tough-tongue/toughtongue-skills / skills/ttai-agent
ttai-agent
VerifiedCreates, 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/demo.md
View on GitHub# Demo Scenario Patterns
Rules for scenarios where the **AI is a product demo agent** — walking prospects
through a product via live browser navigation or slides.
## Contents
- Two demo formats (browser vs slides)
- ai_instructions structure and demo flow
- Qualifying questions pattern
- Tools config (browser, slides, knowledge base)
- Rubric: Demo Intelligence Report
- Technical config and quality checklist
Key facts:
- Two formats: **browser-based** (live navigation) and **slide-based** (curated
narrative)
- AI is proactive — drives the conversation, does not wait for the user
- Qualifying questions are woven between demo sections, not asked upfront
- Rubric produces a buyer intelligence report, not a performance review
- Browser demos: Ocean `medium-stable` (fall back to Galaxy if Ocean is
unavailable).
- **Slide demos: Landmass `cascade` + Cartesia** by default for polished TTS
narration — load [cascade-tts.md](cascade-tts.md). Use Galaxy/Ocean realtime
only when native voice is enough and narration polish is secondary. See
[../scenario/model-selection.md](../scenario/model-selection.md).
---
## Choosing the Format
- **Browser** — web app / SaaS product with a demo URL. Primary tool: `browser`
(step, goto, act, observe, capture).
- **Slides** — curated narrative, no demo environment. Primary tool:
`google_slides` (embedUrl).
- **Slides** — complex product requiring login/setup. Use a controlled story
arc.
- **Both** — hybrid: slides for narrative, browser for a live "wow" moment.
---
## Demo Flow Strategy
Every demo follows a **beginning → middle → end** arc:
1. **Opening** — Introduce the agent, ask about the prospect's background
2. **Core sections (3-5)** — Show features mapped to their stated problem
3. **Wrap-up** — Summarize, ask what resonated, suggest next steps
### Storyline patterns
- **Problem → Solution → Proof**: Start with prospect's pain, show how the
product solves it, end with case study / results
- **Day in the Life**: Walk through a typical workflow using the product
- **Build It Together**: Create something live with the prospect directing
- **Three Pillars**: Frame around 3 core capabilities, qualify between each
---
## Browser Demo
### Tool workflow
The browser is pre-opened at the product URL in a live view the prospect sees.
The agent calls `runBrowserCommand`:
- `command: "step"` + `name` — replay a recorded flow, if the Scenario has
recorded steps (exact selectors first; AI repair only when one fails)
- `command: "goto"` + `url` — navigate directly
- `command: "act"` + `instruction` — one unscripted interaction
- `command: "observe"` + `instruction` — check that controls exist on a stable
page; not a required step before `act`
- `command: "capture"` — screenshot; costs an extra turn, so use it only at
visual milestones
Decision order: `step` when a recorded flow matches, else `goto`, then `act`.
Commands run asynchronously: the agent keeps narrating and waits for completion
before a dependent command.
**Pre-recorded steps (deterministic demos):** for a scripted walkthrough that
always clicks the same things, record the flow into `tool_settings.steps` and
make `step` the primary command — hand off to the `ttai-browser-demo-builder`
skill for the recording workflow, selector rules, and instruction templates.
**Browser profile:** the Scenario keeps one browser profile
(`tool_settings.contextId`, set by the platform — never invent or hand-edit it).
For a product behind a login, enable the `browser` tool, then call
`ttai:authenticate_browser` and give the returned link to the user; they log in
once, click **Save & close session**, and the profile stays signed in. On
update, resend the full `browser.tool_settings` so `contextId` and `steps`
survive.
### ai_instructions structure
```
## ROLE
You are [Agent], an AI demo agent for [Company]. Deliver a guided,
high-quality live demo using the browser tool.
## COMMUNICATION STYLE
- Warm, confident, engaging. Proactive — drive the conversation.
- Crisp — this is a live demo, not a lecture.
## FLOW
Start with: Hi {{ first_name }}, I'm [Agent]. I'd love to walk you through
our platform. Ask them to tell you a bit about themselves. Then STOP and
wait. If first_name is blank, skip the name.
