memory-consolidation

Consolidate a project memory store by merging duplicates, retiring stale entries, and proposing a review diff.

2|Updated Mar 3, 2026
One-click install
npx skills add https://github.com/DanielH2018/server --skill memory-consolidation-danielh2018
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: memory-consolidation
Source: https://github.com/DanielH2018/server/tree/main/.claude/skills/memory-consolidation
Command: npx skills add https://github.com/DanielH2018/server --skill memory-consolidation-danielh2018

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Project memory stores grow monotonically because every session can append entries but nothing ever demotes or removes them, leading to bloated indexes, duplicate facts, and stale pointers that every session pays for in injected context. ## Core Features & Use Cases - Survey-driven audit: Runs a memory survey script to surface dead index links, orphaned files, near-duplicates, and unreferenced entries, then reads each candidate before judging it. - Four-part proposal: Emits a diff under Merge, Retire, Re-verify, and Promote headings, with evidence (overlap scores, file:line citations, dates) for every entry. - Safe retirement: Moves retired entries to memory/archive/ with a pointer in archive/README.md instead of deleting them, so a wrong call is one git mv away from reversal. - Promotion and ownership checks: Identifies entries that should become executable checks (pytest guards, prek hooks) or move into a single agent's own memory directory. - Use Case: When the SessionStart hook reports MEMORY.md is over its line cap, run this pass as a standalone session to produce a shrink proposal the operator applies by hand. ## Quick Start Run a memory consolidation pass over the project memory store and propose which entries to merge, retire, re-verify, or promote.

Frequently Asked Questions about memory-consolidation

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I consolidate a Claude project memory store?▼

Run the memory survey script to get dead links, orphans, near-duplicates, and unreferenced entries, then read each flagged file and emit a proposal under Merge, Retire, Re-verify, and Promote headings. The operator applies the diff; the pass never writes to the store itself.

How do I reduce MEMORY.md size without losing information?▼

Retire entries by moving them to memory/archive/ with a pointer in archive/README.md rather than deleting them, and promote mechanically checkable facts into pytest guards or validation hooks. Promotion removes the index line while the check enforces the fact.

Why should memory consolidation not edit MEMORY.md directly?▼

Multiple worktree sessions append to MEMORY.md concurrently, so a pass that edits it directly races them and can silently drop entries. The skill emits a proposed diff for the operator to apply instead.

When should a memory entry be retired versus kept?▼

Never retire solely because the survey marks an entry unreferenced; age is a reason to read it, not drop it. Retire only when the entry is superseded, for example by an executable check, and re-verify borderline entries against a cited file:line.

When should a memory entry move to an agent-specific memory directory?▼

Move an entry when exactly one agent reads it, so the shared index stops charging every session for it. Confirm the real consumer from the skill that reads the file, and keep cross-domain entries in the shared store.