Claude Session
📖 技能介绍
name: claude-session description: | Claude Code session management. Topics — id (current session UUID), list (enumerate sessions), search (keyword + result validation), import, summarize, analyze (stats), archive (move to ~/.claude/projects/.bak/ with flat naming), classify, clean-profanity (sanitize text in session JSONL), split (topic boundaries), compress (UTCP/code-mode), destroy, dual-sync (Windows/WSL memory path mapping), install (hook), memory-trim (MEMORY.md index byte-budget trim), migrate (project to worktree), move (with cwd update), purge (dead sessions), rename (custom title), repair (chain/tool_result/UUID), url (web URL). Use when: "session id", "current session", "session list", "list sessions", "session search", "find session", "session classify", "session compress", "session migrate", "session move", "session repair", "chain repair", "session rename", "session split", "session purge", "dead session", "session url", "session analyze", "session import", "session summarize", "session archive", "archive session", "session clean", "clean profanity", "sanitize session", "redact session", "worktree session", "session cleanup", "memory trim", "MEMORY.md over budget", "dual sync", "WSL memory", "Windows memory path" metadata: author: es6kr version: "0.1.5" depends-on: [utcp]
Session
Integrated skill for managing Claude Code sessions.
Topic Dispatch
When this skill is invoked with a topic specifier (e.g., /claude-session id or Skill("claude-session", "id")), load and follow only the matching topic file (id.md). Do not echo the Topics table or summarize other topics in the response. The Topics table below is an index for invocations without a topic specifier — it is not user-facing output when a topic is named.
HARD STOP — unknown topic word + extra args
When the first token of args does NOT match any topic in the Topics table below, do NOT silently fall back to rename (or any other topic) by matching argument-count heuristics.
The signature <uuid> <text> matches both rename-session.sh <id> <title> and several other session-content operations (clean-profanity, repair, summarize, etc.). Args alone are not enough to decide.
Decision procedure:
- If args contain
<uuid>followed by an action verb / imperative phrase (e.g.,<uuid> remove profanity,<uuid> remove X,<uuid> clean Y,<uuid> sanitize Z), the user wants a session-content operation. Match againstclean-profanity,repair,summarizeby intent. - If args contain
<uuid>followed by a noun phrase / descriptive label (e.g.,<uuid> "Auth refactor PR #42"), the user likely wantsrename. - If intent is unclear, invoke
AskUserQuestionlisting candidate topics — never default torename.
| # | Don't | Do |
|---|---|---|
| 1 | Route <uuid> remove profanity to rename because the arg signature matches rename-session.sh <id> <title> |
"remove profanity" reads as action verb → route to clean-profanity (or AskUserQuestion if uncertain) |
| 2 | Treat any non-topic-word trailing text as a title for rename | Classify text as verb-phrase (action) vs noun-phrase (title) first |
| 3 | Silent fallback when no topic matches | AskUserQuestion: list 2–4 candidate topics that fit the apparent intent |
Topics
| Topic | Description | Guide |
|---|---|---|
| analyze | Session statistics, tool usage patterns, optimization insights | analyze.md |
| archive | Move a completed session out of the active project key to ~/.claude/projects/.bak/<project-key>_<uuid>.jsonl (flat naming, single backup root) |
archive.md |
| classify | Classify project sessions (delete/keep/extract) | classify.md |
| clean-profanity | Sanitize profanity tokens in a session JSONL file (in-place replacement) | clean-profanity.md |
| split | Analyze topic boundaries and recommend session split points | split.md |
| compress | AI-compress sessions via UTCP/code-mode | compress.md |
| destroy | Delete current session and restart IDE | destroy.md |
| dual-sync | Windows ↔ WSL project memory path mapping — keep both environments' memory in sync | dual-sync.md |
| id | Look up current session ID (UUID) | id.md |
| import | Pipeline session data to other agents/skills | import.md |
| install | Register session-id-inject hook in settings.json | install.md |
| list | Enumerate current-project sessions (UUID + mtime + size) | list.md |
| memory-trim | Trim project memory index (MEMORY.md) pointer-line hooks to a byte budget (script: scripts/trim-memory-index.py) |
memory-trim.md |
| migrate | Move sessions between projects (main repo → worktree) | migrate.md |
| move | Move specific sessions by ID to another project + update cwd | move.md |
| purge | Delete dead sessions (hook-only, no assistant response) permanently | purge.md |
| rename | Assign and look up custom title for session | rename.md |
| repair | Restore session structure (chain, tool_result, UUID) | repair.md |
| search | Keyword session search with result validation (verb/path/class checks) | search.md |
| summarize | View and summarize conversation content from other sessions | summarize.md |
| url | Generate claude-sessions web URL from session ID | url.md |
Quick Reference
Summarize (View/Summarize Sessions)
/session summarize # select project/session then summarize
/session summarize <session_id> # summarize a specific session
Import (Pipeline Delivery)
/session import --hookify # deliver to hookify
/session import --analyze # analysis pipeline
/session import --to <agent> # deliver to specific agent
Analyze (Session Analysis)
/session analyze # analyze current session
/session analyze <session_id> # analyze specific session
/session analyze --sync # sync to Serena memory
Archive (Move Session Out of Active Project)
这个技能的最新版本可以在7w4.net小葱技能站找到。
/session archive <session_id> # move to ~/.claude/projects/.bak/<project-key>_<uuid>.jsonl
bash ~/.claude/skills/claude-session/scripts/archive-session.sh <session_id> # direct script call
bash ~/.claude/skills/claude-session/scripts/archive-session.sh <session_id> --dry-run # preview only
Moves to ~/.claude/projects/.bak/<project-key>_<uuid>.jsonl (flat naming, single backup root shared with transient backups). UUID portion preserved unchanged. Updates INDEX.md ledger.
Split (Topic Split Recommendation)
/session split # Recommend split for current conversation
/session split <session_id> # Recommend split for specific session
/session split --execute # Execute recommendation immediately
Classify (Session Classification)
/session classify # classify current project sessions
/session classify --depth=medium # required when classifying sessions scheduled for split
/session classify --execute # execute immediately after classification
⚠️ --depth=medium or higher required before split — fast only reads the last 3 messages, so it may miss different topics at the end of the session.
🔍 RAG MCP auto-detection — If a vector store MCP (Qdrant / Chroma / Weaviate / Pinecone / etc.) is registered in the current context, classify additionally recommends sessions worth saving to RAG. Vendor-agnostic — uses whichever store tool is detected. See Section 8 of classify.md.
Move (Move Specific Sessions by ID)
/session move <session_id> <target_project_path> # default: --cwd-mode first
/session move <session_id> <target_project_path> --cwd-mode all # update all cwd occurrences
/session move <session_id> <target_project_path> --cwd-mode first # update only first cwd
Move explicit session IDs to another project directory and update cwd references via Python script. Cross-platform (Windows + macOS/Linux). Unlike migrate, no classification — just direct move.
Migrate (Move Sessions Between Projects)
/session migrate # classify + move code sessions to worktree
/session migrate --dry-run # preview only
/session migrate <source> <target> # specify source/target projects
Classifies sessions as CODE/INFRA/TINY/READ, then moves CODE sessions to worktree project and optionally deletes TINY sessions.
Compress (Session Compression)
/session compress <session_id> # compress specific session
/session compress # batch compress sessions containing "hookEvent":"Stop"
Register claude-sessions-mcp with UTCP, then call via code-mode.
List (Enumerate Current-Project Sessions)
/session list # list current-project sessions (UUID + mtime + size)
/session list --all-projects # summary across all projects
/session list --limit 20 # top N by mtime
Non-destructive enumeration. For categorization or cleanup, use classify or purge instead.
ID (Current Session ID Lookup)
/session id # look up current session ID
Current session ID — fast path (handle here, do NOT read id.md):
- Hook injection (preferred) — Look for
Current session ID: {uuid}in conversation context. SessionStart hook~/.claude/hooks/session-id-inject.shinjects it asadditionalContextat session start. If present, return the UUID immediately. - File path UUIDs — Scan recent tool results for UUID pattern
[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}in subagent/task/transcript paths. Return if found. - Marker fallback — Only if 1 & 2 fail: read
id.mdfor the marker→grep procedure.
