Shed
PassAudited by ClawScan on May 1, 2026.
Overview
This instruction-only context-management skill is coherent and benign, but it may cause the agent to save task summaries and delegate work, so users should avoid persisting secrets.
This skill is safe to install as an instruction-only aid, but treat its saved memory files as project artifacts: keep them in a known location, review them for accuracy, and avoid storing secrets or sensitive data.
Findings (3)
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.
The agent may save condensed information from tool outputs to files, which could include sensitive task details if not reviewed.
The skill encourages file writes based on tool outputs. This is central to its context-hygiene purpose, but users should be aware it may create local files containing task-derived information.
When a tool returns large output ... immediately write the key facts to a file or compress into bullets.
Use a clear workspace for these files and avoid saving secrets, credentials, or unnecessary private data in extracted notes.
Future agent sessions may rely on saved summaries, and those summaries may retain private or inaccurate context.
The skill explicitly creates persistent memory breadcrumbs for future use. This is purpose-aligned, but persistent summaries can become stale, incomplete, or contain sensitive information.
Before switching: write what you did, what's next, and where the files are to `memory/YYYY-MM-DD.md`.
Review memory files periodically, keep them scoped to the project, and do not include secrets or sensitive personal data.
A delegated sub-agent may receive project details and perform work based on a summarized prompt, so mistakes or omissions could affect results.
The skill recommends delegating work to sub-agents. It includes a useful scoping instruction, but delegation still transfers task information and authority to another agent context.
Spawn fresh-context sub-agents for complex sub-tasks. ... Give them a clean prompt with just what they need.
Provide sub-agents only the minimum necessary context and require explicit approval for any sensitive or high-impact actions.