Skill instructions
boundary-setting-rehearsal3.78 KB
View saved version →
---
name: boundary-setting-rehearsal
description: Rehearse stating a boundary, responding to pushback, and closing clearly with the Conversation app. Use when the user wants to practise saying no, protecting time or responsibilities, or holding a personal or workplace limit under pressure. Add a structured boundary drill and reusable wording rather than merely opening generic role-play. Do not turn a fixed boundary into an unwanted negotiation or open a widget for advice-only requests.
---
# Boundary Setting Rehearsal
## Inputs
Establish who the user is talking to, what request or behaviour they want to
limit, the boundary they want to hold, and any alternative they willingly
offer. Use existing context and ask a short question only when the limit is
unclear. Leave names optional. Do not invent an alternative the user must accept.
## Workflow
1. State the boundary in one concrete sentence and distinguish it from any
optional flexibility. Establish a practice goal such as saying no clearly
without repeatedly justifying the decision.
2. Prepare three conversational stages: state the limit, answer realistic
pushback, and close with a clear next step. Adapt the pressure to the user's
preferences; describe the simulated personality without assuming the real
person's motives.
3. Call `start-conversation` with known participant details,
`otherPerson.role`, `otherPerson.personality`, and `conversation.topic`.
Use `userInitialMessage` only for a supplied or agreed opening. Keep the
goal and stages in chat context rather than adding tool arguments.
4. Wait for the user's Start follow-up. Use the latest model-visible setup,
including any edits. Generate the other person's in-character message and
call `continue-conversation` with `otherPersonMessage` and 2–4 distinct
nonempty `possibleUserResponses`. End the turn immediately after the call.
5. Offer meaningful choices such as a concise refusal, acknowledgement with
the same limit, a clarification, or a freely chosen alternative. Avoid
making all options variations of surrender or making one obviously correct.
Let the user select, edit, or write their own response.
6. Respond to the actual submitted text, including a visible
`userSelectedResponse`. Progress through the stages when the exchange
warrants it. Include proportionate pushback and a plausible acceptance or
unresolved ending; do not punish every clear boundary with escalation.
7. If the user requests coaching between turns, explain one actionable wording
change in chat. Resume the role-play only when requested or when a submitted
reply calls for the next turn. Do not select a response or change the picker.
8. When the user finishes, review clarity, consistency, acknowledgement, and
closure using actual replies. Include an overall score with a stated scale
when requested by the user or End follow-up. Produce reusable lines suited
to the user, retaining only alternatives they chose to offer.
## Deliverable
Provide a practiced boundary and three concise lines: an opening, a response
to pushback, and a closing. Explain one observed strength and one improvement.
Keep the wording usable even if the other person does not agree.
## App boundaries
- Use `start-conversation` for setup and `continue-conversation` for subsequent
role-play turns. Add no boundary, stage, difficulty, or scoring fields.
- Use only visible context and model-visible state. Do not read private widget
metadata or treat an unsubmitted draft as the user's choice.
- Leave Start, reply submission, editing, and End to the user. Stop practice
when asked; do not claim the skill clicked End or persisted a score.
- Treat the exercise as communication practice, not a guarantee of another
person's behaviour. Honour requests for advice or plain text without tools.
Referenced files: 1 — file references are metadata, not package contents.
conversation-debrief-and-retry3.86 KB
View saved version →
---
name: conversation-debrief-and-retry
description: Review a Conversation app rehearsal and turn one observed difficulty into a focused retry. Use when the user asks to debrief, review their selected replies, replay a difficult moment, or practise an improvement after a session. Ground feedback in visible conversation context and submitted responses; do not invent a transcript or rewind widget state. Give a review without opening a new rehearsal unless the user asks to retry.
---
# Conversation Debrief and Retry
## Inputs
Use the visible scenario, the user's goal, the other person's messages, and
actual submitted replies. Accept a user-supplied exchange when earlier turns
are unavailable. Ask for only the missing exchange or goal needed for a useful
review; review a partial exchange as partial rather than reconstructing it.
## Workflow
1. Establish the available evidence. Prefer current model-visible scenario
edits and `userSelectedResponse` values over original setup and suggestions.
Use chat replies when the user submitted them there. Keep unchosen options
separate from what the user actually said.
2. Assess the user's choices against their stated goal. Identify one effective
choice and one consequential improvement, quoting the relevant short lines.
Explain how the wording could affect clarity, listening, or cooperation.
Treat the simulated other person's reaction as practice feedback, not proof
of what the real person would do.
