跳到正文
cccskills
免费在 GitHub 公开

session-management

Manages Claude Code sessions with naming, checkpointing, and resume strategies. Use when organizing long-running work or resuming across sessions.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md7.8 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Session Management

When To Use

  • Managing session checkpoints and state preservation
  • Resuming work across Claude Code sessions

When NOT To Use

  • Short sessions that do not need checkpoints
  • Fresh starts where no prior session context exists

Overview

Claude Code supports named sessions for better workflow organization. Use this skill to manage complex, long-running work across multiple sessions.

Available Commands

CommandDescription
/renameName the current session (auto-generates name if no argument given, 2.1.41+)
/resumeResume a previous session (REPL)
claude --resume <name>Resume from terminal

Workflow Patterns

1. Debugging Sessions

Name debug sessions for easy resumption:

# Start debugging
/rename debugging-auth-issue

# ... work on the issue ...

# If you need to pause, session is auto-saved
# Resume later:
claude --resume debugging-auth-issue

2. Feature Development Checkpoints

Create checkpoints during long feature work:

# After completing milestone 1
/rename feature-x-milestone-1

# Continue in new session
# Reference old session if needed

3. PR Review Sessions

For complex PR reviews that span multiple sittings:

# Start review
/rename pr-review-123

# Take breaks without losing context
# Resume:
claude --resume pr-review-123

4. PR-Linked Sessions (Claude Code 2.1.27+)

Sessions are automatically linked to PRs when created via gh pr create. Resume PR-specific sessions later:

# Resume session for a specific PR
claude --from-pr 156
claude --from-pr https://github.com/org/repo/pull/156

# Workflow: review → pause → resume with full context
/rename pr-review-156
# ... review work ...
# Later:
claude --from-pr 156

5. Investigation Sessions

When investigating issues that may require research:

# Start investigation
/rename investigate-memory-leak

# Pause to gather more info externally
# Resume with full context:
claude --resume investigate-memory-leak

Resume Screen Features

The /resume screen provides:

  • Grouped forked sessions: See related sessions together
  • Keyboard shortcuts (defaults, customizable via /keybindings):
    • Preview session content
    • Rename a session
  • Recent sessions: Sorted by last activity

6. Resume Hint on Exit (Claude Code 2.1.31+)

Claude Code now shows a resume hint when you exit, displaying the command to continue your conversation. This makes session resumption more discoverable: users no longer need to know about --resume beforehand.

Best Practices

Naming Conventions

Use descriptive, hyphenated names:

PatternExampleUse Case
debugging-<issue>debugging-auth-401Bug investigation
feature-<name>-<milestone>feature-search-v2Feature development
pr-review-<number>pr-review-156PR reviews
investigate-<topic>investigate-perfResearch
refactor-<area>refactor-api-layerRefactoring work

When to Name Sessions

Name sessions when:

  • Work will span multiple days
  • You might need to pause unexpectedly
  • The session contains valuable context
  • You want to reference it later

Session Cleanup

Unnamed sessions are eventually garbage collected. Named sessions persist longer. Periodically clean up old named sessions you no longer need.

Integration with Sanctum

Combine session management with other Sanctum skills:

  1. Before starting: Run Skill(sanctum:git-workspace-review) to capture context
  2. Name the session: /rename <descriptive-name>
  3. Work: Use appropriate skills for the task
  4. Resume if needed: claude --resume <name>

Troubleshooting

Session Not Found

If a named session isn't appearing in /resume:

  • Check for typos in the name
  • Sessions may expire after extended inactivity
  • Use /resume screen to browse available sessions

Duplicate Sessions in VS Code

If you see duplicate session entries when resuming in VS Code:

  • Claude Code 2.1.38+: Fixed: resume now correctly reuses the existing session without creating duplicates
  • Older versions: Ignore the duplicate entries; they point to the same underlying session

Lost Context After Resume

If context seems incomplete or resume is slow:

  • Claude Code 2.1.30+: 68% memory reduction for --resume via stat-based session loading with progressive enrichment, especially impactful for users with many sessions. Also fixes hangs when resuming sessions with corrupted transcript files (parentUuid cycles).
  • Claude Code 2.1.29+: Fixed slow startup when resuming sessions with many once: true hooks: saved_hook_context loading is now optimized
  • Claude Code 2.1.21+: Fixed API errors when resuming sessions interrupted during tool execution: previously these sessions could fail to resume entirely
  • Claude Code 2.1.20+: Session compaction/resume is now fixed: resume correctly loads the compact summary instead of full history
  • Use /catchup to refresh git state
  • Use /debug (Claude Code 2.1.30+) for session troubleshooting diagnostics
  • Re-run Skill(sanctum:git-workspace-review) if needed
  • If on older versions: resumed sessions may reload uncompacted history, increasing context usage unexpectedly

macOS Orphaned Processes (Claude Code 2.1.46+)

Previously, disconnecting from a terminal on macOS could leave orphaned Claude Code processes running. This is now fixed in 2.1.46+. If you encounter stale CC processes on older versions, manually check and kill them:

# Find orphaned claude processes
ps aux | grep -i claude | grep -v grep

7. Automatic Memory (Claude Code 2.1.32+)

Claude now automatically records and recalls memories as it works. Session summaries, key results, and work logs are captured implicitly and recalled in future sessions. This provides passive cross-session continuity without manual checkpoint management.

  • No action required: Memory recording is automatic on first-party Anthropic API
  • Complements named sessions: Automatic memory handles implicit continuity; named sessions provide explicit organization
  • Token overhead: Recalled memories add to baseline context, factor this into MECW budgets

8. Agent Persistence on Resume (Claude Code 2.1.32+)

--resume now re-uses the --agent value from the previous conversation by default. Agent-specific workflows that are resumed will continue with the same agent configuration without needing to re-specify it.

# Start with a specific agent
claude --agent my-agent

# Resume later — my-agent is automatically used
claude --resume

See Also

  • /catchup - Refresh context from git changes
  • /clear - Start fresh session
  • Skill(sanctum:git-workspace-review) - Capture repo context

Exit Criteria

  • Session named via /rename using a hyphenated descriptive name matching one of the documented patterns (e.g. debugging-*, feature-*, pr-review-*)
  • Session resumes via claude --resume <name> without loss of prior context (verified by presence of earlier conversation)
  • If a session name is not found on resume, the error is surfaced with troubleshooting steps (check typos, session expiry)
  • For PR-linked sessions (Claude Code 2.1.27+), claude --from-pr <number> correctly reopens the session associated with that PR
  • If session context seems incomplete after resume, Skill(sanctum: git-workspace-review) is re-run and its output confirms current repo state

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Shapes turns action-first: action leads, steps numbered, state restated. Use for ADHD-friendly output. Do not use to trim tokens; use response-compression.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

Inverts burden of proof for code additions. Use when reviewing PRs, planning refactors, or running unbloat to challenge every addition's necessity.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

Tracks per-agent token usage and flags waste in parallel dispatch. Use when evaluating parallel agent efficiency or after a multi-agent run.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

Coordinates Claude agent teams via filesystem protocol. Use when orchestrating parallel agents with task dependencies. Do not use for single-agent tasks.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

Evaluates API surface design, consistency, and exemplar alignment. Use when reviewing public API changes or before releasing a new API surface.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

Selects architecture paradigm via research before scaffolding. Use when architecture is undecided and the choice needs justification and documentation.

中文简介正在准备中。当前显示原文说明。

athola/claude-night-market3432026年10月17日 更新

查看 athola 的全部技能

报告此技能的问题