home / skills / git-fg / thecattoolkit / operating-claude

operating-claude skill

safe

/plugins/sys-cognition/skills/operating-claude

This skill helps you optimize Claude runtime context usage, troubleshoot performance, and understand session mechanics for smoother operations.

npx playbooks add skill git-fg/thecattoolkit --skill operating-claude

Review the files below or copy the command above to add this skill to your agents.

Files (5)

SKILL.md

1.0 KB

---
name: operating-claude
description: "Operational manual for driving the Claude runtime. Use when needing to optimize context usage, troubleshoot performance, or understand runtime mechanics. Do not use for architectural definitions or building plugins."
allowed-tools: [Read]
---

# Operational Mechanics

## Runtime Operations

### 1. Context Management
- **Monitoring**: Check tokens via `/status` or observing response latency.
- **Optimization**: Use `managing-context` for compression strategies.
- **Degradation**: Performance drops >40% usage. Reset or fork session.

### 2. Session Control
- **Plan Mode**: Double `Shift+Tab` for high-reasoning tasks.
- **Forking**: Isolate heavy tasks to avoid polluting main context.
- **Checkpoints**: Use `_state.md` to persist progress across sessions.

## Architecture References

For strict definitions of how the system is built, refer to the core documentation:

- **Plugin Architecture**: `docs/guides/infrastructure.md`
- **Skill Standards**: `docs/guides/skills.md`
- **Engineering Patterns**: `docs/REFERENCES.md`

Overview

This skill is an operational manual for driving the Claude runtime, focused on practical guidance for context management, session control, and troubleshooting performance. It helps operators optimize token usage, isolate heavy tasks, and persist progress across sessions. Do not use this for architecture design or plugin development.

How this skill works

The skill inspects runtime indicators such as token consumption and response latency to identify context pressure and performance degradation. It provides concrete actions: check token counts via the /status endpoint, apply compression strategies from managing-context, fork sessions to isolate load, and reset or create checkpoints when usage exceeds safe thresholds. It highlights quick controls like Plan Mode and session forking to reduce context pollution.

When to use it

When response times increase or latency becomes inconsistent.
When token usage approaches capacity or you need to monitor /status.
When running heavy reasoning tasks that may pollute the main session.
When you need to persist intermediate progress across restarts.
When performance drops more than ~40% and a reset or fork is required.

Best practices

Regularly check /status for tokens and correlate with latency metrics.
Use compression strategies (managing-context) proactively to extend context life.
Fork sessions for isolated experiments or heavy processing to protect main context.
Use Plan Mode (double Shift+Tab) for complex reasoning to improve output quality.
Create checkpoints (_state.md) to persist progress and enable reliable recovery.

Example use cases

Optimizing a long-running assistant conversation by compressing prior messages to free tokens.
Forking a session to run a high-cost data extraction task without affecting the primary workflow.
Detecting a >40% performance drop, then resetting or forking the session to restore responsiveness.
Using Plan Mode for multi-step reasoning tasks to improve answer coherence and accuracy.
Saving periodic checkpoints in _state.md to allow rollback after experiments or outages.

FAQ

How do I detect context pressure quickly?

Monitor /status for token counts and watch for increased response latency; both are clear indicators of context pressure.

When should I fork a session instead of resetting?

Fork when you want to preserve the main session state but isolate heavy or experimental tasks; reset when you prefer a clean slate.