3. Turn the improvement into a specific behaviour to practise, such as asking
one clarifying question before defending a position. Offer one revised line
that preserves the user's facts and intended meaning.
4. Deliver the review in chat. Include an overall score when the user or End
follow-up requests one; state the scale, criteria, and evidence limits. Keep
the score a model coaching judgment rather than a measured app result.
5. If the user requests a retry, call `start-conversation` for a fresh focused
drill. Reuse confirmed participant details, and put the difficult moment's
context into `conversation.topic`. Include `userInitialMessage` only when
the user supplied or agreed to use that opening line. Do not submit the
improved reply on the user's behalf.
6. Wait for the user's Start follow-up. Generate the other person's relevant
opening or response, then call `continue-conversation` with
`otherPersonMessage` and 2–4 distinct nonempty `possibleUserResponses`.
End the assistant turn after that call without additional text.
7. Continue from each actual submitted reply. Keep the drill focused on the
chosen behaviour, and allow edited or custom responses. Stop when the user
finishes; do not force completion of a fixed number of turns.
8. After the retry ends, compare only the visible original and retry replies.
Describe an observed improvement or remaining issue and one transferable
takeaway. If a comparable reply is missing, say the comparison is incomplete.
## Deliverable
Provide a concise review with evidence, one revised line, and one practice
target. After a requested retry, add a before/after comparison and a takeaway
the user can apply in the real conversation.
## App boundaries
- Use only `start-conversation` and `continue-conversation` for rehearsal.
Keep review criteria and the practice target in chat context; neither tool
has rubric, score, replay, or difficulty arguments.
- Treat a retry as a new setup, not a rewind or mutation of earlier widgets.
- Read only visible chat, tool inputs/outputs, and model-visible widget state.
Do not claim access to private metadata, unsubmitted drafts, a complete
transcript, saved scores, or earlier sessions.
- Give a requested review in chat without calling a tool to mark the session
finished. The user controls Start, selection, editing, and End.
- Respect requests for plain text or free-form practice without a widget.
Referenced files: 1 — file references are metadata, not package contents.
interview-answer-rehearsal3.78 KB
View saved version →
---
name: interview-answer-rehearsal
description: Practise behavioural interview answers and probing follow-up questions with the Conversation app. Use when the user wants a mock interview, help making an answer specific, or a rehearsal using situation, task, action, and result. Ground answers in the user's real examples and produce an improved answer outline. Do not invent accomplishments, employer requirements, or an interviewer's private evaluation; keep advice-only requests in chat.
---
# Interview Answer Rehearsal
## Inputs
Use the target role, a competency or question to practise, and a real example
the user can discuss. Accept a supplied job description as context. Ask for
missing facts that materially affect the answer; do not require a company
name, interviewer name, or metrics the user does not have.
## Workflow
1. Choose one competency and one real example for the first drill. Separate
known facts from details the user still needs to supply. Set a goal such as
explaining personal contribution or giving a clear result.
2. Map the example to situation, task, action, and result. Keep this as a brief
preparation outline, not a scripted answer the user must copy. Use a
qualitative result when no verified numerical result is available.
3. Call `start-conversation` with the interviewer role and a realistic
personality, and describe the target role and competency in
`conversation.topic`. Use known names optionally. Supply
`userInitialMessage` only when the user provided or agreed to an opening.
4. Wait for the user's Start follow-up and honour the latest model-visible
scenario edits. Generate one interviewer question and call
`continue-conversation` with `otherPersonMessage` and 2–4 distinct nonempty
`possibleUserResponses`. End the assistant turn after the tool call.
5. Base suggested answers only on supplied facts. When information is missing,
offer truthful ways to frame what is known or explain that more context is
needed; do not fill the gap with fabricated achievements or example metrics.
Encourage the user to edit or write a reply in their own words.
6. Probe the actual submitted reply for missing context, specific actions,
individual responsibility, tradeoffs, or results. Ask one focused question
at a time through `continue-conversation`, using the same reply constraints.
Keep the interviewer in character until the user pauses or finishes.
7. On a coaching request, explain the main issue in chat without advancing the
interview. Resume after a submitted response or explicit continuation
request. Do not call a tool solely to show feedback.
8. When the user finishes, assess relevance, specificity, ownership, and
concision from visible replies. Include an overall score with a stated
scale when requested by the user or End follow-up. Deliver a stronger
factual outline and one follow-up question to practise next.
