Give Cleo back her afternoons

How it works

FluidVoice runs speech recognition entirely on your Mac's own processor, so the audio never leaves the machine or touches a server. While the app is active, whatever you say into a microphone is transcribed and typed straight into whichever text field has focus — a Claude chat window, an email, anything. Because the model runs locally, there's a short pause for processing but no network round-trip, and it keeps working without an internet connection.

Setting it up in full

1. Run the Homebrew command shown above to install the FluidVoice cask. Homebrew downloads and places the app in your Applications folder.
2. Open FluidVoice from Applications or Spotlight. macOS will prompt for Microphone access — approve it, otherwise dictation stays silent.
3. Grant Accessibility permission when prompted (System Settings, Privacy & Security, Accessibility). This is what lets the app type into other applications rather than just its own window.
4. Set or confirm the dictation hotkey in FluidVoice's preferences.
5. Test it: click into any text field, hold the hotkey, speak a sentence, and confirm the words appear correctly. If nothing appears, re-check the two permissions above.

Where it fits at Berkeley Credit

Dictate a fast case summary into Claude after a call with a broker rather than typing it up from notes. Talk through a complex exit strategy or adverse-credit history while it's still fresh, instead of writing bullet points later. Useful for anyone who thinks better out loud, especially when drafting the first rough version of a valuation note or client update.

Good to know

• Mac only — there is no Windows equivalent bundled with this tool.
• Accuracy drops with strong background noise, overlapping speech, or heavy accents on financial jargon and place names; always proofread before it goes anywhere client-facing.
• Running on-device is genuinely good for confidentiality since no audio is sent anywhere, but the typed text still ends up wherever you dictated it, so treat it with the same care as typed text.
• Requires an Accessibility permission grant, which some company IT policies restrict; check before rolling it out on a managed device.

• "Summarise what I've just dictated into a short case note for the file."
• "Turn this into a professional email to the broker, keep my tone but tidy the grammar."

How it works

FlashClip sits quietly in the background and records everything you copy — text, and often images — into a searchable history, not just the single item macOS normally remembers. A menu-bar icon or hotkey opens that history so you can scroll or search back through past copies and paste any of them again, even ones from several actions ago.

Setting it up in full

1. Run the Homebrew command shown above to install the app.
2. Launch it from Applications; it will add itself to the menu bar and may ask to start automatically at login — accept if you want it always running.
3. Grant Accessibility permission if prompted, which lets it paste directly into the active application via its shortcut.
4. Set your preferred hotkey for opening the clipboard history in preferences.
5. Test it: copy three or four different pieces of text, then trigger the history and confirm you can find and paste an earlier one.

Where it fits at Berkeley Credit

Keep your best-performing Claude prompts, such as the checklist you use for reviewing a credit file, ready to paste back in without retyping. Store a few stock client replies (delay notice, document request, offer confirmation) and reuse them across cases. Also handy mid-task: copy several figures from a valuation report in sequence, then paste them into Claude one after another without losing the earlier ones.

Good to know

• Mac only.
• Clipboard history is stored locally on your machine, but that still means client financial details, account numbers, or personal data you copy can sit in that history for a while — clear it periodically if you handle sensitive information often.
• Some clipboard managers offer a free tier with a paid upgrade for longer history or extra features; check FlashClip's own terms rather than assuming it's entirely free.
• Not a Claude feature — it's a general Mac utility that happens to make working with Claude faster.

• "Using the checklist I just pasted, review this credit file and flag anything missing."
• "Rewrite this stock reply so it fits a bridging case instead of a term mortgage."

How it works

OpenUsage is a small menu-bar utility that estimates how much of your current Claude usage allowance you have left and when it resets, without you having to open Claude and check manually. It's a community-built tool rather than an official Anthropic product, so it works from locally available usage signals rather than requiring a separate login — treat its numbers as a useful estimate rather than a guaranteed exact figure.

Setting it up in full

1. Run the Homebrew command shown above to install it.
2. Open the app once from Applications; it should place an icon in the menu bar and continue running quietly from then on.
3. Click the menu-bar icon to see the current reading, typically a percentage or a time until reset.
4. Leave it running in the background; no account linking or sign-in is needed.
5. Confirm it's tracking correctly by comparing its reading against any usage warning Claude itself shows during a long session.

Where it fits at Berkeley Credit

Glance at it before starting a long document review or a big multi-case catch-up, so you're not caught out mid-task by hitting a limit. It's also what makes the morning scheduled-task trick work well: running your heaviest jobs early, right after the usage window resets, rather than in the afternoon when you might already be running low.

Good to know

• Mac only.
• Unofficial and community-maintained; figures are estimates and can lag behind what Claude itself is tracking internally.
• No account credentials or sensitive data are involved, so the risk of running it is low.
• It doesn't do anything on its own — it's purely informational, there to help you plan your own use.

• "I've only got a little of my usage window left, what's the highest-priority thing to get through first?"
• "Give me the shortest possible version of this summary, I'm close to my limit."

How it works

1Password is an encrypted vault for logins, secure notes, and API keys, unlocked by a master password (and, on Mac, Touch ID). The relevant add-on lets Claude retrieve a named secret, for example an API key stored for a search tool, directly at the point it's needed, so the actual value is read into the command being run and never appears as plain text in the visible chat transcript.

Setting it up in full

1. Install the 1Password app if you don't already have it, and sign in to your account.
2. Run the command shown above to install the 1Password CLI, which is what lets other tools, including Claude, request a stored item by reference.
3. Sign in to the CLI once from Terminal so it's linked to your vault.
4. Create an item for each key or credential you want Claude to use, giving it a clear, memorable name.
5. Enable the relevant connector or integration so Claude can see that a vault is available.
6. Test it by asking Claude to use one stored item and confirming the task completes without the value ever being printed back to you.

Where it fits at Berkeley Credit

Store lender portal logins, e-signature platform passwords, and any free API keys, such as a search tool's, in one place instead of scattered across notes or memory. It's also the right home for anything a task needs Claude to authenticate with, since the credential itself stays inside the vault rather than being typed into a chat message.

Good to know

• A paid subscription product, though many people already have it for personal use.
• Works on both Mac and Windows.
• The safety benefit only holds if you consistently use it instead of ever pasting a password or key into chat by hand; one slip undoes the protection.
• Never enter your 1Password master password itself into any chat, form, or field outside the official app.

• "Use the API key stored in 1Password to run that search, don't ask me to paste it."
• "Check whether I already have a saved login for this lender's broker portal."

How it works

The Gmail connector uses Google's own sign-in (OAuth) to grant Claude a defined level of access to your inbox: reading and searching messages, and composing drafts. It cannot send anything on its own; every message it produces sits in your Drafts folder until you personally review it and press send, which is a deliberate safety gate rather than a technical limitation.

Setting it up in full

1. Open Claude's connector settings and choose to connect Gmail.
2. You'll be taken to Google's standard consent screen; sign in and review exactly what access is being requested before approving.
3. Select which Google account to link, if you have more than one.
4. Confirm the connector shows as connected in Claude's settings.
5. Test it by asking Claude to find a specific recent email; if it returns the right thread, the connection is working.

Where it fits at Berkeley Credit

Ask Claude to draft a chase to a broker who hasn't sent requested documents, in your usual tone. Get a quick summary of everything that landed overnight from lenders before your first coffee. Draft a client update once a valuation has been instructed, ready for you to check and send. Sort a cluttered inbox by which threads relate to which live case.

Good to know

• Nothing sends automatically; you always approve the final draft, which matters when dealing with client-sensitive financial correspondence.
• The access granted is often broader than a single thread, since it can search across the inbox, so it's worth knowing what you approved on the consent screen.
• You can revoke access at any time from your Google account's security settings, not just from Claude.
• Cloud-based, so it behaves identically on Mac and Windows.

• "Find the last email from the broker on the Oak Street case and draft a polite chase for the missing bank statements."
• "Summarise every email from lenders that came in overnight."

How it works

The Calendar connector, Google Calendar, with a separate one available for Fantastical, grants Claude access to your events so it can check free and busy time and, when you ask it to, create a new event directly. Unlike the email connector, calendar changes are generally applied straight away rather than sitting as an unconfirmed draft, so it pays to be specific about time, date, and attendees in the request.

