---
title: Skills in the Venom plugin for Claude Code
description: The skills that come with the Venom plugin for Claude Code, in every project once it is installed - plain writing, two management audits, doc sync, Linear plans, RCAs, cloud migrations, security audits, technical due diligence and Venom Studio artifacts.
section: cloud
order: 44
audience: [user, operator, agent]
features: []
sources: [clients/claude-code/skills, clients/claude-code/test/skills.test.mjs, .claude-plugin/marketplace.json]
---

The [Venom plugin for Claude Code](/docs/cloud/claude-code-sessions.md) also ships skills. Once the plugin is installed, Claude Code
has them in every project, not only in the Venom repo. Claude picks a skill when your request matches its description,
and you can call one by name: `/venom:humanizer`, `/venom:rca-create`.

Install or update the plugin as in [Claude Code sessions](/docs/cloud/claude-code-sessions.md#install), then run `/plugin` to check
that version 0.4.0 or newer is installed (0.3.0 has every skill but `studio-artifact`; 0.2.0 also lacks `tech-due-diligence`). The skills need no Venom Cloud sign-in and make no network call of their own.

## The skills {#skills}

| Skill | Use it to |
|---|---|
| `humanizer` | Rewrite text that sounds machine-written, without changing what it says. Venom runs every PR description, issue and message through it before it goes out. Vendored from blader/humanizer (MIT). |
| `audit-drucker` | Audit a meeting, calendar, plan, decision record or backlog against Peter Drucker's *The Effective Executive*. |
| `audit-eliyahu` | Pressure-test a plan, process or investment case through Goldratt's Theory of Constraints: find the constraint and whether a change moves the goal. |
| `doc-sync` | After a change, update the `CLAUDE.md`, `AGENTS.md` and `docs/public` pages it affects, with scoped edits, and run the feature inventory. Outside Venom it works from the docs that repo has. |
| `linear-plan` | Turn a spec or idea into Linear issues for team VEN, label `batch-1`, one issue per pull request, in dependency order, with an "Agent notes" fold and reserved migration numbers. |
| `rca-create` | Write a blameless root cause analysis from GitHub Actions runs, PRs, Linear and Venom Incidents, as a note under `Incidents/Postmortems/` or a markdown file. |
| `create-migration` | Write the next Venom Cloud migration with a number no open PR has taken, in the house style, and run the lint and migration tests. |
| `security-audit` | Audit Venom's attack surface (cloud routes and server actions, MCP and OAuth, the daemon mesh and join tokens, the vault, webhooks, self-hosting, the browser extension relay, the harness sandbox) or a branch's diff, reporting only findings with a traced exploit. |

| `tech-due-diligence` | Review someone else's code safely: a prospective client's, an acquisition's, a vendor's. It gathers the context from Venom People, Granola, Notes and Tasks and asks for what is missing, clones into a quarantine folder, checks for malware from metadata and file shape before reading, never installs, builds or runs anything, and writes the reports. See [below](#due-diligence). |
| `studio-artifact` | Make a Venom Studio deck, page or HTML page worth opening: a brief, research where needed, the outline in the chat first, the build from the catalog, a review, and an answer with sources and what is unverified. Needs Venom Studio's MCP server. See [Studio for agents](/docs/studio/agents.md#workflow). |

`doc-sync`, `linear-plan`, `create-migration` and `security-audit` are written for the Venom repository and name its paths,
teams and conventions. The other six work anywhere.

## Technical due diligence {#due-diligence}

`tech-due-diligence` is for code you did not write and cannot trust yet. Ask for it in words ("do a due diligence on this
client's GitHub org", "is this repository safe to open") or call `/venom:tech-due-diligence`.

What it does, in order:

1. **Gathers the context.** It looks for the person in Venom People, the meeting in Granola or Notes, and open issues in
   Venom Tasks, then asks one question for whatever is missing: the repositories, what the client says is wrong, what the
   review is for, where the reports go. A tool that is not connected is named, and you are asked for that input directly.
2. **Clones into quarantine.** A folder outside every project, with git hooks, submodules and LFS downloads off.
3. **Checks for malware before reading.** `scripts/dd_scan.py` reads git's object store and reports shapes, never content:
   editor tasks that run when a folder is opened, code hidden behind a long run of spaces, fonts that are not fonts, loaders
   in server entry files, install hooks, environment files, and commits whose dates do not add up. When something is found
   it stops and tells you, then works out which accounts pushed it from the host's push log, because commit names can be forged.
4. **Reads the code as text.** Nothing is installed, built, linted, tested, run or opened in an editor.
5. **Writes the reports**: an overview with a diagram, a report per repository, a plain-language page for the client, who
   maintains what, a requirements draft when the repositories have none, and an incident note when malware was found.
   They go to a Notes folder for the client when Venom is connected, else to markdown files. Fees and phasing stay in an
   internal note.
6. **Hands over**: what was and was not done to the code, the limits of the review, drafts of the messages you may want to
   send, and an offer to delete the clones.

It never contacts the client, never puts a payload or a secret's value in a report, and searches the web only with generic
indicators, never the client's name or code. The scanner finds known tricks; it does not prove a repository is clean.

The same skill is in every organisation's [Skills registry](/docs/cloud/console-skills.md#first-party), so venom and the agents page have
it without the plugin.

## Working on Venom itself {#repo}

In the Venom repository the skills live once, in the plugin (`clients/claude-code/skills/`). `.claude/skills/<name>`
links to each, so agents working in the repo find them without installing the plugin, and the two copies cannot drift.
Edit a skill in the plugin's folder. `make test-claude-code` checks that every skill has a valid `name` and `description`
(the name equals its folder; the description is at most 1,024 characters) and that every link resolves.
