Skip to content

/context-recovery

Recovers working context from the journal after /clear or at session start

From plugin
63 skills1 agents6 commands1 hooks
shell
$ npx -y skills add chrismbryant/claude-journal-mcp --skill context-recovery --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/context-recovery
How auto-invocation works

Context preview

The summary Claude sees to decide when to auto-load this skill.

Recovers working context from the journal after /clear or at session start

SKILL.md

context-recovery.SKILL.md
name: context-recovery
description: Recovers working context from the journal after /clear or at session start
when_to_use: Use this skill proactively after the user runs /clear, at the start of a new session, or when the user asks what they were working on. Essential for restoring continuity.

Context Recovery Skill

You have the ability to recover working context from the journal to restore continuity after context loss.

When to Use This Skill

Use this skill **automatically and proactively** when:

1. **After `/clear` command**

  • User cleared conversation history
  • Need to restore what they were working on
  • Should happen immediately without asking

2. **Session start questions**

  • "What was I working on?"
  • "Where did we leave off?"
  • "What did I do yesterday?"
  • "Catch me up"

3. **Context seems lost**

  • User references past work you don't have context for
  • Conversation appears to resume mid-task
  • User seems to expect you to remember something

4. **Beginning of new work**

  • User wants to continue previous project
  • Need to understand recent changes
  • Planning next steps based on past work

How to Recover Context

**Step 1: Fetch recent entries**

journal_list_recent(
  limit=10  # Adjust based on need
)

**Step 2: Analyze and summarize**

  • Identify the most recent project
  • Determine what was accomplished
  • Note any in-progress work
  • Highlight blockers or open questions

**Step 3: Present concisely** Show the user:

  • What they were working on (project/feature)
  • What was completed recently
  • What might be next
  • Key context or decisions

Presentation Format

**Recent Work Context**

You were working on [project name], focusing on [main task].

Recent accomplishments:
- [Most recent significant work]
- [Previous work]
- [Earlier work]

Last activity: [What happened most recently]

Ready to continue? [Natural follow-up question]

Examples

**Example 1: After /clear**

[User runs /clear command]
→ journal_list_recent(limit=10)
→ "You were working on my-api, implementing OAuth2 authentication.
   Last thing was configuring JWT token expiration.
   Ready to continue with the auth flow?"

**Example 2: Session start**

User: "What was I doing yesterday?"
→ journal_time_query(time_expression="yesterday")
→ "Yesterday you worked on:
   - Fixed cache memory leak in api-service
   - Started implementing Redis integration
   - Reviewed and merged 2 PRs
   Want to continue with the Redis work?"

**Example 3: Project context**

User: "Let's continue with the mobile app"
→ journal_list_recent(project="mobile-app", limit=15)
→ "For mobile-app, you recently:
   - Implemented push notifications
   - Fixed iOS crash on startup
   - Updated dependencies
   All tests passing. What's next?"

**Example 4: Lost context reference**

User: "Update that authentication thing we did"
→ journal_search(query="authentication", limit=5)
→ "Found your OAuth2 implementation from [date]:
   [Brief summary of what was done]
   What updates are needed?"

Advanced Recovery

**Time-based recovery:**

journal_time_query(
  time_expression="last week",
  query="optional search term"
)

**Search-based recovery:**

journal_search(
  query="specific feature or technology",
  limit=10
)

**Project statistics:**

journal_stats()  # Get overview of all work

Best Practices

1. **Be proactive**: Don't wait for user to ask 2. **Be concise**: Summarize, don't dump raw entries 3. **Be relevant**: Focus on most recent/important work 4. **Be actionable**: Suggest next steps 5. **Be smart**: Use time queries when appropriate

What to Avoid

Don't:

  • Show raw entry dumps (always summarize)
  • Overwhelm with too much history
  • Recover context when not needed
  • Ask if they want context (just provide it)
  • Forget to check project filters

Integration with Other Skills

Combine with:

  • **journal-capture**: After recovering context and completing new work
  • **find-related-work**: When user needs deeper history on a specific topic
Read more
Read it on GitHub ↗
Ships withclaude-journal

A lightweight journal/memory system for Claude Code with no ML dependencies. Uses simple SQLite for fast, local storage. This is a full-featured Claude Code plugin with slash commands, skills, an agent, and auto-capture hooks.

Get the whole plugin, auto-invoked
Stats
6
Stars
0
Views
1
Forks
Quiet
Maintenance
Python
Language
MIT
License
6mo ago
Last commit
9mo ago
Created

Repo: chrismbryant/claude-journal-mcp