The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes data from external APIs and user-provided pastes which could contain malicious instructions.
Ingestion points: External telemetry from connectors (Hacker News, Product Hunt, App Store, GDELT) and user-pasted content in SKILL.md.
Boundary markers: Includes a clear instruction for the agent to treat all API responses and pasted content as untrusted input and to ignore any embedded instructions.
Capability inventory: Utilizes the WebFetch tool and performs read/write operations on local files within the memory/ directory.
Sanitization: Explicitly warns against following instructions found in scraped or external content, maintaining a security boundary.
[EXTERNAL_DOWNLOADS]: The skill uses the WebFetch tool and specialized connectors to retrieve launch telemetry from well-known services including Hacker News (via Algolia/Firebase), Product Hunt, and GDELT.
[DATA_EXFILTRATION]: The skill interacts with local data stored in the memory/ directory, such as memory/launch-registry/ and memory/hot-cache.md. While it reads these files, there is no evidence of unauthorized exfiltration to external domains beyond the intended monitoring APIs.

launch-monitor