Token Budget Guard

Automatically manages and compresses context to optimize token usage by summarizing, selectively loading, and budgeting for tool schemas, history, and tasks.

This skill appears to be what it claims — a set of best-practice instructions to reduce token usage — but exercise caution before installing it into a live agent environment. Recommendations: - Review the SKILL.md yourself for any lines that explicitly instruct an agent to overwrite its system prompt, ignore safety rules, or call out to unknown endpoints; the scanner found a possible 'system-prompt-override' pattern which may be a false positive but should be checked. - Do not copy the file into global or shared agent skill directories until you confirm it only contains benign guidance; prefer installing into a sandboxed/test agent first. - If you let an agent follow the examples (jq, grep, head), ensure the agent is not given blanket filesystem access to sensitive directories (e.g., /etc, ~/.ssh, cloud credential files). Limit the agent's file-read scope to project folders. - Monitor the agent's behavior during initial runs: log commands the agent issues and any files it reads, and validate summaries before they are used for privileged actions. - If you need higher assurance, ask the skill author for provenance or a signed release; absent that, keep confidence moderate and treat the skill as useful but requiring manual oversight.

Type: OpenClaw Skill Name: token-budget-guard Version: 2.0.0 The bundle is a collection of Markdown-based instructions and templates designed to optimize token usage and manage costs for AI agents. It promotes efficient practices such as progressive disclosure of tool schemas, conversation summarization, and selective file reading using standard utilities like grep and jq. No malicious code, data exfiltration attempts, or harmful prompt injections were identified across SKILL.md, README.md, or the cost tracking template.