Prompts you can copy.

You are working with [YOUR NAME], a [YOUR ROLE] at [YOUR COMPANY].

Voice and format rules to apply to every response:
- British English. Use organise, analyse, prioritise. Never the American spellings.
- No em or en dashes. Use commas, semicolons, colons, parentheses, full stops.
- No contractions in formal output. Write "do not", not "don't".
- Lead with the answer. Do not preface with "Sure!" or "Great question".
- When asked for a deliverable, produce it directly. Skip the meta commentary.
- Numbers under ten written as words; ten and above as numerals.
- If you are uncertain, say so explicitly. Do not invent.

When drafting emails or client-facing prose:
- Match the recipient's likely register (formal for new clients, conversational for known ones).
- Never use exclamation marks except in direct quotes.
- Open with the point, not pleasantries.

When producing structured output (tables, lists, code):
- Tables for comparisons of three or more options.
- Bullets for parallel items; numbered lists only when order matters.
- Code in triple backticks with language tag.

Constraints:
- Never include "as an AI" or any meta statement about being a model.
- Never apologise for limitations; just state what you can and cannot do.

Run the prompt-builder skill on the brief above. Default mode (master_prompt). Apply my Consilix house style. Emit the runnable master prompt plus the usage template.

You are my inbox triage assistant. Given the emails below, produce a daily briefing for me to read with my morning coffee.

Format:

## Needs my response today
- One line per email. Sender, what they need, deadline if stated.

## Decisions I need to make
- One line per item. The decision, what is at stake, what info I would need.

## FYI, no action needed
- One line per email. Sender, topic.

## Anything I might be missing
- Threads where the sender is waiting on something I have not addressed.

Be ruthless. If an email is marketing, automated, or noise, do not list it. Keep the whole briefing under 250 words.

Emails:
{paste your inbox export here, or rely on the Gmail connector}

Draft a reply to the email below. Rules:

- Match the sender's register: if they wrote a short, casual message, reply in kind.
- Lead with the answer or the most important point.
- Keep it under 80 words unless the email genuinely requires more.
- British English. No em dashes. No contractions.
- Do not start with "Thanks for your email" or any other filler.
- If there is anything I should confirm before sending, list it as a separate "Confirm before sending" bullet at the bottom.

Original email:
{paste here}

My answer to their question:
{a sentence or two of what you want to say}

Below is an email thread. Extract the actions assigned to me, the actions assigned to others, and the decisions that have been made.

Output:

## Actions on me
- [What I need to do] — by [date if stated, "no date" if not]

## Actions on others
- [Name] — [what they should do]

## Decisions made
- Brief bullets.

## Still open
- Questions or items raised but not resolved.

If the thread is contradictory (someone changed their mind), reflect the latest position only.

Thread:
{paste here}

Below is a meeting transcript or note set. Produce a summary in this exact shape:

## Decisions made
- One line each. The decision, briefly.

## Actions
- [Owner name] — [the action] — by [date if stated, "TBC" if not].

## Open questions
- Things raised but not resolved.

## Recap (3 to 5 sentences)
- Plain English summary of what happened. No buzzwords.

Voice: British English, no em dashes, no contractions, no marketing language. If owners or dates were not stated, mark TBC rather than guess.

Transcript:
{paste here}

You are reviewing a compliance document or contractor submission. Reference materials (project rules, H&S requirements) are in this Project.

Produce a review in this format:

## Verdict
- One line: ACCEPT / ACCEPT WITH CONDITIONS / REJECT.

## Reasoning
- Three to five bullets explaining the verdict against the project rules.

## Gaps the contractor must close
- Numbered list. Each gap names the rule it violates and the corrective evidence needed.

## Risks I would flag for human review
- Anything that is plausible but I cannot verify from the documents alone. Be explicit about uncertainty.

Voice: British English, no contractions. Cite specific clause numbers from the project rules where possible. Never invent a clause number.

Document under review:
{paste or attach}

You are comparing contractor quotes for the same scope of work. The scope of work is described in the Project knowledge.

Produce a comparison table in this exact format:

| Item | Quote A | Quote B | Quote C |
| --- | --- | --- | --- |
| Total price (ex VAT) | £X | £X | £X |
| Labour cost | £X | £X | £X |
| Materials cost | £X | £X | £X |
| Programme duration | X weeks | X weeks | X weeks |
| Inclusions noted | ... | ... | ... |
| Exclusions noted | ... | ... | ... |
| Caveats / assumptions | ... | ... | ... |

Then a section "Where the quotes diverge" listing the top three differences in scope or assumptions that explain price gaps.

Then a section "What to clarify before deciding" listing the open questions I should put back to each contractor.

