Independent reproduction and Anthropic’s response will determine whether Claude Code silently truncates the newest persistent-memory index entries after an undocumented roughly 24 KB or 200-line limit.
state: expiredheat: lowuncertainty: highnovelscott: lowclaude-code agent-memory memory-limitsAnthropic
What is this?
Claude Code’s auto-memory system reportedly uses a flat `MEMORY.md` index that points to topic files. A GitHub feature request and a third-party technical write-up say the index is capped at roughly 200 lines and 25 KB, with excess entries at the bottom silently omitted on reload, leaving their underlying files present but undiscoverable to Claude. The supplied snippets do not establish an official Anthropic response or that a proposed fix was closed as “not planned,” and they differ slightly from the case’s stated 24 KB threshold.
Why it matters to Scott
The report is topically aligned with Scott’s work on coding agents and persistent agent memory, but no supplied wiki or radar hits connect it to a position, project, or existing tracked case. Until independently reproduced or answered by Anthropic, it is an unverified implementation-limit report rather than a development that clearly changes what Scott should build or argue.
queries asked of Scott's wikis
- agent memory index architecture and hierarchical retrieval
- silent truncation and observability in coding-agent memory
- append-only memory ordering and newest-entry loss
- Claude Code persistent-memory harnesses and workarounds
- memory capacity limits versus context-window growth
- governance for agent-maintained files and orphaned knowledge
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-08-07T18:31:27Z
Repeated checks have produced neither an independent reproduction nor an Anthropic clarification, while the disputed newest-memories-first claim remains unsupported. The case has faded into an unverified implementation-limit report with no active development expected.
2026-08-01T14:26:48Z
The newly attached material adds no independent reproduction or first-party clarification; it remains repetitive anecdotal amplification of a plausible 200-line index limit. The stronger claim about silently dropping newest memories is still unsupported and appears inconsistent with the primary issue.
2026-08-01T13:21:23Z
New comments add anecdotal encounters and workarounds around the 200-line constraint, but no controlled independent reproduction or Anthropic clarification. They also do not rescue the disputed newest-memories-first claim, so the discussion is mostly amplification rather than stronger evidence.
2026-07-28T00:21:35Z
No independent reproduction or official Anthropic response has emerged, and the primary issue does not clearly support the Reddit post’s newest-memories-first framing. The case remains a testable report of an undocumented index-display limit, but its claimed failure direction is unsettled.
2026-07-27T22:23:17Z
grounded: novel/low — The report is topically aligned with Scott’s work on coding agents and persistent agent memory, but no supplied wiki or radar hits connect it to a position, pro
2026-07-27T22:22:38Z
origin walked (codex/luna, conf 0.97): anchor reddit.post.1v8dnkl -> echo.github.20acb8d19e by theta55
2026-07-27T22:21:43Z
case created — The report identifies a specific, testable persistent-memory failure mode that is distinct from the existing agent-memory cases.
Decision trace
- 08-08 04:31expireRepeated checks have produced neither an independent reproduction nor an Anthropic clarification, while the disputed newest-memories-first claim remains unsupported. The case has faded into an unverif
- 08-08 04:31alert_silentThe only change is negligible engagement movement after 48 hours; it adds no evidence and creates no consequential delta for Scott.
- 08-08 04:31alert_routeThe only change is negligible engagement movement after 48 hours; it adds no evidence and creates no consequential delta for Scott.
- 08-02 00:26repriceThe newly attached material adds no independent reproduction or first-party clarification; it remains repetitive anecdotal amplification of a plausible 200-line index limit. The stronger claim about s
- 08-02 00:20mark_dirtycomment_update
- 08-01 23:21repriceNew comments add anecdotal encounters and workarounds around the 200-line constraint, but no controlled independent reproduction or Anthropic clarification. They also do not rescue the disputed newest
- 07-28 10:21repriceNo independent reproduction or official Anthropic response has emerged, and the primary issue does not clearly support the Reddit post’s newest-memories-first framing. The case remains a testable repo
- 07-28 10:20mark_dirtyengagement_update
- 07-28 08:23groundThe report is topically aligned with Scott’s work on coding agents and persistent agent memory, but no supplied wiki or radar hits connect it to a position, project, or existing tracked case. Until in
- 07-28 08:22promote_anchororigin walk conf 0.97
- 07-28 08:21createThe report identifies a specific, testable persistent-memory failure mode that is distinct from the existing agent-memory cases.