The Agent Skills Directory

[SAFE]: No malicious patterns, such as prompt injection, persistence mechanisms, or obfuscation, were detected in any of the skill files. The skill functions as a structured orchestrator for marketing planning.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses tools like WebFetch and Bash, but the instructions focus on processing local marketing artifacts (e.g., icp-research.md). No hardcoded credentials or unauthorized network exfiltration patterns were found.
[COMMAND_EXECUTION]: While the Bash tool is requested in the YAML frontmatter, it is used for local file operations within the agent environment. The scripts do not contain commands that execute unsanitized external content or perform privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from research/icp-research.md and user-provided briefs to generate agent prompts. However, the orchestrator employs a critic-agent loop to validate outputs against a quality rubric, which serves as a mitigation layer. Severity: LOW.

campaign-plan