The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes several bash scripts (e.g., agentic.sh, topic-1.sh, internal-ad-pentest-sop-assume-b.sh) that interactively collect multiline user input for inclusion in reports. This input is interpolated directly into a markdown buffer without sanitization or boundary markers, creating an indirect prompt injection surface. If these reports are subsequently processed by an AI agent, malicious instructions contained within the user-provided findings could potentially override the agent's context or behavior.
Ingestion points: The ms_read_multiline function in cli/lib/common.sh, utilized across all CLI workflow and decision scripts to gather user findings.
Boundary markers: None are present in the final output; the user input is appended as raw markdown content.
Capability inventory: The CLI environment includes file-write capabilities via ms_emit_md_file and JSON emission.
Sanitization: No filtering or escaping is applied to the input before it is stored in the report buffer.
[COMMAND_EXECUTION]: The provided bash scripts perform local file system operations to save generated reports. While functional, this involves direct interaction with the host's file system through script execution.
Evidence: The ms_emit_md_file function in cli/lib/common.sh writes the MS_REPORT_BUFFER to a local file using a redirect operator.

cybersecurity-red-team-master