launch-retro-analyzer
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external marketing data (GA4 exports, dashboard screenshots, CSVs, and telemetry from platforms like Hacker News and Product Hunt). While the skill includes a specific warning to treat these exports as untrusted input and never follow embedded instructions, the high level of capability (invoking external scripts, proposing data to registries, and writing to memory) creates a surface for indirect prompt injection if an attacker-controlled export contains malicious instructions.
- Ingestion points: Reads UTM-attributed web analytics exports, dashboard screenshots, and community platform threads (
SKILL.md). - Boundary markers: Explicitly instructs the agent to treat exports as untrusted input and ignore embedded instructions (
SKILL.mdInstructions section). - Capability inventory: Proposes data to
memory/events/launches.ndjsonviaregistry-events.py, executes telemetry connectors (hn.py,producthunt.py,appstore.py,gdelt.py), and writes to local memory paths (SKILL.md). - Sanitization: Relies on a negative constraint (instructional warning) rather than programmatic sanitization or strict schema enforcement of the incoming text/CSV data.
Audit Metadata