The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: No instances of sensitive file path access (such as .ssh, .aws, or .env) or network transmission commands (like curl or wget) were detected. The skill instructions emphasize data minimization and the proper categorization of user context to protect sensitive information.
[REMOTE_CODE_EXECUTION]: The package does not contain patterns for downloading or executing code from external sources. The Python scripts located in the scripts/ directory are local utility tools used exclusively for linting and structural validation of the skill's content.
[DYNAMIC_EXECUTION]: No usage of dangerous runtime execution patterns like eval(), exec(), or subprocesses involving external input was found. The included scripts use the ast module for static parsing and analysis without executing the target files.
[OBFUSCATION]: All instructions, configuration files, and utility scripts are provided in plain text or standard JSON formats. No evidence of Base64 encoding, zero-width characters, homoglyphs, or other obfuscation techniques was identified.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied notes and artifacts. It includes explicit instructions to categorize this context and ignore non-material data, which mitigates the risk of external instructions being inadvertently obeyed. The skill lacks exploitable capabilities like network access or persistent file writes.
[PROMPT_INJECTION]: The skill uses structured role-playing for stakeholder analysis but does not contain instructions intended to override system prompts, bypass safety filters, or reveal internal agent configurations.

prime-vector