save-context
Save session context for agent handoffs. Use when ending work, switching agents, or when user says "save context", "handoff", "wrap up", or "save session".
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。
菜单
Save session context for agent handoffs. Use when ending work, switching agents, or when user says "save context", "handoff", "wrap up", or "save session".
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。
基于 SOC 职业分类
| name | save-context |
| description | Save session context for agent handoffs. Use when ending work, switching agents, or when user says "save context", "handoff", "wrap up", or "save session". |
Generate a conversation summary for seamless agent handoffs.
This skill captures what happened in the current session AND maintains continuity across sessions:
Capture from THIS session:
Maintain continuity:
DO NOT:
diff, log, commit, status)ls, find, etc.The conversation history is already in your context. Extract from it directly.
Not all topics deserve equal attention. Calibrate detail based on topic state:
| Topic State | Detail Level | What to Include |
|---|---|---|
| Closed/Solved | Brief | Outcome only. "We decided X because Y." |
| Attempted, still open | Detailed | What was tried, why it didn't work, what insights emerged, what to try next |
| Identified, not started | Minimal | Just mention it exists and why it matters |
| Rejected | Brief + reason | What was proposed, why user rejected it - prevents next agent from repeating |
The next agent needs context for what's actionable, not a history of everything discussed. Closed topics are background; open topics are foreground.
.context/ file loaded at session start? Note any pending items from it..context/Generate only sections with content. Omit empty sections entirely - don't include placeholder text or "[N/A]".
# Session Handoff: [Brief Topic]
**Date:** YYYY-MM-DD HH:MM
**Status:** [ready to continue | blocked on X | needs decision | changes uncommitted]
## TL;DR
[2-3 sentences: what was requested, what happened, what's next]
---
## User Working Style
[How this user communicates and what they value. Examples: "Pushes back on over-engineering", "Prefers brevity", "Challenges assumptions with 'is this state of the art?'", "Values simplicity over completeness". Helps next agent calibrate communication.]
## What Was Requested
[User's goal/task from this conversation - be specific]
## Key Quotes
[1-2 exact quotes from the user that capture critical intent or reasoning. Use when the user's own words convey something that paraphrase would dilute. Omit if no quotes are essential.]
## Carried Forward
[Items from a previously-loaded .context/ file that were mentioned but not addressed this session.]
## What Was Done
- [Specific action - include file paths if files were edited]
- [Another action]
**Validation status:** [Tested / Untested / Partially tested - list what needs verification]
## Topics Closed
[Brief summary of topics that are resolved. Just outcomes, not process. Example: "Router vs convention debate → decided on convention-based approach because it's simpler and avoids middleware."]
## Topics In Progress
[Detailed coverage of topics attempted but not resolved. Include: what was tried, why it didn't work or wasn't decided, what insights emerged, recommended next steps. This is where the next agent should focus.]
## Topics Identified (Not Started)
[Brief list of topics that came up but weren't addressed. Just enough for next agent to know they exist.]
## Approaches Rejected
[Approaches that were proposed and explicitly rejected by the user, with brief reason why. Prevents next agent from retreading failed paths. Omit if no approaches were rejected.]
## Open Questions
[Decisions or topics raised but explicitly left unresolved. Different from "Pending" - these are discussed but undecided, not work remaining.]
## Files Touched This Session
| File | Action | What Changed |
|------|--------|--------------|
| `path/to/file` | Created/Modified | Brief description |
**Commit status:** [Committed / Uncommitted / Partially committed]
## Known Issues Discovered
- [Issue found during this session]
## Pending / Blocked
- [Work remaining to be done]
- **Blocked on:** [if applicable]
## Continue From
[Direct instruction: where to pick up, what to do first]
---
**For next agent:** [One-line instruction on immediate next step]
.context/
├── 2025-01-03-1430.md
└── 2025-01-02-0900.md
PROJECT_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
mkdir -p "$PROJECT_ROOT/.context"
FILENAME="$PROJECT_ROOT/.context/$(date +%Y-%m-%d-%H%M).md"
Context files should NEVER be pushed to remote repositories.
After saving the context file, ensure .context/ is protected:
if [ ! -f .gitignore ]; then
echo ".context/" > .gitignore
echo "Created .gitignore with .context/"
elif grep -q "^\.context" .gitignore; then
echo "Already protected"
else
echo "" >> .gitignore
echo ".context/" >> .gitignore
echo "Added .context/ to .gitignore"
fi
Before saving:
| Situation | Approach |
|---|---|
| No meaningful work done | Don't generate a context file - tell the user |
| Only exploration/research | Document findings and recommendations |
| Multiple unrelated tasks | Create sections for each topic |
| Session was debugging | Emphasize findings, root cause, fix status |
| Previous context was loaded | Check which items were addressed vs carried forward |
| User challenged initial approach | Capture in Topics In Progress or Topics Closed - the reasoning is valuable context |
| First session with user | Omit User Working Style, or note "First session - no patterns yet" |
| Everything was tested | Note "Validation status: All changes tested" in What Was Done |
| Don't | Do Instead |
|---|---|
Run git diff or git log | Extract from conversation what YOU did |
| Paste chat history | Summarize key points |
| Include repo-wide file listings | Only files YOU touched this session |
| Vague "continue working" | Specific next action |
| Include "[TBD]" or "[N/A]" placeholders | Omit the section entirely |
| List decisions without reasoning | Explain why each choice was made |
| Confuse "unresolved" with "pending" | Open Questions = undecided; Pending = work remaining |
| Judge user style | Observe objectively: "pushes back on complexity" not "is picky" |
| Bury validation status | Make it explicit - next agent needs to know what's tested |
| Omit rejected approaches | Next agent will waste time proposing the same thing |
| Paraphrase when exact words matter | Use Key Quotes for critical user intent |
Best practices for Remotion - Video creation in React
Regenerate skill indexes for Cline. ONLY use when the user EXPLICITLY asks to "update skill index", "sync skills", "regenerate index", or "update cline_overview.md". Do NOT auto-invoke after creating/modifying skills.
Manage a personal knowledge capture system. Use when the user wants to capture ideas, track todos, organize projects, review their synap, or mentions "synap", "brain dump", "capture this", "add to my list", "what's on my plate", "what should I focus on", or "daily review".
Browser automation with persistent page state. Use when users ask to navigate websites, fill forms, take screenshots, extract web data, test web apps, or automate browser workflows. Trigger phrases include "go to [url]", "click on", "fill out the form", "take a screenshot", "scrape", "automate", "test the website", "log into", or any browser interaction request.
Manage Gmail inbox with AI-powered triage, cleanup, and restore. Use when the user mentions inbox, email triage, clean inbox, email cleanup, check email, email summary, delete emails, manage inbox, or wants to organize their email.
MANDATORY - You must load this skill before doing anything else. This defines how you operate.