Skip to content

v0.2.0: install into any repo, from inside any coding agent

Choose a tag to compare

@cerredz cerredz released this 30 Oct 06:47
2fe35cf

Install vidbyte-memory into any existing repo, from inside the coding agents you already use. Every correction you give the agent still becomes a slim fact, a test that fails if the mistake comes back, and a footprint commit. What's new is that this now works in your repo and your agent, not only in a fresh template copy.

Install

From the repo you want to give memory:

curl -fsSL https://raw.githubusercontent.com/cerredz/vidbyte-memory/main/scripts/install.py | python3 -

On Windows PowerShell, use irm … | python -. Add --dry-run to see the plan first. You can also install it as a plugin:

  • Claude Code: /plugin install vidbyte-memory --marketplace cerredz/vidbyte-memory
  • Codex: codex plugin marketplace add cerredz/vidbyte-memory, then codex plugin add vidbyte-memory@vidbyte
  • Gemini CLI: gemini extensions install https://github.com/cerredz/vidbyte-memory
  • Antigravity: agy plugin install https://github.com/cerredz/vidbyte-memory
  • Skills only, in any of 78 agents: npx skills add cerredz/vidbyte-memory

Then ask the agent to "install vidbyte-memory in this repo".

Added

  • A one-command installer, scripts/install.py:
    • It copies the memory in without overwriting any of your files, and backs up every file it edits.
    • It folds your existing AGENTS.md into the new Task section.
    • It wires the turn-end guard for every agent it finds.
    • It lists each write in .vidbyte-memory/install.json. A second run changes nothing.
  • install.py uninstall undoes the install:
    • It deletes what the install created and restores what it changed.
    • It takes its hook back out of agent settings you edited since, keeping your own settings.
    • It keeps the memory your agent wrote unless you add --force.
  • Enforcement in more agents. While tests fail, feedback is pending or work is uncommitted, scripts/guard.py keeps the agent working:
    • Stop hooks written by the installer: Claude Code, Codex, GitHub Copilot, Gemini CLI and Antigravity. Cursor, Devin CLI and Grok Build read the Claude Code hook.
    • New plugins in integrations/:
      • OpenCode and Kilo Code: re-prompts the session on idle.
      • Pi: agent_before_settle.
      • Hermes: a pre_verify plugin.
  • Plugin manifests, so the repo installs as a plugin:
    • .claude-plugin/: the Claude Code marketplace vidbyte;
    • plugin.json (Agent Plugins 1.0): Codex, Copilot, Cursor, Kiro, Devin, Goose and others;
    • gemini-extension.json;
    • an install-memory skill.
  • Skill mirrors in .agents/skills/ and .claude/skills/, where about 25 agents look for project skills.
  • docs/hosts: one page per agent (24), each with the install command, what the agent reads and how much is enforced.

Fixed

  • Footprint commits failed on Python 3.9. scripts/footprint.py --write and --commit crashed there, the oldest version the README supports. Task repos made from v0.1.0: copy scripts/footprint.py.
  • Two of the six skills were invisible to skill loaders. Five files had frontmatter values that YAML cannot parse, so the loaders skipped handle-feedback, the protocol's entry point, and one other skill. The CHANGELOG lists the files to copy.

Credits

Requested by @cerredz. Every release credits whoever asked for each change, so open an issue or a discussion and your name goes in the next one.

Full changelog: CHANGELOG.md