Skip to content

Push-vs-Pull Standards

matt-pocock's taxonomy for injecting coding standards / style / architectural rules into agent workflows. The two modes trade tokens for relevance in opposite directions.

"Push is where you push instructions to the LLM. So you say, okay, if you put something in Claude.md, talk like a pirate, that instruction is always going to be sent to the agent — you're pushing tokens to it. Pull is where you give the agent an opportunity to pull more information. … A skill is something that can sit in the repo, and it has a little description header that says, okay, agent, you may pull this when you want to." — Pocock, 1:28:18–1:28:50

The two modes

  • Push — always-on tokens. CLAUDE.md, system-prompt rules, pre-injected style guides. Every turn pays for them whether needed or not; they pollute the smart zone.
  • Pull — discoverable resources the agent loads on demand. Claude Code skills, docs, fetched files. The header description is the advertisement; the body is only paid for when relevant.

Pocock's placement rule

Different roles in the pipeline want different modes:

  • Implementer → pull. The implementer explores a scoped task; it should pull whichever standard applies (React rules on frontend work, DB rules on schema work) so it stays lean.
  • Reviewer → push. The automated reviewer (see afk-implementation's Sandcastle loop) needs the full coding standards in-context next to the diff, so the rules are pushed unconditionally: "push that information to the reviewer so the reviewer has both the code that's written and the coding standards to compare to." (1:29:20)

Cross-wiki echoes