Setting it up in full

1. Open Claude's connector settings and connect Google Calendar (or Fantastical, if that's what you use).
2. Authorise via the sign-in screen and review what access is being granted.
3. Choose which calendar or calendars to expose if you keep more than one.
4. Confirm the connection is active in settings.
5. Test it by asking Claude what your afternoon looks like, and check the answer matches your actual calendar.

Where it fits at Berkeley Credit

Book a callback slot with a client to talk through their exit strategy without switching apps mid-conversation. Find a gap to speak to a broker about a stalled case. Get a surveyor's inspection booked in against a completion deadline. Ask what free time you actually have this week before promising a call-back slot you can't keep.

Good to know

• Events are usually created immediately once you confirm the request, so check the details Claude proposes before saying yes.
• Which connector to use depends on which calendar app you actually live in day to day, Google Calendar or Fantastical.
• Access can be revoked at any time from your Google account settings, independent of Claude.
• Cloud-based, so it works the same on Mac and Windows.

• "Find me a 30-minute slot tomorrow afternoon to call the broker on the Oak Street case."
• "Add a callback with the client at 3pm Thursday to discuss their exit strategy."

How it works

These connectors grant Claude access to files stored in your Google Drive and let it read, and when asked write, values in Google Sheets. That means it can open a specific document or spreadsheet, read the actual figures in it, and produce a summary or draft based on live data rather than something you've copied and pasted in manually.

Setting it up in full

1. Open Claude's connector settings and connect Google Drive and Sheets.
2. Sign in and review the access being requested; some setups let you limit Claude to specific files rather than your whole Drive.
3. Confirm the connectors show as active.
4. Test it by asking Claude to open a named spreadsheet and read back a specific cell or range, to confirm it can see the right file.

Where it fits at Berkeley Credit

Point Claude at a loan-to-value calculator sheet and ask for a clean, client-ready quote summary based on the figures already in it. Pull the latest numbers from a live pipeline tracker instead of asking a colleague to export it. Draft a new one-page case summary from notes and documents already sitting in a case folder in Drive.

Good to know

• Depending on the access level chosen when connecting, Claude may see only files you've explicitly shared with it, or your whole Drive; check which you set up.
• Writes to a sheet are generally applied directly rather than held as a draft, so review what's changed afterwards.
• Client financial figures pass through Claude's systems during that session, so treat it with the same care you'd give any cloud-based tool handling that data.
• Works the same on Mac and Windows.

• "Open the LTV calculator sheet and work out the maximum loan for a GDV of £450,000."
• "Summarise this month's pipeline tracker into a one-page update for the team."

How it works

Tavily is a search API built specifically for feeding results to an AI rather than a person: it returns clean, ranked snippets and source links instead of raw web pages. Once a key is added, Claude can call it mid-conversation whenever it needs information more current than its training data, run a query, and bring the results straight back into the answer with sources attached.

Setting it up in full

1. Sign up for a free Tavily account and generate an API key from their dashboard.
2. Store that key somewhere safe, ideally 1Password rather than a plain text file.
3. Run the command shown above to add the key to Claude.
4. Confirm it worked by asking Claude something that needs current information and checking that it actually searches rather than answering from memory alone.

Where it fits at Berkeley Credit

Check a lender's currently advertised bridging rates before quoting a client. Look up a company's registered status and directors before relying on figures in an application. Check recent news on a property market or area relevant to a valuation. Confirm a solicitor or broker firm is genuinely who it says it is before proceeding on a case.

Good to know

• The free tier has a limited number of searches per month; check current allowances on Tavily's own site rather than assuming it's unlimited.
• Queries go to a third party, so avoid putting client-identifying details directly into the search text itself.
• The API key must be kept secret; store it in 1Password rather than pasting it into a chat message.
• Works identically on Mac and Windows, since it's just an API key rather than an installed app.

• "Search for [Lender]'s current bridging loan rates and summarise the key figures."
• "Check Companies House for [Company Name]'s registered status and directors."

How it works

This is Anthropic's own web-fetch tool: give Claude a web address and it retrieves the live page, strips it down to readable text, and reads that text as part of the conversation. There is no account and nothing to install, it is switched on by default. It only reads what is publicly visible on the page at that moment; it cannot log in, click through, or see anything behind a paywall or portal login.

Setting it up in full

1. Nothing to install. Paste the web address into the chat, or ask Claude to look something up and give it the link shown on the card.
2. Claude fetches the page and summarises what it found. Check the summary against the page yourself the first few times so you trust what it is reading.
3. If the page needs a login, such as a lender intranet, Claude will say it cannot reach the content. That confirms the tool's limit rather than a fault.
4. For a PDF criteria pack rather than a web page, use MarkItDown (card 9) instead, since fetch is built for HTML pages.

Where it fits at Berkeley Credit

Pulling a lender's published criteria page to check a scenario against it, reading a public Land Registry or planning portal page for a property, checking a solicitor's or valuer's public profile, or keeping across a lender's rate or news page without revisiting it manually.

• Only reads public pages, no lender portals or logged-in dashboards.
• Heavily JavaScript-rendered or paywalled pages may come back incomplete.
• Always sense-check figures and criteria against the source; treat this as a fast first read, not a substitute for the underwriting manual.
• Free, built in, works identically on Mac and Windows.

Try saying

• "Read this lender's bridging criteria page and tell me if a semi-commercial property at 65% LTV fits."
• "Check this solicitor's firm page and confirm they handle commercial conveyancing."
• "Fetch this planning portal page and summarise the application status for this address."

How it works

MarkItDown is a Microsoft open-source converter that Claude can call to turn PDFs, Word, Excel, and PowerPoint files into clean, structured text and tables. Instead of Claude trying to interpret a raw PDF's layout (which is expensive and error-prone with scanned valuation reports and long legal packs), MarkItDown does the extraction first, so Claude reads a tidy text version and spends far fewer tokens getting to the substance.

Setting it up in full

1. It runs as a local command-line tool, installed once via the file-to-markdown skill or the Finder Quick Action mentioned on the card, nothing to sign up for, no account.
2. Drop the document onto the Quick Action, or point Claude at the file path and ask it to convert it.
3. Claude runs the conversion and gets back Markdown text; open the .md output once to confirm tables and figures have carried over correctly, since complex tables in scanned PDFs can still lose structure.
4. Keep the original file too. MarkItDown produces a companion text version, it does not replace the source document for compliance or audit purposes.

Where it fits at Berkeley Credit

Turning a valuer's PDF report into text Claude can search and summarise for underwriting notes, converting a solicitor's legal pack into something Claude can question, for example whether it mentions restrictive covenants, or pulling figures out of a lender's Excel rate sheet without retyping them.

• Works best on text-based PDFs; scanned image-only documents need OCR first and may convert poorly.
• Free and local, nothing is uploaded to a third party during conversion.
• Mac setup here is a Finder Quick Action; Windows users would run the same underlying tool from a command line instead.
• It converts formatting and structure, not meaning; always read the converted summary against the original for anything that will inform a lending decision.

Try saying

• "Convert this valuation report to Markdown and summarise the key figures and any red flags."
• "Turn this legal pack into text and tell me if there are any restrictive covenants."
• "Pull the loan figures out of this Excel rate sheet and list them by LTV band."

How it works

Claude has built-in skills for producing real Office and PDF files rather than just formatted chat text, a genuine .docx with styled headings, a .pptx with laid-out slides, an .xlsx with working formulas, or a properly typeset PDF. It builds the file using the same open document formats Word, PowerPoint, and Excel use, so what you get opens and edits normally in those programs.

Setting it up in full

1. Nothing to install; ask for the document type you want directly, describing the content and how it should be structured.
2. Claude drafts the content and produces the file; it will usually offer it as a download.
3. Open the file in Word, PowerPoint, or Excel and check formatting, page breaks, and any figures before sending it anywhere; treat the first draft as a strong starting point, not a finished client document.
4. Ask for specific revisions in plain English, for example make the header navy or tighten the summary to one paragraph, rather than trying to edit the file yourself first.

Where it fits at Berkeley Credit

Turning loan terms into a client-ready one-page PDF summary, building a short deck comparing three live deals for an investment committee, producing an Excel amortisation or fee schedule for a borrower, or drafting a Word memo for a credit file.

• Good for first drafts and routine documents; complex brand templates or pixel-perfect layouts may still need a manual polish pass.
• No data leaves your conversation to a third-party service, the document is generated by Claude directly.
• Always proofread figures and terms before anything goes to a client or lender; this generates the document, it does not check the numbers against your systems.
• Works the same on Mac and Windows since the output is a standard file, not an app.

