Proactively monitors context levels to save critical session information before AI token compaction occurs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install session-watchdog
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install session-watchdog using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Session Watchdog?
Session Watchdog is a specialized management layer for maintaining long-term project context in AI-driven development. It addresses the inherent limitations of LLM token windows by monitoring usage and intervening before critical information is lost through compaction. As a core utility in the Openclaw Skills ecosystem, it ensures that project decisions, task lists, and unresolved issues are safely archived to persistent storage, allowing for seamless session transitions and deep technical continuity.
Session Watchdog Use Cases
- Monitoring real-time token usage during intensive coding sessions.
- Preserving critical project state before the model reaches its context ceiling.
- Automatically documenting decisions and pending tasks to permanent memory files.
- Querying session health and memory status via Openclaw Skills.
- Managing long-running workflows that exceed standard model context limits.
How Session Watchdog Works
- The skill periodically initiates a context check using the session_status command to retrieve current token levels.
- It evaluates the current context against defined thresholds: Safe (<140k), Warning (140k-160k), and Critical (160k+).
- Once the usage exceeds the 80% threshold, it triggers the checkpoint protocol.
- It reads the current daily memory file and appends vital session metadata including decisions and pending tasks.
- It alerts the developer with a warning status and saves the checkpoint to persistent storage.
- It prompts for user intervention to decide whether to continue, summarize, or restart the session context.
Session Watchdog Setup
Session Watchdog is integrated directly into the agent workflow. Ensure your environment is configured to allow file writes to the memory directory.
# Verify the ability to check session status
session_status
Session Watchdog Data Schema & Taxonomy
The skill organizes data into chronological Markdown files located in the memory/ directory. Each checkpoint follows a specific structure:
| Section | Description |
|---|---|
| Decisions | High-level architectural or logical choices made during the session. |
| Pending | A checklist of tasks that remain uncompleted. |
| Important Context | Project state, configurations, and paths of modified files. |
| Unresolved | Specific bugs or issues requiring follow-up in the next session. |
Session Watchdog Advanced Features
- Automated triggers every 30k tokens to ensure no data loss between checkpoints.
- Context-aware warnings that prevent starting large tasks when the limit is near.
- Standardized memory file formatting for compatibility with other Openclaw Skills.
- Configurable safety thresholds for different LLM context architectures.
SKILL.md
Loading
Related Openclaw Skills
session-wrap-up
An end-of-session automation tool that synchronizes code, extracts technical learnings, and evolves project rules using AI pattern detection.
Self-Hosted Whisper API
A local, privacy-focused speech-to-text solution that leverages OpenAI's Whisper model on Kubernetes for fast audio transcription and translation.
Commute Traffic Checker (TomTom)
A TomTom-powered AI skill that provides real-time traffic analysis, geocoding, and travel recommendations for any route.
Taskleef
Taskleef is a versatile terminal-based task manager that syncs with Taskleef.com to handle todos, subtasks, projects, and visual kanban boards.
BillClaw
A local-first financial management skill that syncs bank transactions and email bills while maintaining complete data sovereignty.
CAN ASK MCP
A local, verifiable naming and storage layer that allows AI agents to stamp and log data retrieved from tools via the Model Context Protocol.
Featured*