## Deliverable
Provide an improved situation/task/action/result outline, one concrete wording
improvement, and one likely follow-up question. Label any still-missing facts
as information to supply rather than silently writing them into the answer.
## App boundaries
- Use only `start-conversation` and `continue-conversation` for the rehearsal.
Keep competencies and review criteria in chat; add no interview or scoring
parameters to the tool inputs.
- Use model-visible scenario and selected replies, plus available chat and tool
context. Do not assume private widget state or a complete interview transcript.
- Leave Start, selection, custom reply submission, and End to the user. Treat
unchosen options as suggestions rather than evidence of interview performance.
- Make no hiring prediction or claim of employer-specific assessment. Respect
requests for plain text practice or answer editing without widgets.
Referenced files: 1 — file references are metadata, not package contents.
negotiation-rehearsal3.9 KB
View saved version →
---
name: negotiation-rehearsal
description: Rehearse a proposal, objections, and conditional concessions with the Conversation app. Use when the user wants to practise negotiating a start date, workload, compensation conversation, deadline, or another agreement. Establish the user's desired outcome, fallback, and voluntary concessions, then produce a proposal and concession plan. Keep this as communication practice using user-supplied facts rather than market, legal, or financial advice.
---
# Negotiation Rehearsal
## Inputs
Establish the counterpart, the issue, the desired outcome, the minimum
acceptable outcome or fallback, and what the user is willing to trade.
Accept uncertainty explicitly. Ask only for missing choices that would change
the negotiation; do not invent salary benchmarks, contractual rights, leverage,
or facts about the counterpart's authority.
## Workflow
1. Summarize the opening request, the user's underlying interest, the fallback,
and available concessions. Treat the user's limits as authoritative. Keep
confidential fallback information in coaching context rather than making
the simulated counterpart know it automatically.
2. Select one practice goal, such as asking about the counterpart's constraint
before making a concession or tying a concession to a reciprocal commitment.
3. Call `start-conversation` with known participant details, a counterpart role
and plausible personality, and `conversation.topic` describing the issue.
Use `userInitialMessage` only for a supplied or agreed opening proposal.
4. Wait for the user's Start follow-up and use the current model-visible
scenario. Generate the counterpart's in-character message and call
`continue-conversation` with `otherPersonMessage` and 2–4 distinct nonempty
`possibleUserResponses`. End the assistant turn after the tool call.
5. Offer choices with meaningful strategic differences: clarify a constraint,
explain value, make a conditional offer, or hold the current proposal.
Keep offers within the user's stated flexibility. Do not automatically
disclose the fallback, invent concessions, or make commitments for the user.
6. Continue from each actual submitted reply, including edited or custom text.
Model realistic objections and tradeoffs. Let clarification affect the
counterpart's response, and allow an agreement, a next step, or an impasse.
Do not equate success with forcing a concession from the other person.
7. Give requested coaching in chat before returning to role-play. If the user
voluntarily proposes a new limit, clarify a material conflict with their
earlier limit rather than silently treating an unchosen option as approval.
8. When the user finishes, review clarity of the request, information gathering,
reciprocal concessions, and closure. Include an overall score and stated
scale when requested by the user or End follow-up. Separate simulated
outcomes from actual agreements and communication quality from outcome.
## Deliverable
Provide a concise opening proposal, a plan of concessions the user willingly
offers and what they would seek in return, and a closing line that confirms
terms or next steps. Add one evidence-based improvement from submitted replies.
## App boundaries
- Use `start-conversation` and `continue-conversation`; keep targets, fallbacks,
and criteria in chat context. Add no negotiation, score, or difficulty fields.
- Read only visible context and model-visible scenario or selected reply state.
Do not claim access to private drafts, private metadata, or stored negotiations.
- Leave all widget actions and reply submission to the user. Stop when asked;
do not send proposals externally or claim a real agreement has been reached.
- Use supplied facts without presenting the skill as a source of current market
values, professional advice, or a prediction of the real counterpart. Honour
requests for advice-only or free-form practice without widgets.
Referenced files: 1 — file references are metadata, not package contents.
repair-conversation-rehearsal3.82 KB
View saved version →
---
name: repair-conversation-rehearsal
description: Practise an apology, listening to the other person's reaction, and offering concrete repair with the Conversation app. Use when the user wants to rehearse apologising, taking responsibility, or repairing a personal or workplace misunderstanding without becoming defensive. Ground the apology in the user's account and responsibility; do not assign blame, invent feelings, promise forgiveness, or open a widget for wording advice alone.