Try saying

• "Turn these loan terms into a client-ready one-page PDF summary."
• "Build a short PowerPoint deck comparing these three bridging deals for the investment committee."
• "Create an Excel fee and interest schedule for this 12-month bridge at 0.85% per month."

How it works

This is Anthropic's data analysis plugin: give Claude a spreadsheet or a set of figures and it writes and runs code behind the scenes to clean the data and render charts, then shows you the finished image or a small interactive dashboard rather than a wall of numbers. It is doing real calculation, not guessing, the chart reflects what is actually in the data you gave it.

Setting it up in full

1. No separate sign-up; it activates automatically when you attach a spreadsheet or paste tabular figures and ask for a chart or dashboard.
2. Upload the file or paste the data, then describe what you want to see, a trend line, a comparison, a monthly breakdown.
3. Claude processes the data and returns the chart; check the axis labels and totals against the source spreadsheet, since a wrong column mapping will produce a confident-looking but wrong chart.
4. For a dashboard with several charts, ask for the specific views you need rather than everything, a focused set is easier to check and to present.

Where it fits at Berkeley Credit

Turning monthly completion volumes into a trend chart for a partner update, visualising the loan book by LTV band or product type for a board pack, or building a small dashboard of pipeline figures ahead of a lender review meeting.

• Quality depends entirely on the input data; messy or inconsistent spreadsheets need tidying first.
• Best for internal reporting and first-draft visuals; a design-critical client-facing chart may still want a manual pass.
• Data is processed within the conversation; do not paste anything more sensitive than you would otherwise share with Claude.
• Works the same on Mac and Windows since it runs in the chat, not as a separate application.

Try saying

• "Chart our monthly bridging completions for the last 12 months from this spreadsheet."
• "Build a small dashboard showing the loan book split by LTV band and product type."
• "Turn this pipeline sheet into a chart of expected completions by month."

How it works

This is a set of Anthropic skills that keep everything Claude produces visually consistent: a brand-guidelines skill that stores your colours, fonts, and logo rules once and reuses them; a theme-factory skill that generates a matching palette and type system from a starting point; and a canvas-design skill for laying out standalone graphics. Together they mean Claude does not reinvent the look every time you ask for something new.

Setting it up in full

1. Tell Claude your brand basics once, company name, primary colours with hex codes if you have them, logo file, and preferred fonts.
2. Claude stores these as a reusable reference so future documents, decks, and graphics pull from the same definitions automatically.
3. Ask for a themed asset, a one-pager, a social graphic, a deck cover, and Claude applies the stored palette and fonts without you repeating them.
4. Review the first two or three outputs closely; once the palette and fonts are confirmed correct, later requests need far less checking.

Where it fits at Berkeley Credit

Keeping client-facing PDFs, decks, and social graphics all in Berkeley Credit's colours and fonts without manually restating them each time, producing a quick branded graphic for a case study or completed deal, or giving a new team member consistent branded templates from day one.

• Only as good as the brand information you provide; vague or missing details produce generic defaults.
• Best for supporting materials and quick graphics; a full brand refresh or logo design still needs a designer.
• No cost beyond your existing Claude plan; nothing to install.
• Works identically on Mac and Windows since it is all inside the conversation.

Try saying

• "Here are our brand colours and font, remember these for anything you design for us."
• "Make a branded one-page graphic announcing this completed deal, using our colours."
• "Build a deck cover slide in our brand style for this quarter's update."

How it works

ui-ux-pro-max is a community-built skill pack that gives Claude structured design judgement when it builds a page or form, sensible colour palettes, font pairings that actually work together, spacing and layout rules, and named design patterns to draw on, rather than picking defaults at random. It is reference knowledge Claude applies while generating HTML and CSS, not a separate design tool with its own interface.

Setting it up in full

1. This is a skill, not an app; it needs to be installed into your Claude setup once via the plugin mechanism shown on the card, after that it is available automatically.
2. Ask Claude to build a web page, form, or landing page as you normally would; the skill's guidance is applied without you needing to invoke it by name.
3. Review the result in a browser, check colour contrast, spacing, and how it looks on a phone-sized screen, since automated design guidance still benefits from a human eye.
4. If you want a different feel, more formal or more playful, say so directly, this steers the skill's choices rather than fighting them.

