Why Does CLAUDE.md Keep Growing? Catastrophic Remembering in Agentic Coding

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Kushal Chakrabarti
• Venue: arXiv: cs.AI
• Published: 2026-08-11
• Collected: 2026-08-12
• URL: http://arxiv.org/abs/2608.11095v1
• PDF: https://arxiv.org/pdf/2608.11095v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The paper discusses the growth of Agentic coding README files like CLAUDE.md in real repositories and attributes this to imperfect recall. The authors refer to this phenomenon as 'catastrophic remembering'. They find that appending an instruction is cheap but deleting an outdated one can be costly, leading to the unbounded growth of instructions over time. The paper also suggests that prompt comments can help reduce this growth by encoding latent reasoning and improving real-world agentic instruction-following.

What the Document Says

• The growth of Agentic coding README files like CLAUDE.md in real repositories is due to imperfect recall, a phenomenon named 'catastrophic remembering'.
• Agentic prompts grow without bound, more than tripling over their lifetime (+226%), gaining +4.9 net instructions every commit.
• The older an instruction gets, the less likely it is to be deleted (log-hazard -0.032/commit).
• Prompt comments can halt the growth of Agentic prompts, with comments encoding latent reasoning removing 99.3% of excess instructions (+211.3% to +1.4%).
• Applying the same approach to WildIFEval shows that prompt comments can improve real-world agentic instruction-following by up to 23.1%.

Key Concepts Mentioned

• Agentic coding
• CLAUDE.md
• imperfect recall
• catastrophic remembering
• prompt comments
• instruction growth

Claims Made by the Document

• The growth of Agentic coding README files is due to imperfect recall.
• Agentic prompts grow without bound over their lifetime.
• The older an instruction gets, the less likely it is to be deleted.

Evidence Used

• Growth observed in 247,694 instruction lifetimes in 1,867 repositories.
• Log-hazard rate provided.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Agentic coding READMEs like CLAUDE.md grow without bound in real repositories, stopping only when the repository retires or someone rewrites the file wholesale. We trace this to imperfect recall: appending an instruction is always cheap, but once an instruction's rationale is gone, deleting it without risking a correctness regression costs O(2^|D|) in a prompt of |D| instructions. We name the resulting divergence catastrophic remembering, the inverse of catastrophic forgetting around which continual learning is organized. First, we characterize this phenomenon across 247,694 instruction lifetimes in 1,867 repositories: agentic prompts grow without bound, more than tripling over their lifetime (+226%), gaining +4.9 net instructions every commit; further, the older an instruction gets, the less likely it is to be deleted (log-hazard -0.032/commit). Then, we show that prompt comments can halt the growth: inverting IFEval yields verifiable worlds whose optimal prompts are known, and there comments encoding latent reasoning remove 99.3% of excess instructions (+211.3% to +1.4%). Finally, applying the same inversion to WildIFEval, we show that prompt comments can improve real-world agentic instruction-following by up to 23.1%. If English is the new code, why don't we have comments yet?