Voice: British English. Numbers as numerals. Never invent figures; if a quote does not state something, write "Not stated".

Quotes:
{paste all quotes here, or attach to the Project}

Build me an interactive sales proposal as an Artifact. The proposal is for {client name} for {scope: refurb / new build / extension / fit-out}.

Required inputs the Artifact should expose as form fields:
- Site area in square metres
- Finish level (Basic / Standard / Premium)
- Number of bedrooms / units
- Programme duration in weeks
- Any optional add-ons (list to be confirmed)

The Artifact should:
- Render a single-page proposal layout with sections: Cover, Scope, Approach, Timeline, Pricing, Next steps.
- Recalculate the headline price live as the inputs change.
- Use realistic UK construction unit prices for the chosen finish level.
- Render any pricing assumptions as small footnotes.
- Use Consilix or {company} brand colours if I provide them; otherwise use a neutral professional palette.

Voice: British English, no em dashes, no marketing language. Lead with the scope and the price.

Output the Artifact directly.

Produce a research brief on [TOPIC] in this format:

## What we know with confidence
- Five to seven bullets. Each one citation-backed. Use (Source, year) inline.

## What is disputed or uncertain
- Bullets covering claims where credible sources disagree.

## What is changing fast right now
- Things where the situation in 2026 differs from a year ago.

## Implications for [MY CONTEXT]
- Three to five bullets specific to my business or decision.

## Sources
- Numbered list. Every source cited above must appear here with a URL.

Voice: British English, no em dashes. Cite specific publishers and dates. If a fact lacks a credible source, mark it "[unverified]" rather than including it as a confident claim.

Topic to research: {topic}
My context: {one-line about your business / decision}

Rewrite the text below applying my voice rules (already set in Custom Instructions): British English, no em dashes, no contractions, no marketing language, lead with the answer.

If the original text uses any forbidden words (transform, leverage, streamline, cutting-edge, world-class, robust, harness, game-changing), replace them with concrete alternatives. Do not soften my points. Do not lengthen. If the original is too long, shorten.

Text to rewrite:
{paste here}

Below are my notes from the week. Turn them into a weekly update for the team.

Format:

## This week we
- Three to five bullets on what got done.

## Wins
- One or two highlights. Specific, not generic.

## Blocked or worth raising
- Anything that needs help, money, or a decision.

## Next week
- Three to five priorities.

Voice: British English, no em dashes, no exclamation marks, no marketing language. Conversational but professional. Under 300 words total.

Notes:
{paste here}

I want to build a Claude skill (a reusable instruction module). Walk me through it conversationally.

Start by asking me:
1. What does the skill do? Describe the task in one paragraph.
2. What inputs does it need from the user each time?
3. What outputs does it produce? Are they prose, structured data, an Artifact, or a side effect?
4. What house rules apply (voice, format, do-not-do)?
5. Are there any reference materials it should always consult?

After my answers, draft a SKILL.md following the Anthropic skill convention:
- Frontmatter with name, description, expected inputs, outputs.
- A body explaining the skill's responsibilities, constraints, and how to invoke it.

Then propose how I install and test it. Be specific: file path, where to drop it, how to verify it loads.

Summarise the document below for a busy senior reader. Format:

## In one sentence
- What this document is and why it matters.

## The five things you must know
- Numbered bullets, each two sentences max.

## What it asks the reader to do
- Explicit asks, deadlines, sign-offs.

## Risks or red flags
- Anything that should prompt a deeper look before agreeing.

## Numbers worth knowing
- Specific figures, with units and dates.

Voice: British English, no em dashes, no filler. Quote exact phrasing for any commitment or liability claim. If the document is unclear on something material, note it.

Document:
{paste or attach}

I am deciding between three options. Help me decide.

Options:
A: {describe option A}
B: {describe option B}
C: {describe option C}

Criteria that matter to me (in order of weight):
1. {criterion}
2. {criterion}
3. {criterion}

Output:

## Comparison matrix
| Criterion | Weight | A | B | C |
| --- | --- | --- | --- | --- |

(Rate each option 1-5 against each criterion. Multiply by weight.)

## My honest recommendation
- One option, one paragraph. Specific to my context.

## What I would change my mind for
- The single piece of new information that would flip the call.

Voice: British English. Direct. Do not present each option as "valid" — pick.

Build a live artifact: a Finance Operations Dashboard for a UK firm running a property portfolio plus a cafe operating business. This is a persistent dashboard intended to live in the Live Artifacts tab and refresh from connected data sources each time I reopen it.

Single-page Artifact with four tabs.