Where it fits at Berkeley Credit

Building a quick client enquiry form, an internal calculator page for deal terms, or a simple landing page for a new product line that looks considered rather than default-Bootstrap.

• Community-made, not an official Anthropic product; quality and maintenance depend on that community, not Anthropic support.
• Improves the starting point; it does not replace a professional designer for anything client-facing at scale or brand-critical.
• No extra cost beyond installation; runs inside your normal Claude usage.
• Works the same on Mac and Windows since the output is just a web page.

Try saying

• "Build a client enquiry form for bridging loans, and make it look professionally designed."
• "Create a simple internal calculator page for deal terms with a clean, modern layout."
• "Design a landing page for our new development finance product."

How it works

frontend-design is an official Anthropic skill focused purely on visual polish: once Claude has built a working page, this skill reviews and adjusts spacing, typography, colour use, and layout details so the result reads as intentionally designed rather than functionally assembled. It does not change what the page does, only how it looks and feels.

Setting it up in full

1. Nothing to install separately from your normal Claude setup; it is an official skill Claude can call on automatically when building or refining a page's look.
2. Ask Claude to build a page, or ask it to improve the visual design of an existing one you already built together.
3. Claude applies the polish pass and shows you the result; open it in a browser and check it at both desktop and mobile widths.
4. Ask for targeted adjustments, for example more whitespace around the form or a stronger heading hierarchy, rather than starting again, this skill responds well to specific feedback.

Where it fits at Berkeley Credit

Turning a functional but plain internal tool, such as a deal calculator or a status tracker, into something that looks properly finished, or giving a client-facing enquiry page the last polish before it goes live.

• It refines look and feel, not underlying logic or data handling; a broken calculation will still be broken, just better-looking.
• Official Anthropic skill, so reasonably well supported, but still needs a human check on contrast and readability for anyone with visual impairments.
• No extra cost; included in normal Claude usage.
• Works the same on Mac and Windows since it produces standard web page code.

Try saying

• "Polish the visual design of this deal calculator page, it works but looks unfinished."
• "Improve the typography and spacing on this client enquiry form."
• "Give this internal dashboard page a proper visual hierarchy."

How it works

Superpowers is a plugin bundle of skills that changes how Claude approaches a task before it writes anything. Instead of jumping straight to a draft, it prompts Claude to ask clarifying questions, break the job into steps, and check its own reasoning as it goes — a structured thinking layer bundled as an add-on, not a new model or connection.

Setting it up in full

1. Open Claude Code and run the /plugin command shown on the card to open the plugin browser.
2. Search for "Superpowers" in the marketplace list and select it to install — this copies the skill files into your Claude configuration folder, no separate download or account needed.
3. Start a new Claude Code session (or restart the current one) so the new skills load.
4. Confirm it worked by giving Claude a vague, multi-part task — for example, "help me review this bridging application" — and checking that it asks a follow-up question or lays out a short plan before diving in, rather than guessing at an answer immediately.

Where it fits at Berkeley Credit

Useful for anything with several moving parts: structuring a loan summary for underwriting, drafting a client explanation of an exit strategy, or working through lender criteria methodically instead of pattern-matching to a plausible-looking answer.

• It slows Claude down slightly on simple requests — for quick one-liners it can feel like overkill.
• It changes behaviour, not knowledge — Claude still needs the correct source documents or figures to get the substance right.
• Free to install and works identically on Mac and Windows since it runs inside Claude Code itself.

Try saying

• "Help me put together a bridging quote for this case — ask me what you need first."
• "Walk through this development exit case step by step before giving me a recommendation."

How it works

Stop Slop is a writing skill — a set of instructions Claude follows while drafting or editing text — that strips out the tics that mark writing as machine-generated: stock openers, false-balance hedging, over-use of words like "delve" or "leverage", and the habit of restating the question before answering it. It doesn't add facts or change meaning; it edits tone and rhythm so the result reads like something a person actually wrote.

