Indie Hacker
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to gather external data from Twitter, Reddit, and competitor sites for validation and research purposes, which constitutes untrusted input.
- Ingestion points: Untrusted external data is ingested during problem confirmation and competition analysis phases, specifically targeting platforms like Twitter and Reddit, and stored in
projects/*.mdfiles. - Boundary markers: Absent. There are no instructions for the agent to treat external research data as data rather than instructions, nor are there delimiters used in the memory templates to separate untrusted content.
- Capability inventory: The skill possesses significant capabilities via Rule 6 in
SKILL.md, authorizing the agent to write code, run scripts, and configure tools (e.g., CI/CD) without user intervention. - Sanitization: Absent. No logic is provided to filter, escape, or sanitize external feedback or competitor data before it is processed by the agent or incorporated into project context.
- [COMMAND_EXECUTION]: Rule 6 ("Execute, Don't Suggest") instructs the agent to perform actions such as running scripts and configuring tools autonomously rather than explaining the process to the user. While intended for developer productivity, this high level of autonomy increases the risk that malicious instructions ingested from external research could lead to unauthorized system actions without human review.
Audit Metadata