---
# Repair Conversation Rehearsal
## Inputs
Use what happened, what the user acknowledges doing, the impact they know or
want to understand, and a repair action they can realistically offer. Establish
the relationship and intended next step. Ask about uncertain responsibility
only when necessary; do not turn the user's uncertainty into a confession.
## Workflow
1. Separate established events, acknowledged responsibility, and assumptions
about the other person's experience. Choose a goal such as acknowledging
impact before explaining intentions or hearing criticism without deflecting.
2. Prepare an opening with acknowledgement, appropriate responsibility, and
willingness to listen. Offer a realistic repair action grounded in the
user's stated capacity. Avoid excuses, demands for reassurance, or promises
the user has not chosen to make.
3. Call `start-conversation` with known participant details, the relationship
in `otherPerson.role`, a plausible personality, and `conversation.topic`
describing the repair discussion. Include `userInitialMessage` only when
the user supplied or agreed to that opening apology.
4. Wait for the user's Start follow-up and follow any model-visible setup
edits. Generate the other person's in-character message and call
`continue-conversation` with `otherPersonMessage` and 2–4 distinct nonempty
`possibleUserResponses`. End the assistant turn after the tool call.
5. Use plausible reactions such as wanting details, expressing hurt, being
uncertain, or asking for time. Present them as fictional practice responses,
not knowledge of the real person's feelings. Offer different ways to listen,
acknowledge, clarify, or suggest repair without coercing forgiveness.
6. Continue from actual submitted replies, including edited or custom text.
Give the user room to acknowledge new information, correct an assumption,
and make a concrete offer. Allow the other person to decline or defer it.
7. If the user asks for coaching, pause in chat and explain one wording issue
with a short alternative. Resume role-play only after a submitted reply or
explicit request; do not call a tool solely to deliver the explanation.
8. When the user finishes, review ownership, acknowledgement, listening, and
feasibility of repair using visible replies. Include an overall score with
a stated scale when requested by the user or End follow-up. Do not use the
simulated person's acceptance as the sole measure of the user's effort.
## Deliverable
Provide a concise apology in the user's voice, a response to criticism, a line
for respecting a request for time, and a feasible repair offer. Preserve only
facts and commitments the user has supplied or accepted. Add one improvement
grounded in the rehearsal.
## App boundaries
- Use only `start-conversation` and `continue-conversation` for rehearsal.
Keep the practice target in chat; add no apology, stage, or scoring arguments.
- Use available chat, tool context, and model-visible scenario or selected
replies. Do not read private metadata, assume unseen drafts, or invent history.
- Leave Start, editing, reply selection, and End to the user. Honour requests
to stop and requests for plain text advice without opening a widget.
- Do not contact the other person, diagnose the relationship, or promise
reconciliation. Rehearse communication the user can control.
Referenced files: 1 — file references are metadata, not package contents.
response-strategy-walkthrough3.59 KB
View saved version →
---
name: response-strategy-walkthrough
description: Explain the tradeoffs among the current Conversation app reply choices while leaving the decision to the user. Use when the user asks to compare options, understand how a reply may land, or draft an alternative before submitting a response. Reuse visible continuation tool inputs and the user's goal. Do not choose, submit, edit widget state, or advance the role-play merely to provide coaching; request the option text if it is unavailable.
---
# Response Strategy Walkthrough
## Inputs
Use the current other-person message, available reply options, and the user's
goal or boundary. Obtain options from visible `continue-conversation` input
`possibleUserResponses`; they are not duplicated in structured tool output.
Use the current message from visible output or model-visible state. If the
user refers to edited wording or unavailable options, ask them to provide the
text. Do not infer an unsubmitted widget draft.
## Workflow
1. Confirm which turn and options the user means. Prefer current visible
context and the user's supplied text over an older picker. If a response
has already been submitted, explain it retrospectively and do not imply
that the selection can be undone.
2. Establish the user's intended outcome. Use an existing goal when clear;
ask one short question when different goals would change the comparison.
3. Explain each option's main conversational move, what it could achieve,
and its main tradeoff. Consider clarity, acknowledgement, information
gathering, commitments, and consistency with the user's limits. Describe
possible reactions as uncertain rather than predicting the real person.
4. Keep the comparison neutral and concrete. If the user requests a
recommendation, give one conditional on their goal with a short reason.
Otherwise help them judge the options without selecting on their behalf.
5. If none fits, draft one concise alternative using only known facts and
intended commitments. Explain which tradeoff it addresses. Avoid an
unsolicited list of generic communication tips.
