The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a prebuilt local binary named newsjack and an internal Python script scripts/build_report.py to process news signals and generate reports. It interacts with the local file system to store artifacts in a designated run directory and manages binary paths such as ~/.newsjack/bin/.
[PROMPT_INJECTION]: The skill monitors untrusted external data from news and social media sources, creating a surface for indirect prompt injection. 1. Ingestion points: News and social media clusters retrieved via the newsjack CLI. 2. Boundary markers: References internal doctrine files such as ETHICS.md and WHY-NOT-SPAM.md as absolute behavioral constraints. 3. Capability inventory: Shell access via the newsjack tool, local file system writes for artifacts, and network operations for content retrieval. 4. Sanitization: The pipeline employs tiered filtering passes (coarse and story-origin) to evaluate signals, but lacks explicit string-level sanitization for ingested content.
[EXTERNAL_DOWNLOADS]: The monitoring engine connects to several established and reputable RSS feeds for data gathering, including official government portals for the FTC, SEC, and GOV.UK, as well as well-known news aggregators like Techmeme and Google News. These connections are documented as safe and functional for the skill's intended purpose.

newsjack-detector