Session Observer

Observe OpenClaw session usage, token consumption, context pressure, and model/runtime state. Use when the user asks about token usage, context size, model i...

Security Scan

The skill is an instruction-only observer that asks the agent to call an existing session_status tool and summarize session/token/context metrics — its requests and instructions are consistent with that purpose.

Purpose & Capability

Name and description match the SKILL.md: the skill's goal is to inspect session state (tokens, context, cache, model). It requests no credentials, binaries, or installs that would be unrelated to that purpose.

Instruction Scope

Runtime instructions are narrowly scoped: call session_status, extract a short set of fields, interpret them, and recommend a small next step. It does not ask to read unrelated files, environment variables, or transmit data to external endpoints.

Install Mechanism

Instruction-only skill with no install spec and no code files — nothing is written to disk or downloaded by the skill itself.

Credentials

No environment variables, credentials, or config paths are required. The skill's needs are minimal and proportional to observing session state.

Persistence & Privilege

Does not request always:true and is user-invocable only; it does not attempt to modify other skills or system-wide settings.

Assessment

This skill is narrowly focused and doesn't request credentials or install anything. The main consideration is that it relies on the platform tool session_status — installing it grants the skill (when invoked) the ability to read whatever session metadata session_status exposes. If you are comfortable with session metadata being viewable to invoked skills, this is coherent; otherwise restrict who can invoke the skill or review what session_status returns before installing.

Overview

Inspect OpenClaw session state with minimal disruption. Prefer session_status first, then summarize token usage, context pressure, cache ratio, model/runtime details, and any obvious follow-up actions.

Workflow

1. Start with session status

Use session_status first when the user asks about:

• token usage
• model in use
• context usage
• cache hit rate
• premium/chat budget remaining
• session health

If the user refers to another session, target that session explicitly.

2. Extract the practical signals

Report the fields most useful to the user:

• current model
• input/output token counts
• context used and total context window
• cache hit ratio
• budget or usage percentage when available
• runtime mode and whether reasoning is enabled

Do not dump raw status cards unless the user asks.

3. Interpret, do not just repeat

Classify the situation:

• healthy: low or moderate context pressure, no obvious issues
• watch: context is growing, cache low, or usage notably climbing
• high pressure: context is large enough that compaction, a new thread, or cleanup may help

4. Recommend the smallest next step

Examples:

• continue as-is
• start a fresh thread for a new topic
• reduce unnecessary logs or verbose outputs
• use a sub-agent for a large parallel task
• check a specific session if the user suspects cost spikes

Output format

Use:

• Current state
• What it means
• Recommended next step

Keep it short unless the user asks for detail.

Safety rules

• Do not invent cost numbers not shown by the tool.
• If a field is unavailable, say unavailable.
• Prefer session_status over guesswork.