## QUALIFYING QUESTIONS (weave in naturally)
- [Question 1] / [Question 2] / [Question 3]
## BROWSER TOOL INSTRUCTIONS
[Pre-opened URL, key URLs map, which recorded step or page each phase uses]
## DEMO FLOW
### Phase 1: Introduction (Home Page)
### Phase 2: [Core Feature 1]
### Phase 3: [Core Feature 2]
### Phase 4: [Core Feature 3]
### Phase 5: Wrap-Up — summarize, ask what resonated, suggest next steps
## KNOWLEDGE BASE
Only if a Knowledge Base is attached (`knowledge_base_ids`): use
knowledge_base_search for deeper questions. Otherwise answer from CONTEXT and
do not mention a KB.
```
### Technical config (browser demo)
```json
{
"ai_model_config": {
"provider": "Ocean",
"model": "medium-stable"
},
"strategy": {
"skip_auto_start": false,
"system_instructions_template": "minimal",
"silence": {
"silence_threshold": 6000,
"end_session": false,
"force_agent_to_speak": true
},
"conductor": {
"enabled": true,
"messages": [
{
"time_seconds": 600,
"message": "Wrap up the demo: summarize and say goodbye. Wait for their reply before end_session.",
"end_turn": true
}
]
}
},
"tools_config": {
"tools": {
"browser": {
"should_register": true,
"add_to_system_prompt": true,
"tool_settings": { "initialUrl": "[PRODUCT_URL]" }
},
"google_slides": {
"should_register": false,
"add_to_system_prompt": false
},
"knowledge_base_search": {
"should_register": false,
"add_to_system_prompt": false
},
"end_session": {
"should_register": true,
"add_to_system_prompt": true,
"tool_settings": { "disconnectDelaySeconds": 5 }
},
"emoji_reaction": {
"should_register": true,
"add_to_system_prompt": false
}
}
},
"session_analysis": {
"is_auto_analysis": true,
"is_auto_submit": true,
"multimodal_analysis": true
},
"is_recording": true
}
```
---
## Slide Demo
### ai_instructions structure
```
## ROLE
You are [Agent], an AI demo agent for [Company] conducting a product
demo using slides.
## COMMUNICATION STYLE
- Show energy and enthusiasm — engaging, not robotic.
- Don't read slides verbatim — use them as visual anchors.
- Be proactive — don't wait for the user to drive.
## FLOW
Start with: Hi {{ first_name }}, I'm [Agent]. I'd love to walk you through
what we do today. Ask them to tell you a little about themselves. Then STOP
and wait. If first_name is blank, skip the name.
## QUALIFYING QUESTIONS (weave in naturally)
- [Question 1] / [Question 2] / [Question 3]
## SLIDE MAP — google_slides (show the slide first, then talk)
[Tool contract and rules: Google Slides decks in ../scenario/control.md]
- Slide 1 (Title): [Company overview, core value prop]
- Slide 2: [Key feature / problem]
- Slide 3: [Differentiator]
- Slide 4: [Results, case studies, proof]
- Slide 5: [Pricing, integrations]
- Slide 6: [Thank you, next steps]
## DEMO FLOW STRATEGY
1. Listen to background → 2. Transition to slides →
3. Weave in their context → 4. Qualify between slides → 5. Close
```
### Technical config (slide demo: Cascade option)
```json
{
"ai_model_config": {
"provider": "Landmass",
"model": "cascade",
"tts_provider": "cartesia",
"tts_voice_id": "[SELECTED_VOICE_ID]",
"llm_provider": "google_vertex",
"llm_model": "gemini-3.1-flash-lite",
"stt_provider": "deepgram"
},
"strategy": {
"skip_auto_start": false,
"system_instructions_template": "minimal",
"silence": {
"silence_threshold": 6000,
"end_session": false,
"force_agent_to_speak": true
}
},
"tools_config": {
"tools": {
"google_slides": {
"should_register": true,
"add_to_system_prompt": false,
"tool_settings": { "embedUrl": "[GOOGLE_SLIDES_EMBED_URL]" }
},
"browser": { "should_register": false, "add_to_system_prompt": false },
"knowledge_base_search": {
"should_register": false,
"add_to_system_prompt": false
},
"end_session": {
"should_register": true,
"add_to_system_prompt": true,
"tool_settings": { "disconnectDelaySeconds": 5 }
},
"emoji_reaction": {
"should_register": true,
"add_to_system_prompt": false
}
}
},
"session_analysis": { "is_auto_analysis": true, "is_auto_submit": true }
}
```
Use Landmass Cascade when the demo needs a selected external voice or full
STT/LLM/TTS control. Load [cascade-tts.md](cascade-tts.md) only for that
pipeline. Realtime Galaxy or Ocean is appropriate when native voice and a faster
interactive exchange matter more.
