If you lead a sales team in 2026, your reps already have Claude open all day. They use it to draft follow-up emails, summarize call transcripts, prep for cold calls, and figure out what to say when a prospect raises an objection they have not heard before. The unspoken truth in most sales orgs: Claude is now the primary interface, and the official enablement portal is the secondary one.
Most CI and enablement tools have not adjusted. They keep publishing battlecards as Notion pages, Highspot tiles, or PDFs that live on a SharePoint somewhere. Adoption metrics look fine for the first month and rot from there. The system fails not because the content is bad but because the surface is wrong.
Anthropic's Claude skill format, plus the Model Context Protocol, fixes this. A skill is a markdown file with structured frontmatter that Claude reads as 'when the user says X, run this workflow.' Combine that with an MCP server that exposes your real CI data, and your battlecards live inside Claude. The rep does not switch tabs. The PMM does not chase adoption. The data writes itself back.
What a Claude skill actually is
A Claude skill is a markdown document with a YAML preamble and a body. The preamble has two fields:
- name: a short identifier, kebab-case.
- description: a single paragraph that tells Claude when to invoke the skill. This is the trigger. Phrasing matters: this is the text Claude matches against in user messages.
The body is the workflow. It tells Claude what tools to call (typically MCP tools), in what order, with what arguments. It also includes the quality bar, what to refuse, and what to ask the user to confirm before writing anything.
Skills are installed once and live in the Claude.ai or Claude Code skill panel. Anthropic publishes them in a directory, and any team can submit first-party skills for the connectors they ship. The format is open and the bar is editorial more than technical.
The three skills a sales team needs
Across enough deployments, the same three workflows account for ninety percent of competitive value a sales rep extracts from AI. We built and ship them as first-party Clinch skills.
1. Pre-call brief
The rep has a competitive call coming up. They want a one-pager grounded in the actual battlecard, the prospect's homepage, and their own recent record against the competitor in the deal.
The skill triggers on phrases like 'I have a call in 20 minutes with Acme, evaluating us vs Klue,' 'prep me for the call,' or detected upcoming calendar events. It composes with Google Calendar or Microsoft 365 MCP connectors to read the next event automatically. It calls the CI tool's generate_deal_brief tool, which produces a structured one-pager: summary, threat assessment, prospect signals, why we win, talk tracks, questions to ask, landmines.
Why this matters: the rep reads it in 60 seconds in the same window they were already in. No tab switch, no PDF, no Sunday-night printout.
2. Post-call capture
The hardest part of any competitive intelligence system is keeping the win/loss data current. Reps will tell you they 'log everything in Salesforce' and they sort of do, but the competitive context is the field that gets the least love. This is the single most important field, because it feeds your battlecards, your VP's win-rate review, and your PMM's positioning work.
The skill triggers on phrases like 'we lost that to Klue, integration story was thin' or 'Klue came up on the pricing question.' It reads the user's words, extracts the competitor, the outcome (won, lost, no decision), and the primary reason verbatim. It confirms the parse with the rep in one short sentence, then writes back to the CI tool via log_deal_outcome or log_competitor_mention.
Two design choices matter: never invent a reason the rep did not say (parse failures are better than fabrication), and always use idempotency keys so an accidental retry replays the prior result rather than duplicates the row.
3. Deal coach
This is the workflow VPs of Sales love most because it answers the question they ask every week: how is each rep doing against each competitor? The skill triggers on 'how am I doing against Klue?', 'what's my win rate?', or 'what should I say when they push back on pricing vs Crayon?'
It calls the CI tool's get_my_competitive_record (a per-rep rollup keyed by user_id, not org-wide aggregates) and get_battlecard. The response leads with the rep's own numbers, names the recurring loss reason, and offers to coach with live talk tracks from the battlecard. It is the difference between competitive enablement-as-PDF and competitive enablement-as-coach.
Why write-back is the unlock most CI tools missed
Until recently, the AI side of competitive intelligence was strictly read-only. Claude could query your battlecard, your timeline, your pricing history. It could not write back. The asymmetry meant every workflow ended the same way: 'and then the rep logs in to the CI dashboard to record the deal outcome.' That sentence killed adoption.
Write-back means Claude can also call log_deal_outcome, log_competitor_mention, request_brief, generate_deal_brief, flag_battlecard_section, and add_competitor. Every write is scoped to the OAuth token (a separate write scope, approved explicitly by the user on consent), attributed to the calling user, and recorded in an audit log. Idempotency keys make retries safe. The CI tool's dashboard becomes a viewing surface; the writing surface is the chat.
If you are evaluating a CI tool in 2026 and it does not expose write tools over MCP, you are buying yesterday's product. The reps will not log into a dashboard to keep your competitive memory alive. They will use the tool that meets them where they already work.
What this looks like in practice
A representative day for a rep at a company that has installed all three skills:
- Morning: opens Claude. Asks 'what does my afternoon look like?' Calendar MCP returns events. One of them is a call with a prospect evaluating Klue. The pre-call brief skill triggers automatically and drops a one-pager into the chat for that meeting.
- Right before the call: re-reads the brief. Goes into the call confident.
- Right after the call: types one sentence to Claude. 'We lost to Klue, their integration story was tighter than ours.' Post-call capture skill logs it back to Clinch with auto-categorized reason tags. The audit trail is intact.
- Mid-afternoon: prepping for a coaching session with their manager. Asks Claude 'how am I doing against Klue this quarter?' Deal coach skill returns their personal record, their recurring loss reason, and an offer to coach on the gap.
- Evening: no admin work. The dashboard filled itself.
Notice what is missing from that day: switching tabs, opening a CI dashboard, copy-pasting data between tools, or typing the same lost-deal reason into three different systems. Notice what is present: actual prep, actual conversations, actual records. The skill format collapses the difference between 'doing the work' and 'recording the work.'
How to evaluate skills for your sales team
When you read a candidate skill (yours or a vendor's), check four things:
- Trigger discipline: is the description specific enough that Claude will not fire the skill on unrelated chat? Vague triggers mean noise.
- Quality bar: does the body of the skill say what the skill refuses to do? A skill that fabricates a deal_size when the rep did not state one is worse than no skill.
- Composability: does the skill explicitly call out which other connectors (Calendar, Krisp, Zoom, CRM) it pairs with? Skills are most powerful when they orchestrate across connectors the rep already has.
- Write safety: does the skill use idempotency keys on every write? Without them, accidental retries create duplicate rows and corrupt the data.
The Clinch skills under skills/ in our repo are open. Read them as templates for your own.
The bottom line
Sales enablement in 2026 is no longer 'where do we put the battlecard.' It is 'how do we put the right answer in front of the rep at the moment they need it.' Claude is now the moment. Skills plus MCP write tools are the mechanism. The CI tools that figure this out get adopted; the ones that do not get cancelled at renewal.
Frequently asked
Questions worth answering
- Are Claude skills the same as plugins or extensions?
- Closer to plugins than extensions. A Claude skill is a markdown file with structured frontmatter Claude treats as a workflow trigger. They are open format, installable in Claude.ai or Claude Code, and Anthropic maintains a public skill directory you can submit to.
- Do skills work with ChatGPT or only Claude?
- The skill format is Anthropic's. Other AI workspaces have similar concepts but distinct formats. Practically, in 2026, Claude's skill ecosystem is the most mature for B2B SaaS workflows.
- Can our reps build their own skills?
- Yes, and they should. Anyone on the team can write a markdown file with the YAML preamble and install it. We ship the three above as first-party skills because they are the workflows we expect every sales team to need; teams that want a fourth (e.g. industry-specific objection handlers) can add it locally.
- How does write-back affect security?
- Write tools require an explicit write OAuth scope approved on the consent screen. Tokens are user-bound and org-scoped. Every write is recorded in an agent_writes audit log with the calling user, the tool name, the payload, and the result. You can audit and undo via SQL today; a UI surface follows.
Stop writing battlecards from scratch
Clinch does the CI work so you can do the rest of your job.
Snapshots in under two minutes. Auto-updating battlecards. Weekly digests your team will actually read. 79 dollars per month.