context-budget-audit
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXPOSURE]: The skill performs local file system reads on agent configuration files and session history logs to measure context usage. It incorporates significant privacy protections by using a script that explicitly filters out files with sensitive keywords (e.g., 'auth', 'secret', 'token', 'key') and instructs the agent to only report usage metadata such as file paths and counts rather than raw transcript content.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from session history files. This risk is mitigated by using a specialized Python script that extracts usage patterns and summarizes evidence before passing it to the agent, avoiding the ingestion of large blocks of raw, un-sanitized text into the LLM context. The instructions emphasize reporting metadata like counts and timestamps to maintain a security boundary.
- [FILE_SYSTEM_ACCESS]: The skill facilitates the removal of unused agent resources (skills, commands, subagents) upon user request. These file system modifications are limited to specific, user-confirmed files and include validation steps, such as ensuring a backup exists in the local repository before removing a global store copy.
- [COMMAND_EXECUTION]: The skill executes a bundled Python script (
scripts/audit_context_budget.py) to perform the budget audit. The script is distributed as part of the skill, uses only standard library components, and is designed for read-only inventory and evidence scanning without performing network operations.
Audit Metadata