Development standards, installed in one command.

• Every file in the release is downloaded and hash-verified into a staging folder first. The version marker is written last, after the copy succeeded, so an interrupted run never describes itself as current.
• Writes into ~/.claude/ (constitution + rules), ~/.copilot/ (prompts, agents, hooks) and ~/.mock2/ (hook scripts, repo templates, version marker). Files retired by a newer version are removed only when they are demonstrably the installer's own, unmodified.
• ~/.claude/CLAUDE.md is never owned: it receives one marked block that imports the constitution; everything else in it is yours.
• A managed file whose contents no longer match what the installer last wrote is kept, reported, and the new version put beside it as <name>.mock2-new. To take the release over your copies (kept under ~/.mock2/backup):

• Installs the GitHub Copilot and Copilot Chat extensions when the code command is on PATH.
• It never closes your window: piped through iex it runs inside your session, prints any failure, and returns you to the prompt. Diagnose an install with node ~/.mock2/bin/doctor.mjs.

A repository you work on over SSH sits on a machine that has never run the installer, and VS Code Remote-SSH reads customizations from that host's home folder. So the agents and prompts go into the repository itself, where VS Code reads them as workspace customizations.

• The installer ends by offering the remote step. Every repository set up from this machine is catalogued in ~/.mock2/remotes.json, so it offers Update for those (one, or all) before the choices for a new host.
• New host: pick one from your .ssh/config, pick a folder (git repos are marked), see what would be added, updated or kept, confirm.
• Writes .github/agents/, .github/prompts/, .mock2/, state/, CLAUDE.md, AGENTS.md and .github/copilot-instructions.md. Files are carried over your existing SSH session; the remote host needs bash, tar and a sha256 tool, no network, nothing installed.
• Version-aware: a pristine file from an earlier release is updated, an edited one is kept with the new version beside it. The apply is journalled and recovers from an interrupted run.
• Commit what it adds. Every later clone then carries the agents and no one has to run anything.

VS Code reads ~/.copilot/prompts, ~/.copilot/agents and ~/.copilot/hooks for every workspace, and Claude Code reads ~/.claude/CLAUDE.md and ~/.claude/rules/ — so on a machine you have installed, the shared standards live once. A repository also carries its own copies of the agents and stage prompts under .github/agents/ and .github/prompts/; that is what makes a clone work on a host where nothing has been installed. The two copies are the same file, generated from core/copilot/ when the standards publish, so an agent is only ever edited in one place. On top of that a repo carries what is specific to it (.github/copilot-instructions.md, state/).

Nothing in the standards may refuse, restrict or delay building or testing. Production-policy questions that come up during work are classified (implementation requirement, configurable capability, checklist item, recommendation) and only an explicit requirement changes what gets built. When requested or when a change is ready, /mock2-check runs the applicable production checklist in an independent reviewer context and writes the change record. Valid recorded evidence is reused; missing or stale evidence is checked. Developer acceptance is recorded separately, and promotion is a person's decision informed by the evidence.

SessionStart fetches manifest.json, re-runs the installer if the version changed, and injects the loaded version into the chat context. Stop appends a line to ~/.mock2/ledger.jsonl (session, folder, latest change record) and, when MOCK2_API is set, posts it to the standards server. The hooks are user level, so a repository seeded onto a remote host does not self-update — bring it current by running the command again and choosing Update, or /mock2-init on a machine that has the installer.

Everything here is mock2/mock2-core, the source repository. To change a standard, edit the file under core/, bump version in manifest.json, and open a pull request. When it merges to main, a webhook updates this site within seconds and every developer picks the change up at their next chat.

Phase 2 (planned): an MCP endpoint at /mcp with get_rules, search_learnings and propose_rule, plus a session feed and a proposals inbox on this site. The reserved entry is already in mcp-config.json.