zime-gtm-skills

If you’re building call-review or deal-inspection tooling in-house, start here. 45 Agent Skills that audit sales call transcripts and CRM exports against GTM rubrics: one file in, one evidence-cited audit out. Runs locally, no credentials, no data leaves your machine. Works with Claude Code, Cursor, Windsurf, OpenAI Codex, or any agent that reads the Agent Skills spec.

What sets these skills apart: they credit a call only for what the buyer actually said, and each one is built around what reps usually skip at that deal stage. A rep claim never counts as evidence. A topic the call never touched scores Missed, not “probably fine”. Every dimension lists what does not count (price worry is not a budget, a date for the next meeting is not a decision date, a job title is not authority) and gives one strong-call question the rep can ask next time.

Two axes: deal stage (discovery through renewal) and initiative (MEDDICC, BANT, pain identification: frameworks that run at any stage).

The rule every skill here is built on: an uncited finding doesn’t ship. Every dimension a skill marks Covered, Partial, or Missed points at a direct quote or timestamp from the transcript in front of it, never a plausible guess dressed up as a finding.

MIT, no gating, no telemetry, no lead capture. Fork it, rewrite the rubrics for your motion, ship it internally: that’s the intended use. Built and maintained by Zime; we’d rather hand over the rubrics than argue anyone out of building.

Contributions welcome: see CONTRIBUTING.md. Found a bug or have a question? Open an issue.

• See it work
• Quick install
• How these fit together
• Available skills
• Repository structure
• Where this stops
• FAQ
• Installation
• Built by Zime
• License

See it work

Real output from three skills, one per output shape, each run against its own bundled sample data: nothing invented for this README. Skim the tables; the full run (quotes and all) is one click away in each `

Invoke it by name, like any Claude Code skill:

/deep-discovery skills/deep-discovery/assets/sample-transcript.txt

CSV pipeline-sweep: deal-risk-digest

Without the skill: eyeball 8 CRM export rows by hand, no ranking, no consistent signal, no tiebreak logic.

With deal-risk-digest against skills/deal-risk-digest/assets/sample-pipeline-deal-risk.csv: ranked, most at-risk first, every flag cited to the cell behind it.

Closing read: 2 of 8 deals rated High risk. The most common signal across flagged deals is past-due-and-unlikely (a close date already behind today, paired with under-50% probability).

/deal-risk-digest skills/deal-risk-digest/assets/sample-pipeline-deal-risk.csv

Document-writer: champion-tracker

Without the skill: a rep’s own notes just say “good engagement from both stakeholders,” no way to tell who’s actually selling internally vs. who just likes the product.

With champion-tracker against skills/champion-tracker/assets/sample-call-1.txt and sample-call-2.txt: one cited action ledger, one verdict per contact.

Champion read: Priya ✅, 3 confirmed Action rows by call 2. Devon ⚠️, every row across both calls is Sentiment: a friend, not a champion. Trend: strengthening, an unconfirmed plan in call 1 became 3 delivered actions in call 2.

/champion-tracker skills/champion-tracker/assets/sample-call-1.txt skills/champion-tracker/assets/sample-call-2.txt

Try any other skill

Same pattern, every skill: clone once, then invoke any skill by name against its own bundled sample.

git clone https://github.com/zime-ai/zime-gtm-skills.git
cd zime-gtm-skills

Then, inside Claude Code or Claude Desktop:

/ skills//assets/

Every skill ships its own sample data and documents the exact filename(s) and command in its SKILL.md’s “Sample data” section. Click through from the skill’s row in Available skills below.

Quick install

# Claude Code plugin (recommended)
/plugin marketplace add zime-ai/zime-gtm-skills
/plugin install gtm-skills@zime-gtm-skills

# or clone and copy into any agent that reads the Agent Skills format
git clone https://github.com/zime-ai/zime-gtm-skills.git
cp -r zime-gtm-skills/skills/* .agents/skills/

See Installation below for the submodule option and the .claude/skills/ vs .agents/skills/ distinction.

How these fit together

45 skills: 24 stage motions laid out across the deal lifecycle, 12 cross-stage initiative skills, and 9 deal-intelligence skills that write something forward — a plan, a digest, a drafted email — instead of grading a call.

New business

┌────────────────┐     ┌────────────────┐     ┌────────────────────┐
│ sql-to-qualify │ ──▶ │ deep-discovery │ ──▶ │ meeting-to-qualify │
└────────────────┘     └────────────────┘     └────────────────────┘
                                                         │
                                                         ▼
┌─────────────────────┐     ┌──────────────┐     ┌────────────────┐
│ technical-discovery │ ◀── │ improve-demo │ ◀── │ qualify-to-poc │
└─────────────────────┘     └──────────────┘     └────────────────┘
           │
           ▼
┌─────────────────────┐     ┌─────────────┐     ┌─────────────────────┐
│ pilot-to-conversion │ ──▶ │ poc-success │ ──▶ │ negotiation-closing │
└─────────────────────┘     └─────────────┘     └─────────────────────┘
                                                           │
                                                           ▼

                                                     Post-sale

                                                ┌────────────────────┐     ┌──────────────────┐
                                                │ onboarding-journey │ ──▶ │ customer-success │
                                                └────────────────────┘     └──────────────────┘

Runs alongside, not in the chain:
  • persona-based-discovery — persona-adapted lens on discovery
  • first-call-rampup       — coaching, at any stage
  • churn-prevention        — renewal risk, post-sale
  • upsell-expansion        — expansion opportunity, post-sale

Unlike some agent-skill collections, there’s no foundational context skill every other one depends on, and no required order. Each skill is independent: one call transcript (or CRM export) in, one evidence-cited audit out.

Company context (optional)

Every skill runs fine with zero setup — it reads whatever file you point it at and nothing else. But without any company context, a skill has no way to know your ICP, your product’s actual differentiators, or which competitor names to expect, so it audits each call in a vacuum.

If a file named .agents/gtm-context.md or .claude/gtm-context.md exists in your working directory, every skill in this repo reads it first and doesn’t ask for anything it already answers — no flag, no config, just drop the file in and it’s picked up automatically. See gtm-context.example.md for the shape: ICP, product/pricing basics, named competitors, and internal terminology or stage names that differ from the generic ones a skill assumes by default.

This file is yours, never the repo’s — it isn’t tracked or read by anything except the skill itself at run time, and nothing in it leaves your machine beyond whatever your agent’s own model call already sends.

Available skills

Flagship skills (hand-written, with the buyer-said evidence rules above): deep-discovery, meddicc, bant, technical-discovery, meeting-to-qualify, improve-demo, follow-up-email-drafter, deal-risk-digest, next-step-commitment, win-loss-brief. The other skills in the list are community preview: they work and pass validation, and their rubrics are still being rewritten to the same standard.

All 24 stage motions are covered; the initiative set is 12 skills today. New skills welcome, see CONTRIBUTING.md. For how these are tested, and what that testing can and can’t prove, see EVALS.md.

Repository structure

zime-gtm-skills/
├── .claude-plugin/
│   ├── marketplace.json       # /plugin marketplace add zime-ai/zime-gtm-skills
│   └── plugin.json
├── skills/
│   └── skill-name/
│       ├── SKILL.md          # required
│       ├── references/       # optional, the rubric lives here
│       ├── assets/           # synthetic sample transcript/CSV
│       └── evals/            # declarative evals.json
├── validate-skills.sh         # local, zero-dep frontmatter/layout check
├── scripts/                    # check-docs-sync.sh, scan-content.py, pr-prep.sh
├── tests/                      # fixture tests for the validators above
├── evals/                       # eval methodology assets, see EVALS.md
├── CONTRIBUTING.md
├── MAINTAINING.md              # process + decisions, for picking this repo back up
├── ROADMAP.md                  # what's built, what's next
├── LICENSE
└── README.md

Where this stops

Everything here runs on one file, in one session, with no memory. That’s a design ceiling, and it’s worth knowing before you build on it:

• One call at a time. You point a skill at a transcript. Auditing every call an org makes, as it happens, is a pipeline problem, not a rubric problem.
• Snapshot, not trend. Each run is independent. Nothing here tracks the same rubric across a rep’s quarter, a team, or a pipeline over time.
• Reads exports, never writes back. A CRM .csv goes in. Nothing updates a deal record or feeds a forecast.
• Rubrics are hand-written, not calibrated. These dimensions come from experience and community correction, not from regression against your own won/lost data. A sound default, not a model fit to your motion.
• No rep or manager surface. This is an engineer running a command. Assignment, coaching loops, adoption tracking are out of scope.

Past that ceiling you’re building a platform, not a skill, and that’s a much larger piece of work. It’s the problem Zime works on.

FAQ

Do I need an API key or any credentials? No. Every skill is a markdown rubric your agent reads and applies to a file you already have. Nothing here calls out to a service, and nothing you feed a skill leaves your machine beyond whatever your agent’s own model call already sends.

My agent isn’t picking up the skill. Claude Code reads .claude/skills/, not .agents/skills/ — if you installed via Option 2 or 3 below for Claude Code specifically, copy (or symlink) into .claude/skills/ instead. For other agents, confirm they read the Agent Skills spec format and that the skill directory is where that agent expects it.

A skill’s output doesn’t look right. If it’s the rubric itself — a missed dimension, a wrong classification — that’s the most useful kind of issue, see CONTRIBUTING.md. If it’s install or tooling breaking, open a bug report.

Installation

Option 1: Claude Code plugin

/plugin marketplace add zime-ai/zime-gtm-skills
/plugin install gtm-skills@zime-gtm-skills

Option 2: Clone and copy

git clone https://github.com/zime-ai/zime-gtm-skills.git
cp -r zime-gtm-skills/skills/* .agents/skills/

Or copy just the one skill you want: cp -r zime-gtm-skills/skills/deep-discovery .agents/skills/.

[!TIP] Claude Code reads .claude/skills/, not .agents/skills/ — see the FAQ if your agent isn’t picking the skill up.

Option 3: Git submodule

git submodule add https://github.com/zime-ai/zime-gtm-skills.git .agents/zime-gtm-skills

Update later with git submodule update --remote.

Built by Zime

Zime builds sales-call analytics; this repo is the rubric layer we’d rather hand over than gatekeep. If it’s useful, a star helps other GTM/RevOps teams find it. Business inquiries: [email protected].

License

MIT, see LICENSE.

Maintained by Zime.