Context Sentinel
ReviewAudited by ClawScan on May 10, 2026.
Overview
Context Sentinel’s goal is understandable, but it asks users or agents to schedule and trust a missing PowerShell script that can switch models or trigger handoff.
Treat this as a review-required skill. Its purpose is plausible, but do not add it to cron or HEARTBEAT.md until the missing script is supplied, reviewed, and constrained to approved model IDs and explicit handoff behavior.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A user could end up running or later adding an unreviewed script that the agent will trust to decide model changes and handoff behavior.
The reviewed artifacts report no code files and no install spec, so the script that would make context decisions is not included for review. That is a provenance gap for a workflow meant to run periodically.
This skill provides a script ... Run the `check_context.ps1` script ... `powershell -File scripts/check_context.ps1`
Do not schedule or run this skill until the actual check_context.ps1 and any supporting files are included, reviewed, and tied to a trusted source.
If enabled, local code at that path could be executed repeatedly in the user's environment and affect the session without fresh review each time.
The skill directs repeated local PowerShell execution, but the target script is not present in the artifact set and PowerShell is not declared as a requirement.
This skill is designed to be run periodically, either via a `cron` job or as part of the main agent's `HEARTBEAT.md` checklist ... `powershell -File skills/context-sentinel/scripts/check_context.ps1`
Only run the command manually after verifying the exact script contents and path; avoid heartbeat or cron execution until the implementation is reviewed and pinned.
The agent may switch away from the user's preferred model or trigger handoff when the protocol threshold is reached.
Changing models and triggering handoff is the stated purpose, but it directly changes the agent's operating mode based on script output.
The script will return ... `SWITCH_TO:<model_id>` ... `HANDOFF_NOW` ... Based on the output, execute the appropriate agent command.
Use an explicit allowlist of permitted model IDs and require user confirmation for handoff or model changes in important sessions.
The skill can continue affecting future sessions until the heartbeat or cron entry is removed.
The periodic behavior is disclosed and purpose-aligned, but it creates ongoing autonomous influence over the session once configured.
This skill is designed to be run periodically, either via a `cron` job or as part of the main agent's `HEARTBEAT.md` checklist.
Install it only if you want ongoing context monitoring, and keep a clear way to disable the scheduled check.