Back to the catalog

crm-workflows

36 CRM-neutral, evidence-first sales workflows built on the Agent Skills standard.

Open source Open in the app JSON README (API)

About

36 CRM-neutral, evidence-first sales workflows built on the Agent Skills standard.

Details

Kind
Plugins
Topic
Productivity
Publisher
sepivip
Origin
gemini
Category
ferramentas
Version
0.1.0
Last push
2026-08-27T10:26:21Z
Repository state
ativo
Language
Python
License
MIT
Added
2026-08-30 14:13:39
Updated
2026-08-30 14:13:39
Origin id
sepivip/crm-workflows

README

<p align="center">
  <img src="./assets/logo.svg" alt="CRM Workflows — 36 evidence-first skills for safer revenue work" width="900">
</p>

<p align="center">
  <a href="https://github.com/sepivip/crm-workflows/actions/workflows/validate.yml"><img alt="Validation" src="https://github.com/sepivip/crm-workflows/actions/workflows/validate.yml/badge.svg"></a>
  <a href="https://github.com/sepivip/crm-workflows/releases/tag/v0.1.0"><img alt="Release" src="https://img.shields.io/github/v/release/sepivip/crm-workflows?color=6C5CE7"></a>
  <a href="https://agentskills.io"><img alt="Agent Skills" src="https://img.shields.io/badge/standard-Agent%20Skills-17152F"></a>
  <a href="./LICENSE"><img alt="MIT License" src="https://img.shields.io/badge/license-MIT-54BFA9"></a>
  <img alt="36 skills" src="https://img.shields.io/badge/skills-36-6C5CE7">
</p>

<p align="center"><strong>Portable CRM judgment for revenue agents—without hard-coding a CRM vendor or silently taking action.</strong></p>

CRM Workflows is a public collection of 36 sales and revenue-operations skills built on the open [Agent Skills specification](https://agentskills.io). The same core instructions can be packaged for Codex and ChatGPT, Claude Code, Gemini CLI, GitHub Copilot, or another compatible agent.

The workflows are intentionally evidence-first. They distinguish facts from inference, preserve ambiguity, cite consequential claims, minimize sensitive-data access, and require review before writes, sends, scheduling, assignments, or other external side effects.

The shared [CRM-neutral semantic model](./references/crm-neutral-model.md) explains how provider-specific records map into those workflows.

## What you get

| Area | Skills | Examples |
| --- | ---: | --- |
| Plan and pipeline intelligence | 8 | Daily briefing, pipeline review, forecast narrative, slip scenarios |
| Deal strategy and execution | 6 | Deal review, stakeholder map, close plan, opportunity updates |
| Account growth and prospecting | 10 | Account plan, lead triage, renewal radar, expansion whitespace |
| Meeting and communication productivity | 6 | Call prep, transcript summary, inbox sweep, follow-up |
| Governance, activity, and data hygiene | 6 | Win/loss review, activity history, internal answers, hygiene audit |

Browse the complete [36-skill catalog](./docs/CATALOG.md).

## Install

### OpenAI Codex and ChatGPT

```bash
codex plugin marketplace add sepivip/crm-workflows
codex plugin add crm-workflows@crm-workflows
```

Restart the host if it asks you to refresh installed plugins.

### Claude Code

```bash
claude plugin marketplace add sepivip/crm-workflows
claude plugin install crm-workflows@crm-workflows
```

For local development, clone the repository and run `claude --plugin-dir ./crm-workflows`.

### Gemini CLI

```bash
gemini extensions install https://github.com/sepivip/crm-workflows
```

### GitHub Copilot or another Agent Skills host

Clone the repository, then copy all 36 skills—or only the ones you want—into a discovery directory supported by your agent:

```bash
python scripts/install_skills.py --target .agents/skills
python scripts/install_skills.py --target .agents/skills --skill call-prep --skill deal-review
```

The installer refuses to overwrite an existing skill directory. See the [compatibility guide](./docs/COMPATIBILITY.md) for personal-scope locations and invocation details.

## Use

Ask naturally when your host supports automatic skill selection:

```text
Brief me for tomorrow's Acme security review.
What materially changed across my open deals this week?
Turn these call notes into a reviewed follow-up package.
Audit this opportunity export for hygiene issues—but do not change anything.
```

Or invoke a workflow explicitly when you want deterministic selection:

```text
Use $call-prep for tomorrow's Acme security review.
Use $deal-review on the Globex renewal.
Use $crm-hygiene-check on this export and remain read-only.
```

Claude Code exposes installed plugin skills as `/crm-workflows:call-prep`, `/crm-workflows:deal-review`, and so on. Invocation syntax varies by host; the skill logic does not.

## The operating contract

Every workflow follows the same safety model:

```text
Scope → Resolve identity → Read minimum data → Normalize semantics
      → Separate evidence / inference / conflict / unknown
      → Draft a reviewable result → Confirm → Act once → Report exactly
```

- CRM-neutral semantics: organization, person, lead, opportunity, activity, task, owner, stage, value, close date, and forecast category.
- Evidence over confidence theater: unsupported probabilities and invented facts are prohibited.
- Least-privilege retrieval: private or sensitive data is excluded unless necessary and authorized; sensitive personal data is never used for prospecting.
- Review-gated mutations: proposals show the target, field-level diff, evidence, and side effects before execution.
- Honest failure handling: no silent retries, no fabricated success, and no more than one bounded retry for a failed write.

The repo does **not** bundle a CRM connector or credentials. An agent can analyze supplied exports immediately; live retrieval and mutations require tools that the host exposes and the user authorizes.

## Quality and evaluations

The repository ships 119 generated evaluation cases and a fictional CRM fixture:

- 36 behavioral cases—one per skill
- 36 implicit-routing cases
- 36 incomplete-input cases
- 8 high-risk safety cases
- 3 unrelated non-trigger cases

Run the full structural suite:

```bash
python -m pip install -r requirements-dev.txt
python scripts/generate_skills.py --check
python scripts/validate_crm_plugin.py
python scripts/run_evals.py
```

Export self-contained JSONL cases for your preferred agent harness:

```bash
python scripts/run_evals.py --group behavioral --skill deal-review --emit-jsonl
```

The validator checks the official Agent Skills schema, all provider manifests, generated-file drift, UI metadata, privacy boundaries, fixture references, and complete evaluation coverage. Behavioral model outputs still require an external agent harness or human grading; the repository does not claim model-independent semantic pass rates.

## Build from the source catalog

`scripts/generate_skills.py` is the reviewed source of truth for skill content and generated tests:

```bash
python scripts/generate_skills.py
```

This regenerates every `SKILL.md`, every OpenAI UI manifest, the evaluation catalog, and the human-readable skill catalog. Read [CONTRIBUTING.md](./CONTRIBUTING.md) before proposing a workflow change.

## Project map

```text
skills/                         36 portable Agent Skills
  <skill>/SKILL.md              provider-neutral workflow
  <skill>/agents/openai.yaml    optional Codex/ChatGPT UI metadata
evals/                          routing, behavior, safety, and non-trigger cases
references/                     shared CRM-neutral semantic model
scripts/                        generator, validator, evaluator, installer
.codex-plugin/                  OpenAI plugin manifest
.claude-plugin/                 Claude plugin and marketplace manifests
gemini-extension.json           Gemini CLI extension manifest
```

## Contributing and security

Issues and pull requests are welcome. Start with [CONTRIBUTING.md](./CONTRIBUTING.md); report vulnerabilities privately according to [SECURITY.md](./SECURITY.md).

Released under the [MIT License](./LICENSE).

More