Cross-Session Agent Memory with Hybrid Retrieval
Build or fix automatic coding-agent memory: hook capture, typed memory units, vector + keyword + graph retrieval, budgeted injection.
Before & after
Same prompt, same agent — the only difference is the skill.
I'd need more context to help you find that decision. Can you give me some details about what kind of auth discussion you're referring to? Was it about OAuth, JWT, session management, single sign-on, or something else? What was the general context - were you talking about implementing a new auth system or debugging an existing one? If you can point me to the session or conversation where that happened, I could help you find the specific decision you're looking for.
AGENT MEMORY RETRIEVAL - AUTH DECISIONS VECTOR SEARCH RESULTS (relevance 0.94 - 0.87): 1. Session 47 (2026-06-28): OAuth 2.0 Migration Decision - DECISION: Migrate from custom auth to Auth0, 30-day deadline - Key factors: Support ticket volume down 60% with managed provider - Cost impact: +$400/mo, saves 12 eng-hours/week - Owner: Sarah (auth lead) - Status: In progress, 70% complete 2. Session 44 (2026-06-18): JWT Expiry Windows - DECISION: Set access tokens to 15 min, refresh to 7 days - Trade-off: Security vs. UX friction - Accepted risk: Refresh token theft (mitigated by device fingerprint) - Owner: Marcus - Status: Deployed to staging 3. Session 39 (2026-05-30): Passwordless Auth Trial - DECISION: Pilot magic links for creator onboarding only - Metrics: 40% faster signup, 8% bounce reduction - Rollout: Phase 2 starts July 15 - Owner: Priya - Status: Active, metrics tracked WEB VIEWER: View full decision graph and edit stored context at localhost:5002
About this skill
Your agent either forgets everything between sessions, or your memory file has grown so large that it loads last month's abandoned decisions into every prompt. This skill designs, builds, and debugs a real memory layer for coding agents. It captures sessions automatically, compresses them into small typed memories, and injects only what the current task needs.
What it does
- Places capture at the right Claude Code hook events, so recording happens with no manual logging and without slowing your session
- Compresses sessions into typed memory units (decisions with rationale, gotchas, preferences, open questions) instead of transcripts or vague summaries
- Builds hybrid retrieval: keyword (FTS5) plus vector search fused by rank, then a one-hop graph expansion from the files you are touching
- Stops stale memory with supersede-on-write, file-existence checks, and expiring task state
- Diagnoses symptoms like "injects irrelevant stuff", "slow startup", or "duplicates piling up" and gives the targeted fix
- Tells you plainly when a curated CLAUDE.md is enough and a vector store is overkill
What you get
An architecture sized to your project, working inline code (hook scripts, SQLite schema, retrieval function, settings snippet) or a focused diagnosis and fix, plus a replay test to confirm retrieval actually improved.
Best for
Developers on long-running Claude Code projects who have outgrown markdown session logs, and builders adding persistent memory to their own agents.
SKILL.md preview
---
name: cross-session-agent-memory
… (sign up to view the full skill)More ai & agents skills
View all AI & Agents skills →New Skill Scaffolder
Create a new Claude Code skill correctly: right mechanism, valid name and folder, a triggering description, and a smoke test.
Skill Library Lint and Overlap Report
Lint every skill in a repo at once: frontmatter, name-folder match, weak descriptions, overlapping triggers, broken refs, plus a CI script.
Self-Verifying Agent Loop: Plan, Check, Retry, Close
Turns a goal into a task plan with machine-run checks, executes with retry caps, escalates, and refuses to close until verified.
Skill Test Runner (Trigger and A/B)
Test a skill in fresh sessions: ready-to-run prompts, commands, and rubric for trigger recall, false triggers, and blind A/B.