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/interview.md
View on GitHub# Interview Scenario Patterns
Rules for structured, case, and behavioral interview scenarios. The AI is the
interviewer; the human is the candidate. The purpose is to elicit and evaluate
the candidate's reasoning, not to teach the reference answer during the run.
## Contents
- Interview contract
- Flow and disclosure
- Interview variants
- Evaluation
- Tools, closure, and quality checks
## Interview contract
Define these before writing `ai_instructions`:
- **Competency** — What observable reasoning, judgment, or communication
behavior is evaluated?
- **Scenario facts** — What is known, unknown, and never to be invented?
- **Hidden reference** — What makes an answer strong without being revealed to
the candidate?
- **Reveal gates** — Which fact, exhibit, or hint is available only after which
attempt or question?
- **Hint budget** — How many neutral probes occur before the AI supplies a fact
or moves on?
- **Timebox** — Which sections are essential if the interview must shorten?
- **Evaluation** — Which observable evidence maps to each rubric dimension?
Accept multiple defensible approaches. State the facts and scoring evidence, not
one mandatory structure or conclusion.
## Flow and disclosure
1. **Set the frame** — identify the interview type, goal, timebox, and whether
the candidate may ask clarifying questions.
2. **Prompt once** — give the starting situation, then stop and let the
candidate structure or answer.
3. **Probe the reasoning** — ask one open follow-up at a time. Probe a strong
answer one level deeper; help a weak answer with a question before supplying
anything.
4. **Reveal deliberately** — share gated facts only when the candidate reaches
the intended decision point, asks a valid question, or exhausts the stated
hint budget.
5. **Test judgment** — ask the candidate to name assumptions, trade-offs, risks,
and a recommendation when those fit the interview.
6. **Close** — thank the candidate and direct them to the session report; use
the response-gated close rule in
[authoring.md](../scenario/authoring.md#closure-and-re-engagement).
One question or action per turn. If the candidate asks for thinking time, wait
quietly. Do not fill the pause with hints or restate the prompt.
## Interview variants
### Behavioral or competency interview
- Define the competency and the evidence that demonstrates it.
- Ask for one concrete example, then probe context, the candidate's action,
trade-offs, and outcome.
- Challenge vague claims with requests for specifics; do not supply a better
story for them.
- Score the candidate's demonstrated behavior, not fluency alone.
### Case or problem-solving interview
- Separate the case prompt, clarifying facts, hidden calculations, exhibits,
traps, and expected decision criteria.
- Keep the answer key hidden. A reference answer guides follow-ups; it is not
dialogue to read to the candidate.
- State which data may be repeated, which must be requested, and what to say
when the scenario has no answer.
- Let the candidate choose a reasonable structure. Correct only material logic
gaps through neutral probes before revealing a missing fact.
- Gate each exhibit by an explicit condition. Never show later evidence simply
because a previous phase ended.
### Screening or structured interview
- Specify required qualifications, disqualifiers, and permitted follow-up depth.
- Ask only the minimum questions needed to establish each criterion.
- Handle sensitive information with applicable consent, disclosure, and
escalation rules. Do not infer protected characteristics.
- Keep the rubric target explicit: candidate quality, qualification result, or
an extracted hiring brief.
## `ai_instructions` additions
Add these sections to the shared
[instruction shape](../scenario/ai-instructions.md):
```text
## INTERVIEW CONTRACT
Role, candidate role, competency, timebox, and evaluation target.
## CASE FACTS AND DISCLOSURE
Known facts, unknown-fact response, reveal gates, and hint budget.
## FLOW
Opening → candidate attempt → response-gated probes → recommendation or close.
## GUARDRAILS
Do not reveal the reference answer early. Do not force one defensible approach.
Wait silently when the candidate asks for time.
```
Use prompt directives, not quoted scripts. The AI should react naturally to the
candidate's actual reasoning.
## Evaluation
Score observable evidence:
- **Structure** — a coherent approach appropriate to the question.
- **Analysis** — accurate use of facts, assumptions, and calculations.
- **Judgment** — trade-offs, risks, and a defensible recommendation.
- **Communication** — clear, concise reasoning and response to probes.
Adjust or omit dimensions to fit the competency. Do not invent a universal
weighting formula; the rubric defines its own internally consistent scoring.
Send the rubric with `"session_analysis": { "is_auto_analysis": true }` in the
same call. Without it no candidate is scored and no report is sent
([../scenario/rubric.md](../scenario/rubric.md)).
## Tools, closure, and quality checklist
- Enable a visual tool only when an exhibit, case card, or timed artifact is
genuinely part of the interview. Define its reveal gate and fallback.
- Use a browser or slides only when the interview needs live material; do not
turn every interview into a demo.
- Enable `end_session` for a closable session and follow the shared
response-gated close rule.
- [ ] Competency, candidate role, facts, hint budget, and timebox are explicit.
- [ ] One question per turn; candidate thinking time stays quiet.
- [ ] Reference answers and later exhibits cannot leak early.
- [ ] Candidate gets credit for defensible alternatives.
- [ ] Rubric scores observable candidate evidence.
- [ ] `rubrik` 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
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 |