higgsfield-fx-tuner
Warn
Audited by Gen Agent Trust Hub on May 8, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill implements a self-modification mechanism where it programmatically updates its own source code (SKILL.md) and CHANGELOG.md. This involves generating new instructions, best practices, and vocabulary tables at runtime based on patterns extracted from a log file.
- [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by ingesting untrusted natural language feedback from users and incorporating it into the agent's core instructions. ● Ingestion points: The
feedback-log.mdfile (SKILL.md capture mode). ● Boundary markers: Absent; feedback is stored as raw text strings without delimiters or escaping. ● Capability inventory: The skill uses Bash tools to perform file-write operations, directory creation, and versioned backups across the skill directory. ● Sanitization: Absent; the skill relies on the agent's pattern analysis and a final human approval step rather than technical sanitization or schema validation. - [COMMAND_EXECUTION]: The skill uses shell commands via the Bash tool to monitor and manage its environment. This includes directory existence checks, file line counting (
wc -l), and directory listing (ls) to calculate status and maintain a rotation of the last 5 backups in the.backup/directory.
Audit Metadata