Only 32% of Project Rules Reach Claude Code; Docs Plus Retrieval Lifts It to 70%
AttitudeLatter7306 · reddit · 2026-10-29
A team measured how many project "rules" — constraints not derivable from code — actually reach a coding agent, and open-sourced context-recall (MIT, a Claude Code plugin) to benchmark it.
Key findings
- From 12 real past tasks they extracted 75 rules (6 per task); 57 of 70 rules have no test that fails when broken.
- About a third of the rules real changes needed were written nowhere — they came from deep code paths, bugs the task itself uncovered, or external system behavior.
Results (headless Claude Code, blinded LLM judge)
- No project docs: 32% of rules delivered
- Docs in repo: 54%
- Their simple setup (a short critical-rules doc per module, picked by meaning and read fully before planning): 70%
- Code-indexing tools (Serena, Trail, Graphify) scored 24–29% / 46–47% — no better than bare Claude Code, because these rules aren't in the code.
Reproducible method: snapshot each repo before a past change, run any tool as an arm, judge blind, compare — including before/after a docs edit. Next direction: per-module "dreaming," analogous to how the brain replays experience into long-term memory.
Related event: Coding agents receive only 32% of project rules; docs boost it to 70%(2 posts)→
More from coding & agent
- Video2Game: H3 Max Generates Gameplay Video, Claude Turns It Into a Playable three.js Game — OdinLovis · 2026-10-29
- After 3 Years of Heavy AI Prompting, Domain Vocabulary Beats Prompt Tricks — Dimillian · 2026-10-29
- Agents push compute back to CPUs: $221B server CPU market seen by 2030 — BenBajarin · 2026-10-29
- Cursor removes its beloved Cmd+K command, frustrating power users — jbhuang0604 · 2026-10-29
- Designing a skill like an app: a developer's iteration diary for interface-bts — round · 2026-10-29
- One prompt one-shots a walkable 3D ship where AI agents visibly work around the clock — InternationalGap3698 · 2026-10-29