tushar:~/pack$ ← back to tushardev.com
Free No spam Leave in one click

Claude Workflow Pack

Claude Code will tell you it is done. This is the system I use to check.

Not another CLAUDE.md taxonomy. You already know that file exists, and if you have used it for more than a week you know instructions get followed most of the time, not every time. These are the working files that close that gap.

$ tree claude-workflow-pack/
claude-workflow-pack/
  ├── agents/    # the reviewer that cannot rubber-stamp
  ├── settings/  # allowlists + hooks the harness enforces
  ├── memory/    # CLAUDE.md + AGENTS.md portability
  ├── cost/      # token budgeting, the SRE version
  ├── mcp/       # when a server earns its context cost
  └── skills/    # and when NOT to build one
// what's inside
  • >A reviewer that cannot rubber-stamp you. No write access, forced SHIP / REVISE / REGENERATE verdicts, and an evidence audit that auto-fails invented output. The answer to "you're absolutely right!"
  • >One handbook across Claude Code, Codex and Cursor. Why your rules get silently ignored when you switch tools, and the symlink setup that fixes it. Most-requested gap in the whole tracker, and nobody covers it.
  • >Token budgeting, done like infrastructure. Where the spend actually goes, why context is re-sent every turn, and how to cap the blast radius before a loop surprises you.
  • >Hooks and allowlists, so the rules you keep repeating run every time instead of most of the time. Plus when to reach for a skill vs a subagent vs a hook vs MCP.
// what happens after

You confirm your email with one click, then it opens the repo, star it if it is useful. After that I send The Signal Log: what actually shipped in AI that week, what it breaks if you are running these setups, and a few prompts worth stealing. Usually weekly.

If I have nothing worth saying that week, I do not send anything. That is the whole promise.