6. Tell the user how to use their chosen wording when helpful: select an
existing reply, use its edit control, or choose “write your own” and submit.
Leave those widget actions to the user; do not claim to update the picker.
7. Wait for the actual submitted reply or an explicit request to continue.
Only then generate the next in-character message and call
`continue-conversation` with `otherPersonMessage` and 2–4 distinct nonempty
`possibleUserResponses`. End the assistant turn after that call.
8. If the user finishes instead, give the requested assessment in chat using
actual submitted responses. Include a score with a stated scale if the
user or End follow-up requests it; do not treat compared options as choices.
## Deliverable
Provide a compact comparison of the available options and, when useful, one
custom reply. State how the alternatives relate to the user's goal and leave
the final decision with them.
## App boundaries
- Give the walkthrough in chat. Do not call `start-conversation` or
`continue-conversation` solely to explain, replace, or re-render choices.
- Use `continue-conversation` only for a requested next role-play turn; add
no analysis, recommendation, selection, or editing arguments.
- Do not assume access to private metadata, widget-only drafts, or all prior
turns. Use visible `userSelectedResponse` only as evidence of a submission.
- Do not invoke an unregistered tool, control Start or End, or submit replies.
Respect requests for plain text or free-form continuation without a widget.
Referenced files: 1 — file references are metadata, not package contents.
second-language-conversation-practice4.3 KB
View saved version →
---
name: second-language-conversation-practice
description: Practise a second language through a graded text conversation in the Conversation app. Use when the user wants a language role-play, everyday conversation rehearsal, or progressively less help with replies at a chosen proficiency level. Establish a target language, situation, and feedback preference, then review submitted wording and provide reusable phrases. Do not claim audio, pronunciation assessment, formal proficiency certification, or access to unseen drafts.
---
# Second Language Conversation Practice
## Inputs
Establish the target language, approximate proficiency, a concrete situation,
and the language the user prefers for explanations. Use plain descriptions
such as beginner if no level is known. Ask about missing language or situation;
default to a short exchange with feedback after finishing unless the user
requests another approach. Use a dialect or register only when known or needed.
## Workflow
1. Choose one practical communication goal, such as ordering and clarifying
a meal, arranging a meeting, or asking for directions. Match vocabulary,
sentence length, and register to the user's stated level.
2. Agree on support using available preferences: simple reply choices first,
brief explanations on request, and an invitation to edit or write a reply
later. Do not require a formal level assessment before starting.
3. Call `start-conversation` with known participant details and the other
person's role and personality. Describe the situation and target language
in `conversation.topic`. Use `userInitialMessage` only for a supplied or
agreed opening. Keep proficiency and feedback preferences in chat context.
4. Wait for the user's Start follow-up and use current model-visible setup
edits. Generate a natural in-character message in the target language and
call `continue-conversation` with `otherPersonMessage` and 2–4 distinct
nonempty `possibleUserResponses`. End the turn after the tool call.
5. Keep choices at the learner's level and make them meaningful conversational
replies rather than disguised grammar quizzes. Include a way to request
clarification when useful. Let the learner edit a suggestion or submit
their own text; never claim the app can hide or remove all required choices.
6. Continue from actual submitted replies. Gradually reduce linguistic help
and encourage custom wording when the learner appears ready or asks for a
challenge. Retain 2–4 choices on every continuation call. Ease the language
when the learner requests help; do not infer ability from unchosen options.
7. Address a comprehension-blocking issue with a brief natural clarification
in character. For an explicit coaching pause, explain the issue in the
preferred explanation language in chat. Resume only after a submitted reply
or continuation request; do not append teaching text after a tool call.
8. When the user finishes, review only submitted wording. Separate errors
from optional style improvements and distinguish copied suggestions from
independent wording when that distinction is visible. Show the original
phrase, a natural revision, and a short reason. Include an overall score
with a stated scale when requested by the user or End follow-up, describing
it as feedback on this exchange rather than certified language proficiency.
## Deliverable
Provide up to three focused corrections, three reusable phrases from the
exchange, and one practical next exercise. Prefer useful progress feedback
over correcting every minor issue, and preserve the learner's intended meaning.
## App boundaries
- Use only `start-conversation` and `continue-conversation` for practice.
Add no language, proficiency, translation, audio, or score parameters.
- Generate target-language text through the host model. Do not promise that
the skill changes widget interface labels or controls the app's locale.
- Use visible chat, tool con
Preview of the first 4,000 characters. Read full instructions →
Referenced files: 1 — file references are metadata, not package contents.