메뉴
Hex memory system — save, search, and retrieve persistent memories across sessions. Provides FTS5 + optional vector hybrid search over indexed files.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | memory |
| description | Hex memory system — save, search, and retrieve persistent memories across sessions. Provides FTS5 + optional vector hybrid search over indexed files. |
| tags | memory, search, recall, persistence, knowledge |
| trigger | Agent needs to save or recall information across sessions, or search existing memories. |
| version | 1 |
The hex memory system stores persistent facts across sessions using SQLite FTS5 (full-text search) with optional vector embedding for semantic recall. Memories are markdown files with YAML frontmatter stored in .hex/memory/.
The legacy Python memory scripts were removed — the memory subsystem is native to the
hexbinary now. Use thehex memorysubcommands instead.
hex memory search "query terms"
hex memory search --top 5 "phrase"
hex memory search --compact "keyword"
hex memory search --private "keyword" # excludes me/, people/, raw/
hex memory search --file PATTERN "keyword" # restrict to matching file paths
hex memory search --context N "keyword" # show N lines of context
hex memory index # incremental (skips unchanged files)
hex memory index --full # force re-index all files
hex memory index --stats # show index statistics
hex memory parse-transcripts # parse JSONL transcripts → markdown
Memories are markdown files with YAML frontmatter:
---
name: memory-name
description: One-line description for index lookup
type: user | feedback | project | reference
---
Memory content here.
If sqlite_vec and fastembed are installed, search uses vector embeddings for semantic recall. Falls back to FTS5-only when deps are absent.
Use when asked to audit a repository, assess codebase health, find what's wrong with a codebase, identify technical debt, or produce a prioritized improvement plan for an existing project. Also use when a repo feels untrustworthy — red CI, stale docs, mystery failures — and the user wants a grounded picture before investing, or says "upgrade this project", "what should we fix here", "how bad is this code".
Upgrade hex to the latest version from hex
Idempotently updates and restructures a repository's documentation — agent instruction files (AGENTS.md / CLAUDE.md, root and nested) and human docs (README, docs/) — converging them to a router-plus-topic-doc architecture and fixing doc↔code drift with code-cited evidence; re-running on an unchanged repo (prior run merged) produces zero diff. Use when asked to "clean up the docs", "restructure CLAUDE.md / AGENTS.md", "split this giant instruction file", "docs are stale", "fix doc drift", "set up agent docs for this repo", or after an audit confirms documentation findings worth fixing.
Delegate tasks to BOI v2. Covers TOML spec authoring, dispatch, context wiring, and monitoring. Use when delegating any non-trivial task — code changes, research, generation, or multi-step work.
Run the vibe-to-production pipeline on any Python project. Assesses code quality, generates characterization tests, prioritizes refactoring targets, and dispatches BOI specs for each phase. Use when the user says "vibe to prod", "harden this", "production-ready", "assess this project", "run assessment", or references the vibe-to-production playbook.
Validate hex agent structure and repair issues. Runs health checks against the expected directory layout, reports findings by severity, and auto-fixes safe issues. Also handles data migration from backup directories on first launch after bootstrap migration. Use when the user says "hex doctor", "check health", "fix my hex", "something is broken", or on first launch when .hex/migrate-from exists.