The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various external social media sources to synthesize community sentiment briefs.
Ingestion points: Content is fetched in SKILL.md using web_search, web_extract, x_search, and youtube-content skills.
Boundary markers: A deterministic Python ranking script (scripts/rank.py) acts as a filter and aggregator for the raw search results.
Capability inventory: The skill uses python3 to run the ranker, executes the gh CLI, and utilizes platform tools for network searching and extraction.
Sanitization: The pipeline includes a pruning step for low-relevance items and requires a synthesis step that transforms raw data into prose, though it does not explicitly mention escaping of injected instructions.
[DYNAMIC_EXECUTION]: The skill executes a bundled script at runtime and provides instructions for creating temporary verification scripts.
Evidence: SKILL.md instructs the agent to execute scripts/rank.py using python3 or uv run to process gathered data.
Evidence: references/research-to-experiment.md describes the creation and execution of temporary verifiers with a specific hermes-verify- prefix.
[COMMAND_EXECUTION]: The skill uses command-line tools to interact with external services.
Evidence: SKILL.md utilizes the gh (GitHub) CLI to search repositories and fetch metadata via the GitHub API.

mob-check