Skip to content

JPeetz/agent-skills / skills/ai-agent-engineering/agent-memory-management

agent-memory-management

Verified

Use this skill when designing, implementing, or debugging AI agent memory systems — long-term memory, working memory, episodic memory, vector stores, summarization strategies, state persistence, context window optimization. Activates on "agent memory", "long-term memory", "working memory", "memory architecture", "state management", "context window", "vector store", "memory retrieval", "RAG", "summarization", "memory persistence", "conversation memory", "episodic memory", or "memory optimization".

FileKindSize
CHANGELOG.mdmarkdown4.0 KB
LICENSEother1.0 KB
SKILL.mdskill md29.9 KB
evals/evals.jsonconfig10.9 KB
references/memory-architectures.mdmarkdown15.6 KB
scripts/memory_architect.pyscript41.2 KB

CHANGELOG.md

View on GitHub
# Changelog

All notable changes to this skill are documented here. Format follows
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/); this project adheres to
[Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [1.0.0] - 2026-11-16

### Added

- Initial release: `agent-memory-management` skill under the `ai-agent-engineering`
  category.
- Memory taxonomy with lifetimes and selection criteria: working, episodic,
  semantic, and procedural memory, plus the distinction between episodic events
  and semantic facts and why it changes the store design.
- Four architecture families with ASCII diagrams and explicit failure modes:
  single-store, dual-store (working + long-term), hierarchical (L0–L3), and
  distributed/agentic (event bus + blackboard).
- Context window management: an explicit segment budget with protected output
  reserve and safety margin, four retention strategies (sliding window,
  summarization, selective retrieval, priority-based retention), and a
  five-step compaction order that cuts the least damaging thing first.
- Vector store integration: embedding model discipline (no mixed models per
  index, query/document asymmetry, normalization), a chunking strategy table
  with concrete token defaults, ANN index selection (HNSW / IVF-PQ / flat) with
  recall measurement guidance, hybrid search with RRF and cross-encoder
  reranking, and pre-filter metadata requirements including mandatory tenant
  scoping.
- Persistence pattern comparison across in-process dict, JSONL/file, relational
  DB, vector DB, distributed cache, and object store, with the four rules that
  prevent divergence (single source of truth, append for history, write-ahead
  intent, namespace by agent/session/tenant).
- Retrieval strategy table (last-N, semantic, hybrid, importance-weighted,
  recency-weighted, graph traversal) with a concrete tunable ranking function,
  dedupe-before-budget, pinned-set injection, and per-step query rewriting.
- Memory consolidation as a six-step offline pass: cluster, deduplicate, resolve
  conflicts by superseding, promote episodic→semantic and procedural→declarative,
  roll up old episodes, and prune by documented rule — idempotent and reversible.
- Summarization approach comparison (extractive, abstractive, recursive,
  hierarchical, map-reduce) plus the rules that keep summaries faithful:
  verbatim identifiers, provenance, bounded recursion depth, and the
  narrative-summary + facts-ledger split.
- State management: session state, workflow state schema, self-sufficient
  checkpoints, and five-rule idempotent resumption including derived
  idempotency keys and lease-based fencing; names the phantom-checkpoint and
  optimistic-checkpoint failure patterns.
- Sixteen anti-patterns with description, impact, and fix — including unscoped
  multi-tenant retrieval, post-filtering top-k, mixed embedding models, random
  idempotency keys, similarity-score-as-priority, and summarize-only memory.
- Five reference architectures: personal assistant, support/RAG agent, coding
  agent, long-running pipeline, and multi-agent blackboard.
- `scripts/memory_architect.py` — stdlib-only advisor that validates a JSON
  requirement profile and emits a markdown recommendation covering tier,
  components, retrieval strategy, summarization, context budget, consolidation
  policy, cost estimates, tradeoffs, risks, and first build steps. Includes
  `--help`, `--print-schema`, `--json`, `--out`, and `--self-test` (5 profiles,
  8 invalid-config cases, 45 assertions), plus tolerant parsing so the output of
  `--print-schema` can be edited and fed straight back in.
- `references/memory-architectures.md` — full architecture comparison with a
  scoring matrix, patterns by agent type, cost/latency/accuracy/complexity
  tradeoff axes, migration paths, and a selection checklist.
- `evals/evals.json` — four positive activation cases (architecture selection,
  recall failure, crash-safe resumption, retrieval quality) and two near-miss
  negatives (unit testing, document summarization).
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall