open-coder-ai Token Efficiency Rule
trigger: large command output, broad searches, re-reading unchanged files, front-loading references. avoid: wasting context window on low-signal content.
cap(tool_output): 4000_bytes; cap(search_results): top_3; cap(retry_loops): max_3_iterations
prefer: targeted_reads|structured_output|on_demand_refs; never: re-read(unchanged_file)|load_all_upfront
This skill is advisory: the client reading it has no mechanism to enforce it, and this policy stays advisory even when compiled by chock -- it ships rule text, not a blocking hook. See https://github.com/open-coder-ai/chock