bro get cto opinion on X`) still fire single-consultant flow, not roundtable

Priority

Medium. Doesn't break the chain; reduces discoverability of an existing skill. Best fit for a roundtable-specific PR after PR #64 latency work merges.

Related

• PR 🚧 chore: latency reduction — proposals on review #64 (latency reduction)
• Multi-consultant voting protocol (architect/cto/ceo as advisors, not deciders) #57 (multi-consultant voting protocol — companion design issue; voting is the inside-roundtable mechanic, this issue is about how to OPEN one)

[ZaxShen]

Issue tracking moved to Linear. This issue continues as TRU-12: https://linear.app/trustmybot/issue/TRU-12

GitHub issues are no longer the active tracker for this project — see the Linear workspace for current state. Closed-as-migrated, not closed-as-resolved.

[ZaxShen]

Disregard the prior 'closed-as-migrated' comment. GitHub remains the public issue tracker for this project — file new issues, PRs, and discussion here as normal. TMB team mirrors this issue internally to Linear (TRU-12) for our own workflow, but that mirror is auto-synced from GitHub — contributors don't need a Linear account.

[ZaxShen]

Closing — focusing on v0.8.x. Planning lives on GitLab milestones (v0.8.0 / v0.8.1). Will re-mirror active items to GH once tri-source schema sync (#48) lands.