Sales is the highest signal-to-noise category in the entire skill catalog. There are 1,300+ skills under sales/ and most of them solve the same three or four problems badly. This page is the opposite of a roundup — it's the eleven skills I actually keep installed on the laptops of every AE, SDR, and RevOps person I've worked with, plus the workflows that chain them, plus the anti-stack I rip out on day one.
If you're new to Claude Code skills: a skill is a SKILL.md file with YAML frontmatter that Claude loads contextually. Install one and Claude knows when to use it — you don't have to remember the trigger. Everything below assumes you've installed the Claude Code CLI and have the ~/.claude/skills/ directory ready. If not, start with the install guide and come back.
Every Claude Code skill category has filler. Sales has the most useful filler of any — because the underlying problems are well-defined and repeat at every company. There's a finite list of things a quota-carrying rep does in a week: prospect, write outreach, run discovery, send follow-ups, update CRM, handle objections, draft proposals, prep QBRs. That's it. Every skill in /category/sales/ is trying to compress one of those eight loops.
The signal-to-noise problem isn't that the catalog has too many bad sales skills — it's that the good ones look identical to the lazy ones from the outside. A skill called cold-email could be a 300-line system prompt with a working voice model and a banned-phrases list, or it could be three lines telling Claude to "write a cold email." The frontmatter is identical. The README usually doesn't help. You have to read the body.
The shortcut I use: a sales skill is worth installing if it does one of three things. First, it captures a non-obvious operational detail (the specific shape of a discovery question, the exact sequencing of a follow-up cadence, the calibration of a pricing objection rebuttal). Second, it integrates with a real system you already pay for — Apollo, Gong, Salesforce, HubSpot, LinkedIn Sales Navigator — and saves you from clicking through a UI. Third, it enforces a discipline you'd otherwise skip when you're tired on a Friday afternoon: the follow-up that needs to go out, the call notes that need to land in CRM before the day ends, the deal review checklist nobody actually runs.
If a skill doesn't do one of those three, it's a wrapper around "write a thing" — and you're better off with the raw model. The picks below all clear that bar, and I'll tell you which of the three they clear so you can decide whether the angle matches your motion. Field sales people will weight differently than inside SDRs, and RevOps will care about an entirely different subset. That's fine. The point of a stack isn't that you install all of it — it's that you have a coherent reason for each thing you install.
One more framing note: I'm avoiding any pick whose value proposition is "do outreach at higher volume." The volume game is lost. Buyers have triaged out of mass-personalisation theatre, and every spam filter on Earth has learned what an AI-written cold email looks like at scale. The skills below are calibrated for a working rep doing 30–80 thoughtful touches a day, not a sequencer doing 800.
Install order doesn't matter — Claude loads them all on session start. I've grouped them by job-to-be-done.
1. cold-email — The cleanest cold-email skill in the catalog. Trained on what actually gets replies in 2026: short, specific, no flattery, one ask. Use when you start any new outbound thread.
claude install cold-email"write a cold email to the VP of Eng at Acme Robotics, angle is our Postgres replication monitoring saved a similar team 11 hours during their last incident"2. cold-outreach-sequence — Multi-touch cadence builder for LinkedIn + email combined. Handles connection request → soft-touch comment → DM → email → second email. Crucial because the sequence is the product, not any single message.
claude install cold-outreach-sequence"build a 14-day sequence for series-B fintech CTOs, angle is async incident postmortems, voice should be peer-not-vendor"3. linkedin-outreach — LinkedIn-only when you don't want to mix channels. Takes a lead list from CSV or upstream skill, drafts personalised connection requests with one specific hook each, plus three follow-ups.
claude install linkedin-outreach4. sales-account-research — Single best account-research skill I've used. Pulls together 10-K snippets, news, leadership changes, and recent hires into a one-page account brief. Pair with an enrichment tool if you have one; works standalone with web search if not.
claude install sales-account-research"research Notion for an outbound pitch — focus on their Q1 layoffs, AI features shipped this quarter, and any signal on their data warehouse stack"5. cold-call-scripts — Five-step framework: pattern interrupt, permission, problem hook, micro-commitment, scheduling. Surprisingly hard to get right and this skill nails it.
claude install cold-call-scripts6. discovery-calls — Plan, run, summarise. The prompt asks the right questions before the call (what's the hypothesis on pain, who else is in the room) and produces a structured summary after.
claude install discovery-calls7. objection-crusher — Maps the specific objection to a calibrated response, with three escalation levels: acknowledge, reframe, redirect. Worth it for the framework alone.
claude install objection-crusher8. follow-up-emails — Post-call summary emails with action items and next-meeting confirmation. Boring on the surface; pays for itself the first week.
claude install follow-up-emails9. follow-up-radar — Scans your inbox for threads waiting on you, unanswered questions, and quiet deals. Trigger when you sit down Monday morning.
claude install follow-up-radar10. business-proposals — Proposals, SOWs, RFP responses. Has good defaults for Good-Better-Best pricing tiers and explicit exclusions sections — the part everyone forgets.
claude install business-proposals11. qbr-generator — Synthesises QBR decks from CRM data and customer-success notes. Better than most of the SaaS tools in this space, and it lives where you already work.
claude install qbr-generatorThis is the loop that runs the most often, so it's worth getting right. The mistake most teams make is treating these as separate steps with separate skills. They're not — they're one workflow with three checkpoints.
Step 1: Research before you draft. Don't open the email composer first. Open Claude and run the research skill:
"use sales-account-research on Acme Robotics, I'm targeting their VP of Eng about Postgres replication monitoring"The output should be a one-page brief: company stage, recent funding, three news items from the last quarter, who reports to whom in the eng org, and any public signal about their data stack. If you can't get three concrete hooks from this brief, the account isn't ready and you should move on. Don't try to pad weak research with clever copy — it always reads as weak research with clever copy.
Step 2: Draft the first touch. Now invoke cold-email with the brief in context:
"write a cold email to Sarah Chen, VP Eng at Acme, using the brief above. Angle: their incident on March 14 (mentioned in their engineering blog) is exactly the kind of thing our replication monitoring would have caught 40 minutes earlier. Keep under 90 words, one ask: 15 minutes next Tuesday."Read the output. If it sounds like an AI wrote it, push back: "this is too smooth, give me something rougher and more direct, lose the second paragraph." Good cold-email skills respond to this — they have a voice register you can dial. If yours doesn't, replace it.
Step 3: Build the cadence around the first touch. Now cold-outreach-sequence takes over:
"build a 14-day cadence around the email above. Touch 2 is a LinkedIn comment on something Sarah posted, touch 3 is a short bump email referencing a second incident from their blog, touch 4 is a soft-close — be specific about timing."The cadence skill should produce dated touches, not just "day 3, day 7, day 14" — translate them into your calendar so you actually fire them. I keep this in a single text file per prospect.
Step 4: When they reply, switch skills. The moment a prospect engages, cold-email stops being the right tool and discovery-calls takes over for booking, then follow-up-emails takes over for the post-meeting note. Don't keep using a cold-outreach skill to write a warm follow-up — the voice is wrong and it shows.
Step 5: Triage quiet threads weekly. Friday afternoon, run:
"follow-up-radar — show me everything I owe a response on, plus deals where I haven't touched the prospect in 7+ days"This is the single highest-ROI 15 minutes in a rep's week. Most pipeline doesn't die from rejection — it dies from being forgotten.
Discovery is where the catalog actually beats human discipline. The honest truth is that most reps don't run good discovery — they run a checklist of qualifying questions and call it discovery. The skill stack here is designed to force a deeper conversation and then capture it cleanly.
Pre-call: plan the hypothesis. Twenty minutes before the call, use discovery-calls to plan:
"I'm meeting Acme tomorrow at 2pm. They're a series-B fintech, 80 engineers, current pain seems to be incident response time but I'm not sure that's the real driver. Help me plan the hypothesis tree and the three questions that disambiguate it."The output should give you a working theory (most likely the surface pain is downstream of a deeper organisational problem — incidents are slow because nobody owns runbooks, or because oncall is rotating too fast for muscle memory to build) and the questions that test the theory. Bring the questions to the call. Don't bring the theory — you want to update it during the conversation, not anchor on it.
During the call: take notes, not minutes. Capture verbatim what the prospect said about pain, timeline, who else is involved, what they've tried. Don't paraphrase. Verbatim quotes are gold for the follow-up email because they let you reflect their language back to them, which builds trust faster than any other rhetorical move.
Post-call: structured summary. Paste raw notes into Claude and trigger discovery-calls again:
"summarise this discovery call using the BANT-plus-narrative format. Pull out direct quotes for pain points, flag the three biggest unknowns I still have, and suggest the two highest-leverage questions for the next meeting."This produces a CRM-ready summary plus a prep doc for next time.
CRM update: don't trust the summary alone. Use follow-up-emails to draft the customer-facing recap and then push the structured fields to CRM manually. I know — automation should handle this. In practice, CRM auto-fill from AI summaries hallucinates close dates and inflates deal sizes, and you'll end up cleaning it up in pipeline review anyway. Manual CRM update with AI-drafted email is the right division of labour.
Next-meeting prep: use the summary as the brief. When meeting two comes around, feed the prior summary back to discovery-calls and ask it to update the hypothesis tree. The skill should track what's changed, what new unknowns surfaced, and what the prospect's energy was around each topic. Good reps build a mental model of the buying committee over multiple meetings; this loop externalises that model so it's actually retrievable.
RFPs are the worst use of a senior AE's time on the planet, which means the AI lift here is enormous. The trick is not to use AI to write the answers — AI-written RFP responses read as AI-written RFP responses, and procurement teams have been calibrating against them since 2024. Use AI to do the assembly and positioning work that consumes the hours, then human-write the parts that matter.
Step 1: Inventory the questions. Drop the RFP PDF into Claude and ask business-proposals to parse it:
"parse this RFP. Categorise every question into: factual (we have a canonical answer), positioning (we need to differentiate vs competitors), or judgement (the answer depends on which deal version they're imagining). Flag any question where the answer is genuinely uncertain."This sorting alone saves you 2-3 hours. Factual questions get answered from a content library (your security questionnaire, SOC 2 report, integration list). Positioning questions need real thought. Judgement questions need a conversation with the prospect before you answer.
Step 2: First-pass draft for factual. For the factual category, let the skill draft from your existing content. Provide a paste of your security questionnaire and the standard integration list:
"draft answers to the factual questions using the canonical content I just pasted. Match their question phrasing exactly so the eval team can map answers fast. Flag anywhere you're filling a gap with placeholder text."The placeholder flags are the most important output. They tell you what's actually missing from your content library.
Step 3: Positioning pass. This is where objection-crusher earns its install. For each positioning question, ask:
"this RFP question — 'describe how your solution handles X vs alternative approaches' — is a thinly-veiled competitive question. The likely incumbent is CompetitorY. Use objection-crusher framing to draft an answer that doesn't name them but anchors on the two dimensions where we win clearly: $DIMENSION_1 and $DIMENSION_2."The skill should produce three versions: acknowledge-and-reframe (most diplomatic), direct-comparison-without-naming (medium), and aggressive-anchoring (most assertive). Pick the register that fits the deal stage. Early evaluation → diplomatic. Procurement-stage → assertive.
Step 4: Pricing structure. business-proposals has good defaults for Good-Better-Best tiers. Push it to produce three options where Best has explicit risk-reduction language (assumptions, exclusions, termination clauses) and Good has explicit limits (this tier excludes X, doesn't cover Y). The exclusions section is the part 90% of proposals skip, and it's the part that prevents fights at renewal.
Step 5: Human pass on the executive summary. Whatever you do, don't let AI write the executive summary. That's the one section the buying committee actually reads. Use the skill to assemble a list of the top three differentiators with supporting evidence, then write the prose yourself. Twenty minutes well spent.
For every skill above, there are five in the catalog that look superficially similar and will actively hurt your numbers. Here's what I uninstall on sight.
Spammy sequencers. Any skill whose pitch is "send 500 personalised emails a day" is selling the volume game, which is over. The skills to avoid usually have at-scale, volume, mass, or blast in the slug or description. Mass-personalisation outputs are pattern-recognised by every modern spam filter and by every prospect who's been on the receiving end. They will hurt your domain reputation and your reply rate in the same week.
Over-eager personalisation gimmicks. The "I noticed you posted about X on LinkedIn 6 minutes ago" opener was clever in 2023 and lethal in 2026. Prospects know it's automated and have been trained to ignore anything that opens this way. Skills that pitch "hyper-personalised" openers based on social signals are mostly producing this output. The honest replacement is: open with the reason you're reaching out, in your own voice, with the specific operational hook. No flattery, no "I noticed."
Anything that breaks human voice. Some skills try to enforce a corporate tone — "professional," "polished," "executive register." This produces the worst possible voice for outbound: a generic mid-Atlantic accent that reads as nobody in particular. The cold-email skill in the picks above is calibrated for a specific recognisable voice register; if your skill produces output that could have come from any rep at any company, replace it.
Skills that automate CRM updates from call audio. The pitch is great — record the call, AI transcribes it, AI updates Salesforce. In practice the AI consistently inflates deal size, hallucinates close dates, and miscategorises stage transitions. RevOps then spends Friday cleaning it up. The right pattern is: AI drafts the summary email, human pushes structured fields to CRM. Don't automate the field updates.
Anything claiming to coach reps based on call sentiment. The sentiment models are not good enough to coach off. They confuse a prospect who's enthusiastic about a different vendor with one who's enthusiastic about you. They miss sarcasm 60% of the time. They flag false-positive risk signals that put deals through unnecessary save motions. gong-conversation-intelligence exists because the underlying Gong product is real; an AI-only "sentiment coach" skill without a real platform behind it is not.
Single-message generators. Anything that produces one email at a time with no cadence context. Sales is a cadence game. A skill that doesn't think in sequences is solving the wrong problem.
Skills that don't disclose what they don't know. Good skills have a "when not to use" section and a list of things they can't do. Skills that promise to solve every sales problem from prospecting to closing are usually thin wrappers around a single prompt. Check the SKILL.md body. If there's no scope discipline, there's no skill.
The stack above is a starting point, not a destination. The reps who get the most out of Claude Code build their own skills for the specific shape of their motion within six months. Three signals tell you it's time.
Signal 1: You're rewriting the same prompt context every session. If every time you ask Claude to draft an email, you're pasting in the same paragraph about who your ICP is, what your three differentiators are, and the voice register you want — that paragraph belongs in a SKILL.md file. Five minutes of writing your own skill will save you ten minutes a day forever.
Signal 2: The off-the-shelf skill is 80% right and you keep correcting the other 20%. This is the classic "close but wrong" pattern. cold-email might produce great outputs but always default to a sign-off you don't use. follow-up-emails might always lead with "Hi {{name}}, great chatting today" when your style is direct. Fork the skill, change the parts that bother you, and use your fork. Most well-written skills are explicitly forkable — the SKILL.md is short enough to read in five minutes.
Signal 3: You have a workflow that doesn't exist in the catalog. Maybe you run a very specific deal review format every Thursday. Maybe your team uses an unusual qualification framework that's not BANT, MEDDIC, or any of the named variants. Maybe you have a proposal template with 14 specific sections and the catalog skills only know about Good-Better-Best. These are exactly the cases where a 200-line SKILL.md will dramatically outperform any generic skill. The skill doesn't need to be sophisticated; it needs to be specific to your motion.
Writing your own is not as scary as it sounds. The frontmatter is just a name, a description, and the trigger conditions. The body is a Markdown document that tells Claude what to do, what not to do, and what good output looks like. The two best techniques: include 2-3 concrete examples of your own past work as exemplars, and write an explicit "when not to use this skill" section. Both massively reduce the rate of bad outputs.
One thing to avoid when writing your own: don't try to encode every possible variation. A skill that handles "all sales emails" will be worse than a skill that handles "first-touch cold emails to VPs at series-B fintech companies." Specificity is the friend. Build narrow skills and chain them — that's the architectural insight that makes the catalog approach work at all.
If you build something useful, the catalog accepts submissions. There's a real chance your motion-specific skill helps another team with the same shape of motion, and the cross-pollination is what makes the catalog get better over time.
If you run RevOps, you're not just thinking about individual rep productivity — you're thinking about consistency across the team, pipeline hygiene at scale, and what shows up in the forecast. The stack above can support all three, but the deployment pattern is different from how an individual contributor would use it.
Standardise the skill list, not the prompts. The temptation is to write team-wide prompts and force every rep to use them. This backfires — reps lose the muscle of articulating intent to Claude, and the prompts get stale. The better pattern is to standardise the installed skill set (everyone has the same eleven skills above) and let each rep develop their own prompt habits within those skills. The skill's SKILL.md provides the rails; the prompt provides the steering. Your job is to maintain the rails.
Audit installed skills quarterly. Skills go stale. Catalog skills get rewritten, sometimes for the worse. New picks emerge that obsolete old picks. Once a quarter, do a 30-minute audit: which skills are people actually using, which have been forked into local versions, which produce output that's drifted away from your house style. The Claude Code CLI has an installed-skill list — run it on a representative laptop, not on your own.
Forecast hygiene depends on what reps don't automate. The anti-stack section above isn't aesthetic. Skills that auto-update CRM from call audio degrade pipeline data quality and your forecast quality with it. The RevOps win is enforcing the boundary: AI drafts the customer-facing artefact, human pushes structured data to CRM. This is unpopular with reps who want full automation, but the forecast accuracy gain is worth the friction.
Build one team-specific skill in-house. The leverage move is to write one custom skill that encodes your specific qualification framework, deal stages, and forecast categories. Call it deal-review or pipeline-check. It should take a CRM record (or a paste of one) and produce a structured assessment matching your team's framework exactly. This is the skill nobody else can write for you, and it pays back enormously in pipeline review meetings.
QBR season: lean on qbr-generator hard. The QBR skill in the stack above is genuinely good and worth the prep time. Feed it CRM data plus customer-success notes, get a first draft, and spend your time on the narrative and the recommendations rather than the data assembly. The hours saved here are RevOps hours, which makes them strategic hours.
One thing to watch. Some of the catalog skills assume specific platforms (Apollo, Gong, Salesforce, HubSpot). If your tech stack doesn't include them, those skills will produce confidently-wrong output that references features you don't have. Either replace the skill with a platform-agnostic alternative, or fork it and strip out the platform-specific assumptions. Don't ignore the mismatch — it will surface as confused reps a week later.
Found a bug or want a topic covered? Email [email protected] or open an issue via GitHub.
SKILL.md files, not affiliated with, endorsed by, or sponsored by Anthropic.