agentsop-context-scope-discipline
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains legitimate instructions for context management and token budget discipline. No attempts to override safety filters, bypass system prompts, or implement jailbreak patterns were detected.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill does not access sensitive local files, hardcode credentials, or perform network operations. URLs referenced point to official documentation for well-known developer tools.
- [OBFUSCATION]: Analysis of the text and metadata found no evidence of Base64 encoding, zero-width characters, homoglyphs, or other techniques used to hide malicious content.
- [REMOTE_CODE_EXECUTION]: No remote script downloads or piped command executions are present. The skill references standard CLI commands (/add, /drop, /tokens) for use within a coding agent harness.
- [DYNAMIC_CONTEXT_INJECTION]: The skill does not utilize dynamic shell execution patterns (e.g., the
!commandsyntax) to execute arbitrary code at load time. - [INDIRECT_PROMPT_INJECTION]: While the skill involves the agent reading external files, it establishes strict boundaries and suggests minimizing context volume, which reduces the potential surface area for indirect injection attacks compared to unrestricted ingestion.
Audit Metadata