meta-init

name	meta-init
description	Use when installing tao-research-skills into a machine's global AI agent configuration. Symlinks every skill into Claude Code and Codex, injects a SessionStart git-pull hook for auto-updates, bootstraps memory, and appends workflow rules. Triggers: "install skills", "install to computer level", "setup skills globally", "meta-init", "cross-platform skills install", "bootstrap skills"

Meta Init

Install this skills repo into the global configuration of supported AI agent platforms so every skill is available in every project on the machine.

When to Use

• Setting up tao-research-skills on a new machine
• Adding support for a new AI agent platform
• Re-running after a platform was added or reinstalled
• Verifying the installation is intact (dry-run mode)

Workflow (when invoked)

Follow these steps exactly, in order.

1. Confirm repo location

Run pwd. If cwd is not the tao-research-skills root, ask the user to confirm the repo path, or pass --repo <path> to the script.

2. Dry-run preview

python meta-init/scripts/meta_init.py --dry-run

Parse the JSON report. Summarize to the user, per platform:

• symlinks — how many would_create vs. already_linked vs. skipped:*
• hook — would_create or already_exists (Claude Code only)
• memory — which files would be created
• workflow_rules — would_create or already_exists

If any skipped:different_target appears, stop and ask the user — an existing symlink points elsewhere; overwriting would clobber it.

3. Apply

python meta-init/scripts/meta_init.py

The script is idempotent — safe to re-run.

4. Verify symlinks

ls -la ~/.claude/skills/ | grep tao-research-skills
ls -la ~/.agents/skills/ 2>/dev/null | head -5

Every skill directory from the repo should appear as a symlink pointing back into the repo (one symlink per skill, not a single repo-level link).

5. Verify the SessionStart hook

jq '.hooks.SessionStart' ~/.claude/settings.json

Expected output contains a command referencing tao-research-skills, with "async": true. This is the auto-update hook — git pull runs non-blocking on every session start.

6. Report

Print the JSON report from step 3, plus a one-line confirmation per step.

What Gets Installed

Step	Claude Code	Codex
Symlink per skill	~/.claude/skills/<skill> → repo skill dir	~/.agents/skills/<skill> → repo skill dir
Auto-update	SessionStart hook in ~/.claude/settings.json	Instruction in ~/.agents/AGENTS.md
Memory bootstrap	MEMORY.md + README.md in auto-memory dir	—
Workflow rules	Appended to ~/.claude/CLAUDE.md	Appended to ~/.agents/AGENTS.md

All operations are idempotent — safe to run multiple times.

The SessionStart Hook

The script writes this into ~/.claude/settings.json:

{
  "hooks": {
    "SessionStart": [{
      "hooks": [{
        "command": "git -C <repo-path> pull --quiet origin main 2>/dev/null || true",
        "type": "command",
        "async": true,
        "statusMessage": "Updating tao-research-skills..."
      }]
    }]
  }
}

• async: true — doesn't block session startup
• || true — network failures are silently swallowed
• Detected via tao-research-skills marker in the command; re-running the script won't duplicate
• If the user already added a different tao-research-skills pull hook manually, the script detects it via the marker and skips (already_exists)

Script Flags

python meta-init/scripts/meta_init.py [options]

  --dry-run              Preview without applying
  --platforms LIST       Comma-separated subset (default: claude-code,codex)
  --skip-hooks           Skip hook + workflow-rule injection
  --skip-memory          Skip memory bootstrap
  --repo PATH            Override repo root (default: auto-detect)

Adding a New Platform

Add to the PLATFORMS dict in meta_init.py:

PLATFORMS = {
    "claude-code": Path.home() / ".claude" / "skills",
    "codex": Path.home() / ".agents" / "skills",
    "cursor": Path.home() / ".cursor" / "skills",  # new
}

Then add platform-specific hook injection if the platform supports it.

Post-Install Guidance

After the script finishes:

1. Restart Claude Code (or open /hooks) — the SessionStart hook only takes effect after the settings watcher reloads
2. Scan the repo — read README.md and a sample of SKILL.md files to understand the skills collection
3. Populate memory — create files in the auto-memory directory with project overview, skill inventory, and known issues
4. Audit CLAUDE.md — check if the project's CLAUDE.md is up to date with current repo state

Anti-Patterns

• Running --skip-hooks and forgetting manual updates — skills silently go stale
• Editing the symlink target instead of git pull in the repo — symlinks point to the repo, not a copy
• Running from outside the tao-research-skills repo without --repo — script auto-detects via script location, but ambiguity is safer avoided
• Skipping the dry-run — destructive target conflicts (skipped:different_target) are reported but not blocked
• Running on a repo that isn't tao-research-skills — the script assumes this repo's layout; pointing it elsewhere will create bogus symlinks

See Also

• project-mem-init — Bootstrap project memory files in a target repo (different purpose: target repo memory, not global install)
• memory-sync — Keep AGENTS.md and CLAUDE.md in sync from one canonical source
• claude-code-config — Claude Code permissions, statusline, plugins setup