# memory-to-llms-txt Turns an agent's persistent memory store — Claude Code auto-memory, a .remember/ time-decay pyramid, a napmem-style pyramid, or an ad-hoc project MEMORY.md — into a well-formed llms.txt family, preserving type/recency structure as facets and flagging entries that have drifted from reality. A memory store isn't a scratch dump — it already has real structure: a type per entry (user/feedback/project/reference), an index linking entries to files, explicit recency tiers (`now` vs `today` vs `archive`), dated entries the memory system itself already compressed once. [notes-to-llms-txt](/skills/notes-to-llms-txt/) assumes none of that exists and re-derives topic clusters from an unstructured pile; `memory-to-llms-txt` assumes the opposite — the structure *is* the input's main asset, and the job is carrying it into the llms.txt grammar without flattening it away. ## How it works 1. **Identify the shape** — Claude Code auto-memory (`MEMORY.md` index + frontmatter'd entry files), a `.remember/` pyramid (`now.md` → `today-*.md` → `recent.md` → `archive.md` → `core-memories.md`), a napmem-style pyramid queried through its own retrieval tool, or an ad-hoc project `MEMORY.md` with no formal typing. 2. **Ingest with provenance** — every linked/cross-referenced file, not just the index; record source, type, recency tier, and any `[[wikilinks]]` per unit. 3. **Verify against reality** — the step this skill has that [notes-to-llms-txt](/skills/notes-to-llms-txt/) doesn't need: for every fact naming a checkable thing (a path, a version, a command), a best-effort check against the live system. A memory is a durable claim about system state in a way a scratch note rarely is, and durable claims go stale — a path that moved, a version that shipped since the note was written. 4. **Draft by type/tier**, not by re-clustering into new topics — a `feedback`-type entry and a `project`-type entry about the same subject are different *kinds* of claim (a standing rule vs. a point-in-time fact), and collapsing that distinction away loses real information. 5. **Compile and hand off** to [llms-deep-optimizer](/skills/llms-deep-optimizer/) for the multi-pass audit, same handoff `notes-to-llms-txt` makes. A drifted entry is never silently corrected or dropped — it's exported with its original claim intact plus a `## Drift flags` line naming what changed and when it was checked. Same redaction discipline as its sibling: anything secret-shaped is scrubbed to `[REDACTED]` before any output file, since a memory export is exactly the kind of thing that ends up queryable by more than just its author later. **Use it for:** "make MEMORY.md queryable", "export the .remember pyramid as llms.txt", "build a servable index of what this agent remembers". **Not for:** unstructured scratch notes or meeting dumps with no maintained typing ([notes-to-llms-txt](/skills/notes-to-llms-txt/)) · a repo's code or docs rather than its memory files ([crawl-repo-to-llms](/skills/crawl-repo-to-llms/)) · an llms.txt family that already exists and just needs a pass ([llms-deep-optimizer](/skills/llms-deep-optimizer/)).