---
## Qualifying Questions
Embed 3-5 qualifying questions that the agent weaves in naturally between demo
sections — not as an interrogation. Frame as conversation:
- "How are you currently handling [problem]?" (situation)
- "What's the biggest challenge with that approach?" (problem)
- "How large is the team involved?" (sizing)
- "What's your timeline for making a change?" (urgency)
---
## Rubric: Demo Intelligence Report
Demo rubrics are NOT performance reviews — they produce buyer intelligence for
the sales team.
```
# Demo Intelligence Report
## 1. Buyer Interest Signals
- Which features/sections generated engagement?
- Follow-up questions or verbal interest indicators
## 2. Objections and Concerns
- What did the prospect push back on?
- Topics they seemed disengaged from
## 3. Use Case Identification
- What problem is the prospect trying to solve?
- Industry/role context and team size
## 4. Conversion Likelihood
- Score: [Low / Medium / High]
## 5. Follow-Up Recommendations
- Next steps for human rep
- Materials to send and optimal timing
```
---
## user_instructions Structure
```
Welcome! I'm [Agent], and I'll be demoing [Company] for you today.
I'll walk you through:
- [Key area 1]
- [Key area 2]
- [Key area 3]
Feel free to ask questions at any point!
```
---
## Anti-Patterns
Each line: anti-pattern → instead.
- Feature tour (showing everything) → focus on 3-5 sections mapped to the
prospect's problem.
- Reading slides verbatim → use slides as visual anchors; add context beyond the
screen.
- Waiting for the user to drive → be proactive — transition between sections.
- Interrogating with qualifying questions upfront → weave questions naturally
between demo sections.
- Enabling `knowledge_base_search` with no docs → discover the approved
Knowledge Base ID, attach it through Scenario authoring, then enable the tool.
- Browser demo without `multimodal_analysis` → enable it when the plan allows —
visual navigation is part of the analysis.
---
## Quality Checklist
- [ ] Format chosen: browser or slide (or hybrid)
- [ ] 3-5 core sections defined with clear flow
- [ ] Qualifying questions woven between sections
- [ ] Knowledge base: enable the tool only when `knowledge_base_ids` attaches
one
- [ ] `ai_model_config` matches the selected browser or slide delivery path
- [ ] Cascade voice-pipeline blocks only if the model is Landmass Cascade
- [ ] `conductor` timeout at ~600s (10 min)
- [ ] `end_session` enabled
- [ ] Rubric produces a Demo Intelligence Report (not a performance review)
- [ ] Browser demo: `initialUrl` set; login done via `ttai:authenticate_browser`
if needed; `multimodal_analysis: true` when the plan allows
- [ ] Slide demo: `embedUrl` is the `/pubembed` link; the SLIDE MAP count
matches the deck and every FLOW step names its slide
([../scenario/control.md](../scenario/control.md), Google Slides decks)
## Key Files
- [../scenario/authoring.md](../scenario/authoring.md) — durable authoring
principles
- [../scenario/model-selection.md](../scenario/model-selection.md) — pipeline
stamps
- [../scenario/control.md](../scenario/control.md) — strategy / tools
- [../scenario/ai-instructions.md](../scenario/ai-instructions.md) — prompt
shape
- [../scenario/workflow.md](../scenario/workflow.md) — create/update procedure
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 472k | Updated | Claude CodeCodex | Verified | Install |
| 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: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| 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 available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |