# Claude Code for SDRs, A Workday Rebuilt as Skills > Canonical: https://www.yalc.ai/blog/claude-code-for-sdrs/ Every hour of the SDR day mapped to a real Yalc skill, the MCPs it calls, and the output it drops on your desk. Claude Code for SDRs is a way to rebuild the sales development job as inspectable markdown skills that call your data and messaging APIs directly. One skill pulls the morning list, one qualifies it, one drafts LinkedIn and email, one triages replies, and one writes the weekly report. Prompts stay editable, sends stay compliant. ## Why the SDR job is the easiest one to rebuild in skills The SDR day is the most rules driven job in modern GTM. Source a list against an ICP, enrich it, score it, personalize a note, send, watch replies, log to the CRM, repeat. Every step has a defined input and a defined output. Every step has a rule the rep already carries in their head about what to keep and what to drop. That structure is what makes it the easiest role to migrate onto Claude Code. Not because the work is trivial, but because the work is decomposable. You can name each block, tell the agent what good looks like on that block, and hand it the API keys it needs to execute. What you get back is not a chatbot pretending to be an SDR. You get five or six markdown files, each doing one part of the job, each editable when you learn something new. That is the shape of [Claude Code for the sales function](/blog/claude-code-for-sales/) applied to the SDR role specifically. Most SDR tooling content treats Claude as a copy assistant, a place to draft a subject line. That framing underuses it. The operator move is to treat every recurring block of the day as its own skill, run those skills from one prompt in the morning, and spend the rest of the day on the two things that never delegate, the discovery call and the judgment call about who to write next. ## What Claude Code is for an SDR, and what it is not Claude Code is a command line environment where the model reads and writes files on your machine, calls tools you approve, and executes multi step work without a canvas or a vendor UI. For an SDR, that means the difference between a chat window that types drafts back at you and an operator that opens Crustdata, pulls the accounts you told it about, runs your scoring rules, drafts the note in your voice, and stops on your desk waiting for approval. What it is not, is another entry in the [AI SDR tools landscape](/blog/ai-sdr-tools/). It does not warm your domain. It does not own an inbox. It is the orchestrator, not the sender. You keep [Instantly](/tools/instantly/) or your existing sequencer for the wire, keep the CRM for the record, and let Claude Code be the layer that reads, decides, and drafts. That separation is the safety net. Claude drafts, your sequencer sends, and the split keeps deliverability rules where they belong, inside a tool built for them. Pricing tracks that framing. Anthropic's public plans put the Pro tier at 17 dollars per month on annual billing and 20 dollars monthly, Max 5x from 100 dollars per month, and Team seats at 20 to 25 dollars per seat per month, per [Claude's pricing page](https://claude.com/pricing) fetched 2026-07-08. Compared to a full replacement quote near 36,000 dollars per year for a managed AI SDR, or a Clay Growth seat at 495 dollars per month, running the orchestration layer yourself in Claude Code sits at a different order of magnitude. ## Morning list pull, the Crustdata skill The workday starts with a list. Not who could I write today, but a specific slice of the ICP that changed in the last twenty four hours. New hires in the target seniority. Fresh funding rounds. Job posts that hint at the pain your product solves. The morning skill is a markdown file that knows your ICP, calls the [Crustdata MCP](/mcps/crustdata/) for the change, deduplicates against yesterday's list and against everyone already in an active sequence, and drops the surviving rows into a scratch file with the trigger note attached. The rep opens their laptop at 9 a.m. and sees fifty rows, each with the reason it made the list. Written as a skill, the input is a natural language prompt. Pull today's list against the ICP. The output is a markdown table with company, contact, trigger, and a one line why. The MCPs called are Crustdata for signals and people, and [FullEnrich](/tools/fullenrich/) for the email gap when Crustdata cannot resolve a work address. The skill runs in under two minutes because the model is deciding which rows to keep, not typing them. The reason this works is that the rules for keeping a row are the same rules the SDR has been applying by hand for years. Write them down once, and the skill applies them every morning without drift. ## Qualification scoring, the qualify-leads skill The list is not the deliverable. The scored list is. Yalc ships a public [lead qualification skill](/skills/qualify-leads/) that reads a scoring rubric out of your config folder and ranks each row against it, cold, warm, hot, with the reason written next to the score. An SDR can install this skill in ten minutes. The scoring rubric is a markdown file, not a database column. Change the weight on hired first VP of sales in the last 30 days from 5 to 8 and the next run reflects it. There is no vendor to email, no config panel, no product manager to lobby. Two properties matter. First, the rubric is auditable. When a rep asks why lead 47 got scored a 6 and not a 4, the answer sits in one file. Second, the rubric evolves. Every week the SDR looks at what scored high and did not convert, edits the rubric, and reruns. That is compounding, in the sense that the operator playbook for [faster lead qualification](/blog/ways-to-qualify-leads-faster/) argues the middle mile has to compound or it decays. A CRM field that never changes because nobody wants to break the pipeline dashboard is not compounding. ## Personalization at scale, the Unipile skill Personalization is where most SDR copy dies. A merge tag on a first name and a company is not personalization. The template dresses up but the reader still knows what it is. That gap is why manual personalization caps at fifteen to twenty emails a day per rep before the quality collapses, per [Claude Code HQ](https://www.claudecodehq.com/blog/claude-skills-sales-teams). The personalization skill reads three inputs, the scored row, the trigger, and a voice file that captures how the rep actually writes. It then drafts a LinkedIn invite note under 250 characters, a first touch email under eighty words, and a follow up under forty. Each draft cites the specific trigger, not the generic pain. It calls the [Unipile MCP](/tools/unipile/) for LinkedIn context on the recipient, and it stops before sending. The rep reads five drafts in the time it used to take to write one. The compound effect is what matters. A rep with a working personalization skill can put out fifty first touches a day at higher quality than fifteen they used to grind out by hand. Reply rate goes up, not down, because every touch is anchored in a real change at the account. The voice file lives in the same folder as the skill, so any rep on the team can borrow it and stay on brand. ## Send and reply triage, the sequencer skill The next skill hands the drafts off to the sender, not the model. That handoff is the deliverability firewall. Instantly or a similar warmed sending platform owns SPF, DKIM, DMARC, one click unsubscribe, and complaint rate monitoring. Google and Yahoo have required all of those since February 2024, with a spam complaint rate below 0.3 percent, per [Google's bulk sender guidelines](https://support.google.com/a/answer/81126). A model drafting well does not exempt anyone from that rule. The reply side is where the skill earns its keep. Every reply that lands in the connected inbox gets read once by the classification skill and tagged, positive, objection, out of office, wrong person, unsubscribe. Positive replies get an alert to the rep with the thread context and a suggested next step. Objections get a matched response draft, never sent, only queued. Out of office get a soft bump three days later. Wrong person gets a routing note. Unsubscribes get logged to the [HubSpot MCP](/mcps/hubspot/) and quietly removed. The rep opens the inbox once at 11 a.m. and once at 4 p.m. and clears the queue in twenty minutes. That is the difference between reactive inbox chasing and a queue that is already sorted for them. ## Weekly reporting, the monthly-report skill Friday afternoon is the weakest hour in the SDR week. The rep is tired, the pipeline call is Monday, and the numbers everyone wants live in six different tools. The reporting skill reads from HubSpot, the sequencer, and the CRM history, and writes a weekly report that follows the same template every time. Sends, replies, positive rate, meetings booked, top three trigger types that worked, top three that did not, and a proposed change to the personalization file or the qualification rubric for next week. That last piece is the point. A report that only lists numbers is a status update. A report that proposes a change is a system that gets sharper. The rep reads the proposal in ten minutes, edits, and commits. Same shape as [ways to scale an SDR team without hiring](/blog/ways-to-scale-sdr-without-hiring/), where the compounding gain sits in the config, not the headcount. ## Claude Code vs a full AI SDR platform The obvious comparison is against a managed AI SDR like 11x or Artisan. The obvious answer is that they solve different problems. A full replacement platform sells you an outcome without visibility. Claude Code sells you visibility without a wrapper. The trade for an SDR who wants to keep their job and get sharper at it is easy. A managed platform hides the prompt. When a send goes off brand or an objection is fumbled, the fix is a support ticket. In a skill folder the prompt is a file. The fix is an edit and a rerun. That property matters most when a Google spam complaint threshold moves or a message pattern stops working, because the operator can respond the same day instead of waiting for a vendor sprint. There is a second trade that shows up on the pricing page. A single Max seat at 100 dollars per month can drive multiple skills for a rep. A managed AI SDR quote at 36,000 dollars per year is the same money spread across one hidden config the rep does not own. If the rep leaves, the config leaves. The skill folder stays in the repo. That is the same argument the [AI sales agents](/blog/ai-sales-agents/) piece makes for keeping the intelligence layer in the operator's hands. ## What still needs a human Two blocks of the day never delegate. The first is the choice of who to write. Claude Code will happily draft to any row on a list. The judgment about which trigger is real, which company is worth the pursuit, and which message angle to test this week, is a first mile decision. Skills apply rules. Humans set them. The second block is the discovery call and every conversation after it. No skill takes a discovery call. No skill closes. What the skills do is protect the rep's attention so they arrive at the call rested and specific, having read a brief the reporting skill wrote for them ten minutes before the meeting. Everything between those two blocks, the sourcing, enriching, scoring, drafting, sending, triaging, logging, and reporting, is the middle mile. That middle mile is where Claude Code lives. Yalc's take on [ways to use Claude Code for GTM](/blog/ways-to-use-claude-code-for-gtm/) argues that owning the middle mile in code is the durable advantage of the AI native operator, and the SDR seat is the clearest example of that pattern paying off inside a single workday. ## The five skill starter kit an SDR installs this week Five skills, in this order, get an SDR most of the way to a redesigned day. 1. **Morning list**. Calls Crustdata for signals, deduplicates, drops a scored table. Output is fifty rows a day with triggers attached. 2. **Lead qualification**. Runs the public qualify-leads skill against the rubric. Output is a cold, warm, or hot tag with a reason. 3. **Personalization**. Reads the voice file and the trigger, drafts three touches per row. Output is a folder of drafts the rep approves. 4. **Reply triage**. Classifies every reply, alerts the rep on positives, queues objections and bumps. Output is a cleared inbox in twenty minutes. 5. **Weekly report**. Pulls from HubSpot and the sequencer, writes the report, proposes changes. Output is a Monday morning ready doc. Install the first two on day one. Add the personalization skill on day three. Add reply triage on day five. Add the weekly report the following Friday. By week two, the workday looks different because the rep's hours moved from doing to reviewing. That is the whole play. ## FAQ ### Do I need to know how to code to use Claude Code for SDR work? No. Claude Code takes natural language instructions and executes them against tools you have connected. The skills that ship in the Yalc library are markdown files that read like a playbook, not a program. What a rep needs to learn is how to edit a markdown file and how to describe the input and output of a repeating block of work, which is closer to writing a good SOP than writing code. ### Does using Claude Code for email hurt deliverability? Only if you let it send. The safe pattern is to keep drafting in Claude Code and to hand the send to a warmed platform like Instantly that owns SPF, DKIM, DMARC, one click unsubscribe, and the sub 0.3 percent complaint rate that Google and Yahoo require. Claude drafts. The sender sends. That split is what keeps a rep's domain out of trouble. ### How does Claude Code compare to a full AI SDR platform like 11x or Artisan? They solve different problems. A managed AI SDR sells an outcome and hides the config. Claude Code sells visibility and expects the operator to own the config. The right pick depends on whether the team wants a black box that runs itself, with the risk that a bad send is only fixable by a support ticket, or a skill folder they can inspect and edit the same day something changes. ### Can Claude Code fully replace a human SDR? Not in a way most teams should aim for. The middle mile of the SDR day, sourcing, enrichment, scoring, personalization, triage, and reporting, is where Claude Code earns its keep. The first mile, deciding what to hunt this quarter, and the last mile, running the call and handling the deal, still need a human. The point of the rebuild is to expand the middle mile the skills can carry, not to remove the rep. ### How much does Claude Code cost for an SDR? Anthropic's Pro plan is 17 dollars per month on annual billing and 20 dollars monthly, Max 5x starts at 100 dollars per month, and Team seats run 20 to 25 dollars per seat per month, per [Claude's pricing](https://claude.com/pricing) fetched 2026-07-08. For most reps a Pro or Max 5x seat is enough. That sits well below the third party estimates around 36,000 dollars per year for a managed AI SDR platform, and it keeps the config in the operator's hands.