Search (Keyword Session Search)
/session search Makefile remove # find sessions by keyword
/session search --today ansible/Makefile # only sessions modified today
/session id <keyword> # legacy alias — routed to search
Always run the result-validation gate (verb ambiguity, artifact location, action class) before reporting matches as the answer to "which session did X" — a misplaced artifact path is a "task orphaned" signal, not a successful match.
Destroy (Delete Session)
scripts/destroy-session.sh
Purge (Dead Session Cleanup)
/session purge # dry-run: list dead sessions in current project
/session purge <project_name> # dry-run: specific project
/session purge --all # dry-run: all projects
Dead session = 10 lines or fewer + no "type":"assistant" response.
Script: scripts/purge-dead-sessions.sh <project_name> [--delete]
Clean Profanity (Sanitize Session Content)
# Single session
python3 ~/.claude/skills/claude-session/scripts/clean-profanity.py <session_file.jsonl>
# Multiple sessions
python3 ~/.claude/skills/claude-session/scripts/clean-profanity.py file1.jsonl file2.jsonl
# Resolve UUID → path first (glob catches both bare <uuid>.jsonl and
# archived flat names like <project-key>_<uuid>.jsonl)
find ~/.claude/projects -name "*<uuid>.jsonl"
Replaces matched tokens with **** in place. Patterns loaded from data/profanity-patterns.json. Back up before running; do not run on the live current session.
Repair (Session Recovery)
/session repair # default: current session (uses hook-injected ID)
/session repair <session_id> # repair specific session
/session repair --dry-run # preview only
/session repair --check-only # validate only (no repair)
Primary script (full pipeline: backup → dedup → 400 error → orphan tool_result → chain → validate):
python3 ~/.claude/skills/claude-session/scripts/repair-session.py <session_file>
python3 ~/.claude/skills/claude-session/scripts/repair-session.py <session_file> --dry-run
Repair targets: - Broken chain (missing parentUuid) - Orphan tool_result (no matching tool_use) - Duplicate UUIDs / duplicate message.id (Syncthing conflicts) - API 400 error lines (incl. invalid surrogate pair / thinking-block signature)
Rename (Naming a Session)
Current session → output a copyable /rename list (NO script, NO AskUserQuestion):
Session name suggestions:
1. `/rename Candidate 1`
2. `/rename Candidate 2`
3. `/rename Candidate 3`
/rename is a Claude Code built-in command — it cannot be invoked via Bash or the Skill tool. The user copies and pastes the desired line. Do NOT call rename-session.sh for the current session (it is reserved for other sessions by ID).
Other session (session ID specified) → apply via script:
# Assign a name to a specific session
bash scripts/rename-session.sh <session_id> "name"
# Check current title
bash scripts/rename-session.sh --show <session_id>
# List named sessions in current project
bash scripts/rename-session.sh --list
Project Name Conversion Rules
| Actual Path | Project Name |
|---|---|
/Users/es6kr/works/.vscode |
-Users-es6kr-works--vscode |
/Users/es6kr/Sync/AI |
-Users-es6kr-Sync-AI |
Rule: all non-alphanumeric characters → - (i.e., replace(/[^a-zA-Z0-9]/g, '-'))
Requirements
- claude-sessions-mcp MCP server required
- Serena MCP server (when using analyze --sync)
🤖 AI 评测
这个 Skill 质量不错，提供了从会话查看、搜索到归档、修复的一整套工具，文档清晰且有中文指南，使用前建议仔细阅读风险提示。脚本运行较稳定，支持预览模式避免误操作。不足之处在于部分高级功能依赖外部工具，压缩会话前需要额外配置；对新手来说功能较多，上手有一定门槛。整体适合需要管理大量会话的高级用户，普通用户可先从基础的查看和搜索功能开始使用。