TAB 1 — Invoices (manual + Gmail-augmented)
Editable table. Columns: party, direction (in/out), amount £, issued date, due date, status (open/paid/overdue), reference, notes. Auto-calculate overdue flag from due date vs today. On reopen, scan Gmail for the last 14 days for emails with subjects matching invoice patterns; add candidate rows below the manual rows in a separate section "Suggested from inbox" for me to promote with one click. Do not auto-modify any row I have already edited.
Footer row: total open £ in, total open £ out, net position.

TAB 2 — Bank reconciliation (manual)
Editable table per account. Columns: account name, last reconciled date, transactions reconciled count, transactions outstanding count, status (current/lagging), assigned to, notes. Status auto-derives: current if reconciled within 7 days, lagging if older. Manual entry only.

TAB 3 — Cashflow (12-week rolling, manual inputs, computed view)
Inputs: opening balance, weekly property rental inflow, weekly cafe net takings, weekly operating costs. Computed: running balance per week, minimum balance across 12 weeks, alert banner if any week dips below zero. Plus a section for one-off outflows (table: week, description, amount, add/remove rows). No MCP refresh.

TAB 4 — Email-sourced tasks (live from Gmail)
The live tab. On every reopen, refresh from Gmail. Editable table. Columns: task description, owner, source (sender name), received date, due date, status (open/chased/done), last chased date, notes.
Refresh logic on open:
- Pull unread Gmail messages received in the last 7 days
- Identify ones containing a request, ask, or deadline directed at me. Skip newsletters and automated mail.
- For each, add a row with task, owner (me by default), source, received date, due date if stated, status (open).
- Do not overwrite rows where status is chased or done.
- De-duplicate by source+task.
Quick actions per row: mark chased, mark done, copy chase template.

GLOBAL FEATURES
- Tabs at top with count badges showing open items per tab
- Inline editing on every cell
- Add/remove rows on every table
- Export current state as JSON button in footer
- Refresh from Gmail now button in tab 4 header
- Last-refreshed timestamp shown bottom-right

STYLE
- British English. No emoji. No marketing language.
- Buttercream #F5F0E8 background, near-black #131314 text, deeper orange #B54A1F accent
- Headings in Fraunces serif (Google Fonts), body in Inter, mono in JetBrains Mono
- Status chips: open = neutral, chased = orange tint, done = muted green, overdue = red tint
- Mobile-responsive: tabs become a horizontal scroll under 720px

PRE-POPULATE on first build with realistic demo data:
- Invoices: 6 rows mixing in/out, 2 overdue, 2 due this week, amounts £1,500 to £18,000
- Bank reconciliation: 3 accounts, one lagging
- Cashflow: opening £215,000, property weekly inflow £45,000, cafe weekly net £12,500, operating costs £40,000 weekly, one £150,000 VAT outflow in week 6
- Email-sourced tasks: 5 sample rows that get replaced or merged on first Gmail refresh

After building, confirm which MCP connectors this Artifact uses so I can verify Gmail is authorised.

Build a live artifact: an interactive tenant proposal one-pager that recalculates pricing live as inputs change. Save it to Cowork so I can reopen per prospect and tweak inputs.

Inputs as editable form fields:
- Tenant name (text)
- Intended use (dropdown: retail, office, restaurant, mixed use)
- Floor area (square feet, number)
- Lease term (years, number)
- Headline rent (£ per sqft per year, number)
- Service charge (£ per sqft per year, number)
- Rent-free period (months, number)
- Deposit (months of rent, number)

Outputs in a clean one-pager layout:
- Headline "Proposal for [Tenant name]"
- Property summary section (the inputs above as bullets)
- Calculated cost table: annual rent, annual service charge, total year-1 cost, total over full term, deposit amount, effective rent after rent-free period
- Standard heads of terms section (placeholder text I can edit)
- Next steps section

STYLE
- British English. £ for currency, commas for thousands separators.
- Buttercream #F5F0E8 background, near-black #131314 text, deeper orange #B54A1F accent.
- Fraunces display, Inter body, JetBrains Mono for amounts.
- Print-friendly layout.

Pre-populate with: Tenant = "Compass Coffee Ltd", retail use, 2,400 sqft, 10 year lease, £45 per sqft headline rent, £8 per sqft service charge, 6 months rent free, 12 months deposit.

Make every field editable. When I change the rent, all calculated figures update live.

Build a live artifact: a contractor compliance dashboard. Lives in the Live Artifacts tab and refreshes when I reopen.

Read documents from the connected folder ~/Romulus/contractors/ and pull contractor correspondence from Gmail with the label "contractors".

Output a single table. Rows: one per contractor (extracted from the folder + email senders, deduplicated by name).
Columns: contractor name, latest RAMS received date, RAMS status (received/missing/expired), insurance verified (Y/N), induction completed (Y/N), sign-off status (cleared/conditional/blocked), last contact date (from email), next milestone, notes.

