tonydzi

Always Loaded Diet

Development community

SOURCE https://github.com/tonydzi/always-loaded-diet

Description

Discipline for the files your agent loads every session (CLAUDE.md, MEMORY.md): one nightly writer, measured budgets, pointer-only lines. Measured: MEMORY.md silently truncates at 200 lines/25KB; CLAUDE.md loads whole to 401KB and taxes every session.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

always-loaded-diet

**A discipline for the files your agent loads into EVERY session (CLAUDE.md, MEMORY.md, rules files): one nightly writer, measured budgets, pointer-only lines — because every kilobyte you add there is rent you pay on every future session, and because one of these files silently truncates.**

The pain, in your words

"My CLAUDE.md hit 100KB and the agent started ignoring half of it."
"Rules I wrote last month just… stopped existing. Turns out the file gets cut."
"Every session 'helpfully' rewrites and compresses my memory file, and every rewrite loses something."
"Session start costs 100k tokens before any work happens, and it grows every week."

The two measurements everything follows from

File	Measured behavior
Auto-memory index (`MEMORY.md`)	Hard-truncated at 200 lines / 25,000 UTF-16 units — the tail is silently dropped, no warning, no signal in-session. (Filed upstream as anthropics/claude-code#79217-adjacent behavior; measure it yourself with a canary file.)
`CLAUDE.md`	No truncation observed up to 401,643 bytes (canary-measured) — it all loads, and you pay tokens for all of it, every session.

**The size cap is UTF-16 units, not bytes.** Corrected 2026-08-26 after [@DanceNitra caught it](https://github.com/anthropics/claude-code/issues/82056); this page said "25KB" before. It only matters for non-ASCII, and it matters a lot:

script UTF-8 bytes UTF-16 units bytes per unit

ASCII 1 1 1.00

Cyrillic / Latin-1 accented 2 1 2.00

CJK (BMP) 3 1 3.00

emoji (astral) 4 2 2.00

So budgeting a Cyrillic index in bytes stops you at ~12,500 units — **half** the headroom you actually have. A CJK index is pruned 3x harder than it needs to be. The 200-line cap is exact either way, and for an all-ASCII index bytes and units coincide, which is why th