Claude Code Plugins

Community-maintained marketplace

Feedback

context-status

@majiayu000/claude-skill-registry
27
0

Checks current context window usage and provides recommendations. Use when sessions feel slow, responses seem degraded, or before starting complex work.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name context-status
description Checks current context window usage and provides recommendations. Use when sessions feel slow, responses seem degraded, or before starting complex work.
allowed-tools Read

Context Status

Monitor context window health. Saturation causes degraded performance, forgotten instructions, and tool neglect.

When to Use

  • Session feels "slow" or less sharp
  • Before complex multi-step task
  • After loading many files
  • Noticing instructions being forgotten
  • Periodically (every 30-60 minutes)

Saturation Symptoms

Symptom Severity
Forgetting earlier instructions High
Not using mentioned tools High
Repeating completed work Medium
Simplified responses Medium
Missing analysis details Critical

Context Levels

Level Indicators Action
Low Session <15min, few files Continue normally
Medium 15-45min, moderate files Monitor, delegate more
High 45-90min, many files /checkpoint, strongly delegate
Critical >90min, heavy loading /compact immediately

Quick Self-Test

  • Can recall safety rules?
  • Can recall available tools?
  • Can recall agent spawn criteria?
  • Remember current task objective?

If any unclear → context likely saturated.

Output

## Context Status

**Level**: [Low/Medium/High/Critical]
**Session Duration**: [estimate]

### Retention Check
- Safety rules: [clear/fuzzy]
- Tool awareness: [clear/fuzzy]
- Task objective: [clear/fuzzy]

### Recommendation
[continue/checkpoint/compact]

### Suggested Action
[specific next step]

Preservation Strategies

  • Delegate: Spawn agents → receive summaries (not raw content)
  • Query lazily: Ask Doc Agent questions (don't load full docs)
  • Progressive disclosure: Return structure + excerpts (not full files)
  • Checkpoint regularly: Every major milestone

Related

If Level Action
Low Continue normally
Medium /checkpoint before complex work
High Consider /compact, delegate heavily
Critical /compact immediately