Overdue logic:
- RAMS more than 90 days old: red flag
- Insurance not verified: red flag
- Sign-off blocked: red flag
- Any green row: cleared to work

Above the table:
- Summary tiles: total contractors, count cleared, count blocked, count pending sign-off
- Filter chips: all / cleared / blocked / pending

Below the table:
- "Outstanding contractor correspondence" section: rows from Gmail unanswered for >3 days

STYLE
- British English. No marketing language. No emoji.
- Buttercream #F5F0E8 background, near-black #131314 text, deeper orange #B54A1F accent.
- Fraunces display, Inter body.
- Mobile-responsive: table collapses to cards under 720px.

Pre-populate (only if the folder is empty) with 5 sample contractor rows so the layout renders. The first real refresh replaces these.

After building, list the MCP connectors used.

Build a live artifact: a quality-controlled inbox triage view. Live tab pulls Gmail on every reopen.

Categorise the last 24 hours of inbox into four sections:
1. Needs my response today (questions, deadlines, asks)
2. Decisions pending (something needs choosing)
3. Compliance-flagged (any email mentioning deadlines, financial commitments, contractual language, regulatory terms) — these get an "AI-flagged risk: verify manually" indicator on the row
4. Automated noise (newsletters, build alerts, no-reply)

Per row: sender, subject, one-line summary, urgency, confidence in categorisation (0-100), the AI-flagged risk indicator if applicable.

Top of view: total emails processed, count needing response, count compliance-flagged.

Quick actions: mark resolved, open email in Gmail, copy reply draft.

STYLE
- British English. No marketing language.
- Buttercream background #F5F0E8, near-black text #131314, deeper orange accent #B54A1F.
- Compliance-flagged rows get a red left border so they cannot be missed.

The key design rule: when a confidence is below 70%, surface that visibly so the user double-checks rather than trusting the AI's call. Quality control over volume.

Build a live artifact: a development pipeline + deal activity tracker.

Three tabs:

TAB 1 — Sites in pipeline (manual)
Columns: site name, location, area (sqft or acres), use class, planning status (none/pre-app/submitted/granted/refused), target acquisition or disposal, owner, notes.

TAB 2 — Planning correspondence (live from Gmail)
On reopen, pull Gmail for the last 14 days filtered for planning-related correspondence (sender contains "planning", "council", or recognised consultants). For each, add a row: sender, subject, site reference if extractable, summary, action required (Y/N).

TAB 3 — Agent enquiries (live from Gmail)
On reopen, pull Gmail for the last 14 days filtered for letting/sales agent activity. For each, add a row: agent, sender, summary, site, status (new/responded/awaiting agent), follow-up due date.

Plus a top-line section: scheduled summary at 07:30 Monday — last week's activity rolled up.

STYLE
- British English. No marketing language.
- Buttercream paper background, near-black text, deeper orange accent.
- Fraunces display, Inter body.

After building, list the MCPs in use.

Build a live artifact: a daily executive briefing.

Auto-refresh on open. Also scheduled to refresh 07:00 weekdays.

Pull from Gmail and Calendar MCPs.

Sections:

1. Today at a glance — date, count of meetings, count of new asks overnight, count of decisions pending.

2. Today's calendar with AI-prepared prep notes — one row per meeting: time, title, attendees, AI-generated 2-sentence prep note based on recent email threads with the attendees.

3. Overnight asks (since 18:00 yesterday) — rows of emails containing a request directed at me. Source, ask summary, urgency, last action.

4. Decisions pending across the team — extracted from email threads where someone is waiting on a decision. Summary, who is waiting, last touch.

5. Action items extracted from this week's email threads — what I have committed to that has not yet been done.

STYLE
- British English. No marketing language.
- Buttercream paper background, near-black text, deeper orange accent.
- Fraunces display, Inter body.
- Print-friendly so I can hit Cmd+P and have a clean briefing on paper.

After building, list MCPs and confirm scheduled-refresh is configured.

Build a live artifact: a team AI adoption tracker.

One single editable table. Columns: team member name, role, this week's AI use (1-sentence each, up to 3 entries), tools used (chips: Claude, ChatGPT, Copilot, Gemini, Perplexity, Other), last updated date.

Footer summary: total tools used across the team, total entries this week, top use case categories.

Below the table: a "week-over-week" trend showing entry count by week for the last 8 weeks, as a simple bar chart.

Every row editable inline. Add row button at bottom.

STYLE
- British English. No marketing language.
- Buttercream paper background, near-black text, deeper orange accent.
- Fraunces display, Inter body.
- Mobile-responsive.

Pre-populate with the Romulus AI task force members as rows so the layout renders. Real entries replace placeholders.