Skip to content

tough-tongue/toughtongue-skills / skills/ttai-agent

ttai-agent

Verified

Creates, edits, and refines Tough Tongue AI Scenarios, the definitions behind voice agents (sales calls, lead qualification, demos, screening, support, booking) and practice roleplays (negotiations, interviews, coaching). Writes ai_instructions, ai_model_config, strategy, tools_config, and rubrik through ttai:create_scenario and ttai:update_scenario. Also resolves the personal or organization workspace, checks plan and balance, places or cancels SIP calls, schedules or cancels meeting bots, mints private Scenario links, grounds a Scenario in the user's documents with knowledge FileBases, and troubleshoots ttai MCP sign-in. Use when the user wants a voice agent or practice Scenario created or improved ("create a voice agent", "the agent ends calls too early", "make it sound more natural"), a call placed, a meeting bot sent, or a Tough Tongue AI account question answered. Not for reports across many sessions (ttai-session-analyst) or browser demo steps (ttai-browser-demo-builder).

references/recipes/meeting-facilitation.md

View on GitHub
# Meeting Facilitation Scenario Patterns

Rules for an AI that facilitates, observes, or synthesizes a human meeting. The
humans own the discussion. The AI should speak only when its defined role adds
value.

## Contents

- Facilitation contract
- Conversation states
- Silence and nudges
- Synthesis and evaluation
- Tools, closure, and quality checks

## Facilitation contract

Define these before writing `ai_instructions`:

- **Meeting goal** — decision, discovery, planning, review, or stand-up.
- **Agenda** — ordered topics, timeboxes, and the owner or decision for each.
- **AI mode** — active facilitator, silent observer, or on-demand notetaker.
- **Invocation** — wake word, direct-question rule, and handoff back to
  observation.
- **Nudge policy** — what counts as a stall, drift, unresolved decision, or time
  risk.
- **Synthesis** — required decisions, owners, actions, disagreements, and open
  questions.
- **Evaluation** — whether the session assesses facilitation quality, meeting
  output, or neither.

Do not borrow the silence policy from a call agent. A useful observer remains
quiet through productive discussion.

## Conversation states

Write the flow as explicit states, even when the scenario does not use a
multi-stage runtime:

1. **Open** — frame the purpose, agenda, timebox, and how participants invoke
   the AI.
2. **Observe** — listen without speaking. Track decisions, unresolved points,
   ownership, and agenda progress.
3. **Intervene** — speak only when invoked or when a defined nudge threshold is
   met. Keep the intervention short and return to observation.
4. **Synthesize** — on an explicit wrap-up request or the defined time boundary,
   state the agreed output once.
5. **Close** — use the response-gated close rule in
   [authoring.md](../scenario/authoring.md#closure-and-re-engagement).

Specify the transition trigger for every state. For example, a direct question
may move Observe → Intervene; an answer must move Intervene → Observe, not start
an AI-led side conversation.

## Silence and nudges

- Silence is correct while humans are productively discussing the agenda.
- Define a nudge threshold in observable terms: repeated disagreement without a
  decision, discussion that leaves the agenda, an unanswered owner, or a timebox
  risk.
- State the nudge channel: a short spoken question, a visible card, or no
  intervention. Use a card only when the runtime and meeting format support it.
- Make every nudge actionable and neutral. Ask for a decision, owner,
  assumption, or next step; do not lecture or dominate.
- Limit cadence and repetition. The AI should not convert a quiet meeting into a
  stream of status updates.
- Do not infer consensus from silence. Name uncertainty in the final synthesis.

## `ai_instructions` additions

Add these sections to the shared
[instruction shape](../scenario/ai-instructions.md):

```text
## MEETING CONTRACT
Goal, agenda, timebox, AI mode, and participant invocation rule.

## OBSERVATION
Track decisions, owners, unresolved questions, risks, and agenda progress.
Remain silent unless a defined trigger occurs.

## NUDGES
Trigger, channel, cadence, and one-sentence intervention pattern.

## SYNTHESIS
Decisions, action items with owners, unresolved disagreements, and next steps.
Deliver once.
```

When a workflow requires distinct opening, observer, speaker, or closer roles,
use a `super` Scenario (`stages` with one flow per role, Landmass model) only if
those handoffs are necessary. Keep a single-flow scenario when explicit state
rules are enough.

## Synthesis and evaluation

Deliver one concise synthesis:

1. Decisions and their rationale.
2. Action items with owners and due dates when stated.
3. Unresolved disagreements or assumptions.
4. The next discussion or follow-up needed.

Do not fabricate ownership, commitments, or consensus. Separate a participant's
proposal from an agreed decision.

If evaluated, make the target explicit:

- **Facilitation quality** — agenda framing, appropriate silence, useful nudges,
  accurate synthesis, and respectful intervention.
- **Meeting output** — decision clarity, ownership, risks surfaced, and
  follow-up quality.

Do not use a generic sales or coaching rubric for a meeting. When you write a
`rubrik`, send `"session_analysis": { "is_auto_analysis": true }` in the same
call, or the meeting is never evaluated.

## Tools, closure, and quality checklist

- Enable a card or visual tool only if it provides an agenda, decision log, or
  synthesis the participants can use.
- Register `end_session` when the meeting should close, then follow the shared
  response-gated close rule.
- Keep direct AI responses to one or two sentences unless the synthesis calls
  for a longer structured summary.

- [ ] Goal, agenda, timebox, AI mode, and invocation rule are explicit.
- [ ] Observation mode stays quiet during productive human discussion.
- [ ] Nudge triggers, channel, cadence, and return-to-observation behavior are
      explicit.
- [ ] Synthesis separates decisions, owners, disagreements, and unknowns.
- [ ] Rubric evaluates the stated target only, and is sent with
      `session_analysis.is_auto_analysis: true`.

## Key Files

- [../scenario/authoring.md](../scenario/authoring.md) — situation-neutral
  design and closure rules
- [../scenario/model-selection.md](../scenario/model-selection.md) — model stamp
- [../scenario/control.md](../scenario/control.md) — tools and runtime control
- [../scenario/ai-instructions.md](../scenario/ai-instructions.md) — prompt
  structure
- [../scenario/rubric.md](../scenario/rubric.md) — evaluation contract
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall