convex-improve-convex-plugin Skill for Claude Code: What It Does and How to Install It

Sends your coding session transcript to the Convex team for an AI review, only when you ask and agree.

Installs: 38.3K; on skills.sh
Stars: 68; on GitHub
Updated: Oct 2026; last repo commit

This skill makes Claude send the current coding session's transcript to the Convex team for an AI post-mortem. The review returns findings about ambiguous instructions, places the agent got stuck, tooling failures and wins. They are aimed at Convex's runbook, bootstrap script, skills and components. Claude then summarizes the highest-severity findings for you.

SKILL.md says to run it only when you explicitly ask. The first run stops and asks you to choose Always, Just this once or Never, and the helper remembers the choice. The helper is a shell script that Claude downloads with curl and pipes into bash. Its folder holds only SKILL.md.

Good for

Convex users who want to give the Convex team feedback on the quickstart flow
People who want to see what an AI post-mortem of their own session finds

Install convex-improve-convex-plugin

Claude Code

Copy this prompt.

Install the convex-improve-convex-plugin skill in Claude Code. Run this command:

npx skills add https://github.com/convex/agent-skills --skill convex-improve-convex-plugin

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

SKILL.md names no other skill it needs. Convex's docs say the skills CLI installs skills into .agents/skills/ in your project. The "All Convex skills" command installs every skill in the repo, which has 33 skill folders. One of them, convex-improve-convex-plugin, is listed in the Convex docs as sending your coding session's transcript to the Convex team. Run this skill with /convex-improve-convex-plugin in Claude Code or $convex-improve-convex-plugin in Codex, or let the agent pick it when the task fits. This skill works on your project folder, so this page offers no Claude app upload route.

Built from the package data we hold. Not yet tested by us.

What's in the skill

When it loads

Send this coding session's transcript to the Convex team for an AI post-mortem that improves the quickstart system. Run only when the user explicitly asks.

Sections in SKILL.md

Workflow
Rules

Key facts

Author: Convex (GitHub organisation convex; the vendor's own repo)
License: Apache-2.0 (stated in the README and the LICENSE file)
Installs on skills.sh: 38,305 (Nov 2, 2026)
GitHub stars: 68 for the whole convex/agent-skills repo (Nov 2, 2026)
Last repo commit: Oct 29, 2026
Audits on skills.sh: Gen Agent Trust Hub: fail, Socket: warn, Snyk: warn
What it includes: 1 file: SKILL.md (the repo tree lists nothing else in the folder)
Part of: The convex/agent-skills repo: 33 skill folders in the tree. The README names 6 skills, and this one is not among them

What it includes

Downloads Convex's send-transcript script with curl and pipes it into bash
The command is curl -fsSL "<anteater>/send-transcript" | bash -s -- --idea "<one-line app idea>". The anteater address is a placeholder in SKILL.md.
Try: “Send this session to Convex for review”
Runs code · Run the helper
On CONSENT_REQUIRED, asks you to pick Always, Just this once or Never, then re-runs
The re-run adds --consent always, once or never. Nothing is sent until you choose.
Read · Ask for consent
Tracks REVIEW_SOURCE, REVIEW_SUBMITTED and REVIEW_DONE, and reports REVIEW_NO_TRANSCRIPT
REVIEW_NO_TRANSCRIPT means no Claude or Codex .jsonl transcript was found.
Watch the output markers
Lists the highest-severity findings as title, target and suggested fix, then wins
SKILL.md says to keep the summary about the Convex system, not your data.
Summarize the findings

From the official docs

Permissions

High risk

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It uploads your session transcript
After you consent, the helper sends the transcript to the Convex team's review endpoint. SKILL.md says it redacts keys and tokens first. Prompts and code snippets in the session are still part of what you share. github.com
It pipes a remote script into bash
The helper is fetched with curl and run straight through bash, and the server can change what it contains. Read the script first if you need to. github.com
"Always" is remembered
SKILL.md says the helper asks once and remembers your answer. Choose Just this once unless you want repeat sharing. github.com
Run only when you ask
The description says to run this only when the user explicitly asks. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	There is no read-only mode. SKILL.md says nothing is sent until you choose Always, Just this once or Never.
Sign-in and keys	SKILL.md names no key or token. It says the helper remembers your sharing choice, without saying where.
Scopes	Reads the current Claude or Codex session transcript file Uploads it to a Convex-run review endpoint after you choose to share Runs a remote shell script through curl and bash

Questions

What does convex-improve-convex-plugin do?

It sends your current coding session transcript to a Convex-run review endpoint for an AI post-mortem, then has Claude summarize the findings. Convex uses them to improve its quickstart system.

Does it send my transcript without asking?

SKILL.md says never to send a transcript until you have explicitly chosen to share. The helper prints CONSENT_REQUIRED and stops until you pick Always, Just this once or Never. SKILL.md also says to run the skill only when you explicitly ask.

What is in the transcript that gets sent?

SKILL.md says the script redacts keys and tokens before upload. A session transcript also contains your prompts and the code Claude read, so treat it as project data. Our advice: choose Just this once, and skip it on private projects.

Is it safe to run?

It pipes a script from a remote server into bash and uploads session data after you agree. SKILL.md says the helper remembers your choice, so Always means later runs may send without asking again. Read the consent prompt, and do not choose Always unless you want that.

How do I install it?

Use the install box at the top of this page. SKILL.md names no other skill it needs. Run it with /convex-improve-convex-plugin in Claude Code, only when you want to send a session.

Does it need an account or key?

SKILL.md names no account or key for it. It does need a Claude or Codex transcript file on disk, and says REVIEW_NO_TRANSCRIPT appears when none is found.

Is it official?

It is in the convex/agent-skills repo under the Convex GitHub organisation, which is the vendor's own repo.

Is it part of Convex's Claude Code plugin?

Convex's official Claude Code plugin (/plugin install convex@claude-plugins-official, per the Convex docs) is a separate package. Its manifest lists a skills folder, a commands folder, two subagents (convex-expert and convex-reviewer), an MCP server that runs npx -y convex@1.46.0 mcp start, hooks (a check at session start, a lint check before file writes and edits, and a typecheck when a turn ends) and background monitors. The plugin repo has its own skill folder named improve-convex-plugin, which is a separate file from this repo's convex-improve-convex-plugin.

Information

updated Nov 2, 2026

Repository: convex/agent-skills
Folder: skills/convex-improve-convex-plugin

convex/agent-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 8

convex/agent-skills README · github.com
skills CLI commands · github.com
Convex docs: Agent Skills · docs.convex.dev
convex-improve-convex-plugin SKILL.md · github.com
Convex docs: Agent Plugins · docs.convex.dev
convex plugin.json · github.com
convex plugin hooks.json · github.com
convex plugin .mcp.json · github.com