From Header · the optimization layer for AI coding agents
One command grades your agent's setup and reads your real sessions to say how your AI coding is really going.
npx skills add Header-inc/Header-skill -g
then run /header in your agent. Works in Claude Code, Codex, Cursor, or any tool with a shell.
LLM agents are non-deterministic. You optimize one the way you'd optimize any such system: measure, enforce, prove.
Find the debt, and the habits.
Two reads, both local and read-only. One scans your setup: config bloat, supply-chain gaps, unused MCP servers, the wrong model tier. The other mines your session transcripts for where the work actually breaks down.
Determinism, enforced.
Mechanisms beat good intentions. A CLAUDE.md rule is a suggestion the agent can forget; a pre-commit gate, a test ratchet, and compounding memory can't be skipped. Header installs the ones you're missing.
Prove what's next.
A/B testing code used to cost engineering hours. With agents, a variant costs tokens. Header queues the uncertain changes, runs them against your own tasks, and merges only the statistically significant wins into your harness.
Not a one-time cleanup. A loop that keeps running as the models and your stack move.
Static config says the setup is messy. It can't say you shipped four fixes this week without running a test, or re-fixed the same bug all afternoon. Header reads your real sessions and pinpoints it.
Every finding comes with the move that ends it: a guardrail, a worktree, a memory entry. How you actually build, plus the fix.
Same skill, two jobs: size up the setup before you build, capture what you learned after.
Audit the setup, so you start from a clean, graded harness instead of yesterday's debt.
Review what the session taught you and write the pitfalls and learnings into committed memory. The next session starts smarter on its own.
Out of the box, enrichment comes from public agentic-coding briefings. After the audit, Header reads your repo, the languages, frameworks, and tools you actually use, and builds a custom topic around them on its own.
That's where Header earns its keep: every run names the change that matters for your codebase, not general advice.
The audit and public briefings are free forever. A custom topic tuned to your repo in Header comes with 3 free briefings, then $15 a month.
npx skills add Header-inc/Header-skill -g
Then /header. Header is the optimization layer underneath.