The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches resources and updates from several external locations, all of which are official vendor sources or well-known CDNs. These include data from 'alva-ai-static.b-cdn.net', library components from 'unpkg.com/@alva-ai' and 'cdn.jsdelivr.net', and version updates from 'github.com/alva-ai/skills'.
[COMMAND_EXECUTION]: Includes a shell script 'scripts/version_check.sh' used to compare the local skill version with the latest release on GitHub. The skill also executes commands through the platform's CLI tool '@alva-ai/toolkit' and uses standard utilities like 'yt-dlp' for transcript processing.
[PERSISTENCE]: The platform provides native persistence mechanisms through 'alva schedule' for future agent turns and 'alva deploy create' for scheduled cronjobs. These are core architectural features designed to maintain ongoing financial monitors and automated data pipelines.
[DYNAMIC_EXECUTION]: Employs a custom 'jagent' runtime based on V8 to execute JavaScript for data transformations and strategy logic. It also supports 'User Defined Functions' (UDFs), ONNX model inference, and scheduled LLM reasoning steps ('alpi'), all of which are managed through the vendor's registration system.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external content from social media (X, Reddit), news feeds, and YouTube transcripts.
Ingestion points: Data is ingested via 'sourceDigestItems' in 'references/fintwit-digest-sdk.md' and various search modules in 'references/search.md'.
Boundary markers: While not enforced in data formats, instruction-level guidance in 'references/user-facing-prose.md' (the 'VOICE block') mandates clear separation between factual data and AI-generated narrative.
Capability inventory: The skill has access to sensitive capabilities including file writing via 'alfs', network requests via 'net/http', and financial trading execution via 'alva trading'.
Sanitization: Relies on robust prompting guidelines and explicit labeling requirements to mitigate the risk of external content influencing agent behavior or trading decisions.

alva