# Memory Management

> Framework-agnostic project memory infrastructure for engineering tasks. Use for sparse recall, stale-memory handling, secret-safe durable checkpoints, and MCP/CLI fallback without consuming specialist slots.

- Skill: `soden46/memory-management-2` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add soden46/memory-management-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/soden46/memory-management-2/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: soden46 (https://skillmd.com/u/soden46)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/soden46/memory-management-2

---


# Memory Management

Use memory as conditional infrastructure, not specialist expertise.

## Preflight

Run memory preflight only when prior project/session/workflow context could materially affect correctness. Skip self-contained tasks.

Prefer an active MCP memory tool when the host provides one. Otherwise use:

```bash
node skills/engineer-flow/infrastructure/memory-management/scripts/memory.mjs auto --cwd <project-root> --query "<task intent>" --limit 5
```

The command prints `decision: RUN` or `decision: SKIP`. On `RUN`, use only sparse relevant memory and verify it against current code/config.

## Rules

- Memory never consumes primary or support specialist slots.
- Current code/config beats old memory.
- Do not store secrets, `.env` values, raw credentials, or personal data.
- Preserve provenance and status.
- Checkpoint only durable reusable knowledge.

## Checkpoint

After meaningful work, checkpoint only when durable knowledge changed:

```bash
node skills/engineer-flow/infrastructure/memory-management/scripts/memory.mjs checkpoint --project <alias> --summary "<durable summary>"
```