marketing-orchestrator-skill
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSTIME_DELAYED_CONDITIONAL
Full Analysis
- [DYNAMIC_EXECUTION]: The skill implements a 'Revision Loop' that analyzes session logs and prompts the agent to 'Update relevant sub-skill' files. This logic enables the agent to dynamically rewrite its own instructions located in SKILL.md files.
- [PERSISTENCE]: Modifications to sub-skill instructions are written back to the file system, ensuring that changes (including potentially malicious ones) persist across different sessions and become part of the agent's permanent logic.
- [INDIRECT_PROMPT_INJECTION]: The orchestrator ingests untrusted data from
references/revision-log.mdandreferences/client-profiles/. This data influences the self-improvement process, creating a surface where malicious log entries could inject instructions into the skill's source files. Ingestion points: revision-log.md, client-profiles/*.md. Boundary markers: None. Capability inventory: File-write (updating sub-skills), shell command execution. Sanitization: None. - [COMMAND_EXECUTION]: The 'GitHub Deploy' sub-skill generates shell commands for user execution, including software installation (
brew install git) and global environment configuration (git config --global). - [EXTERNAL_DOWNLOADS]: The installation instructions direct the user to clone a repository from a personal GitHub account (
github.com/bryanengel72/marketing-orchestrator.git) to install the tool and its sub-skills. - [TIME_DELAYED_CONDITIONAL]: The 'Revision Loop' logic is triggered conditionally based on a session counter, prompting for analysis and updates specifically every third session.
Audit Metadata