Setting it up in full

1. Load the skill as shown on the card, either by name in a prompt or via the skill picker in Claude Code.
2. Give Claude the text you want written or a rough draft to clean up, and mention you want the AI-sounding phrasing removed.
3. Claude drafts, then re-reads its own output against the skill's checklist of common AI tells and rewrites the flagged lines.
4. Check the result reads naturally aloud — no stiff transitions, no repeated sentence shapes, no jargon that wouldn't appear in a normal email — and ask for another pass if a phrase still feels off.

Where it fits at Berkeley Credit

Client-facing emails, quote cover notes, and broker correspondence benefit most — text that needs to sound like it came from a person at Berkeley Credit, not a chatbot. Also useful for tidying internal case notes or valuation summaries before they go into a file.

• It is a style pass, not a fact-checker — it will not catch a wrong rate or an incorrect LTV, only clunky phrasing.
• Free and built into Claude — no separate account or cost.
• Works the same on Mac and Windows; it is just a prompt-level instruction set.
• Best used as a final polish step, after the substance of the email or document is already right.

Try saying

• "Rewrite this client email so it doesn't sound like AI wrote it."
• "Clean up this valuation summary — make it read like I wrote it, not a bot."

How it works

CLAUDE.md is a plain text file that Claude automatically reads at the start of every session in a given folder or project. There is no install: you simply create the file and write your preferences into it — your role, how you like things written, recurring facts about your work — and Claude treats it as standing context every time it opens that project.

Setting it up in full

1. Create a file named exactly CLAUDE.md in the folder you work from most (or use the location shown on the card).
2. Write short, plain statements: who you are, what you do at Berkeley Credit, tone preferences ("keep emails brief, no jargon"), and any standing facts Claude should never have to be told twice (lender panel, typical deal sizes, terms you always use).
3. Save the file as plain text — no special formatting required, though simple headings and bullet points help keep it scannable.
4. Confirm it worked by starting a fresh conversation in that folder and asking Claude something that depends on the note (for example, "what's my usual tone for client emails?") — it should answer from the file without you repeating it.
5. Update the file whenever a preference changes; Claude re-reads it every session, so edits take effect immediately next time.

Where it fits at Berkeley Credit

One-time note covering your writing style, standard disclaimers, common lender criteria, or house terminology means every future quote, email draft, or case summary starts from the right baseline without you re-explaining it each time.

• It is only read for sessions started in or under that folder — a note saved elsewhere on the computer will not be picked up.
• It is plain text Claude trusts as instructions, so avoid putting anything sensitive (client data, passwords) in it — treat it as a preferences note, not a case file.
• No cost, no install, works identically on Mac and Windows since it is just a file.
• It shapes tone and habits, not facts — it will not remember a specific client's numbers unless you write them in.

Try saying

• "Create a CLAUDE.md for this folder with my email tone and our standard bridging disclaimers."
• "Add a note to my CLAUDE.md that I always want LTV shown as a percentage, not a ratio."

How it works

claude-mem is a plugin that saves a summary of what happened in each Claude Code session to a local memory store, then automatically feeds relevant past sessions back into a new conversation when it starts. Unlike CLAUDE.md, which holds fixed preferences you write once, claude-mem captures what you actually did — the case you were working on, decisions made, where you left off — without you writing anything down.

Setting it up in full

1. Run /plugin as shown on the card and install claude-mem from the marketplace list.
2. Use Claude Code as normal for a session on a piece of work — the plugin records a summary in the background as you go, no extra steps needed mid-session.
3. Start a new session later on the same project and ask Claude to recall the earlier work — it should pull the relevant summary back in automatically.
4. Confirm it worked by checking Claude references specifics from the earlier session (case name, decisions made) without you re-explaining them.

Where it fits at Berkeley Credit

Picking a bridging case back up after a week away, resuming a half-finished valuation review, or continuing a multi-day client correspondence thread without re-typing the background each time.

• Community-made plugin, not an official Anthropic product — reasonably mature but worth treating as a convenience layer, not a system of record.
• Memory is stored locally on the machine it ran on; it will not follow you to a different computer without extra setup.
• Do not rely on it for anything that must be auditable — keep case files and compliance records in your proper systems, not just in Claude's memory.
• Free to install; works on Mac and Windows since it runs inside Claude Code.

Try saying

• "Pick up where we left off on the Smith bridging case."
• "What did we decide about the exit strategy on that development deal last week?"