skills/mukul975/anthropic-cybersecurity-skills/hunting-for-process-injection-techniques/Gen Agent Trust Hub
hunting-for-process-injection-techniques
Warn
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: Deceptive metadata identified in the skill documentation and license.
- The
authorfield inSKILL.mdis listed as "mahipal", while theLICENSEfile attributes copyright to "mukul975". - The
LICENSEfile contains a post-dated copyright year of "2026", which is deceptive regarding the creation date of the work. - [INDIRECT_PROMPT_INJECTION]: The skill processes external log data which creates a potential vulnerability surface.
- Ingestion points: The
scripts/agent.pyscript reads and parses user-supplied Sysmon JSON logs via thelog-fileargument. - Boundary markers: No boundary markers or "ignore" instructions are present to prevent the agent from potentially obeying instructions embedded within log fields such as
SourceImage,TargetImage, orStartFunctionif the output is processed by an LLM. - Capability inventory: The skill possesses file system read and write capabilities to process logs and generate reports.
- Sanitization: The script performs standard JSON parsing but includes no validation, filtering, or sanitization of the textual content within the log entries.
Audit Metadata