用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/bdarbaz/claude-stack-plugin --skill s-resume命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
Autonomous workflow loop - runs the full skill chain automatically with circuit breaker and stuck detection
Freeform router - accepts natural language and routes to the appropriate skill(s)
Show all available skills, agents, and workflow guides - quick reference card for claude-stack
基于 SOC 职业分类
正在显示 SKILL.md
| name | s-resume |
| description | Resume work from a previous session - reconstructs context from state files and recent history |
Reconstruct full working context from persistent state files so the user can seamlessly continue where they left off. This skill reads everything relevant and presents a structured summary of the project's current position.
Check if .planning/STATE.md exists.
If it does NOT exist:
/s:new to initialize a project."Read the following files in order:
.planning/STATE.md - Current position, active phase, decisions, blockers.planning/ROADMAP.md - Big picture: all phases, overall progressExtract from STATE.md:
Determine the active phase number from STATE.md. Then read:
.planning/phases/{N}/PLAN.md - What tasks are planned for this phase.planning/phases/{N}/CONTEXT.md - Phase-specific context and notesIf these files do not exist, note that the phase may not have been fully set up.
Read the 3 most recent files (by filename date prefix) from each directory:
docs/brainstorms/*.md - Recent brainstorm outputsdocs/plans/*.md - Recent implementation plansdocs/solutions/*.md - Recent compound learningsFor each file found, extract the title and first 5-10 lines to get the gist. If a directory is empty or does not exist, skip it silently.
Run these commands:
git log --oneline -10 - Last 10 commits for recent activity trailgit diff --stat - What files have uncommitted changes right nowgit status --short - Untracked and modified filesRead .continue-here.md at the project root if it exists. This file is created by
/s:pause and contains explicit instructions about what was being done and what to
do next.
If it exists, its content takes PRIORITY over inferred context - it is the most specific record of where work was interrupted.
Present the reconstructed context in this format:
=== CONTEXT RESTORED ===
Where we left off:
Phase {N}: {phase name}
Task: {active task description}
Status: {status}
What was in progress:
{description from .continue-here.md if available, otherwise inferred from
STATE.md and git diff}
Recent activity:
- {last 3-5 commits, one line each}
- {uncommitted changes summary}
Key decisions (recent):
- {decision 1 from STATE.md decisions table}
- {decision 2}
Knowledge context:
- Last brainstorm: {title and date}
- Last plan: {title and date}
- Last learning: {title and date}
Blockers:
{blockers from STATE.md, or "None"}
What's next:
{next steps from .continue-here.md, or inferred from phase plan}
Based on the reconstructed context, recommend the single most useful next action:
/s:build to resume implementation"/s:review to review phase {N}"If .continue-here.md was found and read, delete it now. Its purpose is served -
the context has been reconstructed. Keeping it around would cause stale data on
the next resume.
Run: rm .continue-here.md (only if the file existed)