Agent Debugger
PassAudited by ClawScan on May 1, 2026.
Overview
This is a benign instruction-only debugging guide, with user-directed examples that should be applied carefully because they can affect exec-tool approvals and persistent memory use.
This skill looks safe to install as an instruction-only debugging aid. Before applying its examples, review any changes to exec-tool permissions, memory-search behavior, or persistent agent instructions, and avoid storing or retrieving sensitive information unnecessarily.
Findings (2)
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.
If a user changes exec settings to broader modes such as full access or disables prompts, the agent could run local commands with less oversight.
The skill documents exec-tool security settings, including options that could reduce command approval prompts. This appears as user-directed troubleshooting guidance for permission errors, not hidden execution.
"exec": { "security": "ask", // or "allowlist" or "full" ... "ask": "on-miss" // or "always" or "off" }Keep least-privilege settings, prefer explicit approval or allowlists, make any permission changes temporarily, and restore safer settings after debugging.
Saved memory may influence future answers or expose prior workspace details in the current session if used too broadly.
The skill encourages retrieving persistent memory files for certain answers. This is coherent with debugging memory behavior, but persistent memory can contain stale, private, or poisoned context.
Before answering anything about prior work, decisions, dates, people, or todos: run memory_search on MEMORY.md + memory/*.md
Search only relevant memory, verify retrieved information before relying on it, avoid storing secrets or sensitive personal data, and clear outdated memory entries.