Shed

Context window hygiene for long-running LLM agents. Decision rules for when and how to compress, mask, switch, or delegate context — backed by research (JetB...

This skill appears coherent and matches its purpose, but before installing: (1) confirm where 'memory/YYYY-MM-DD.md' and other files will be stored, who can read them, and how long they are retained; (2) test it in a sandboxed agent runtime to verify file-IO and sub-agent behavior; (3) ensure the runtime masks or filters any sensitive data before writing; (4) note the package has no source/homepage listed and an anonymous owner — if you need provenance or accountability, request the source or prefer a skill with clear authorship. If those controls are acceptable, the skill's instructions are reasonable for managing context.

Type: OpenClaw Skill Name: shed Version: 1.0.0 The 'shed' skill bundle provides instructions for an AI agent to manage its context window, including directives to 'write key facts to a file', 'write results to file', and 'leave breadcrumbs' in `memory/YYYY-MM-DD.md` (SKILL.md). While the stated purpose is context hygiene, these instructions grant the agent broad file write capabilities. If the agent processes sensitive information, these directives could lead to confidential data being written to local files without explicit sanitization or content restrictions, posing a data leakage risk. This is classified as suspicious due to the potential for unintentional sensitive data exposure through file writes, even though there is no clear evidence of malicious intent like exfiltration to external servers or backdoor installation.