The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes financial news article headlines and descriptions retrieved from external sources, which constitutes a potential attack surface for indirect prompt injection where malicious content in news articles could attempt to influence the agent.
Ingestion points: Articles fetched from the newsapi.org API (Step 3a) and headlines gathered via WebSearch from external sites like Reuters, Bloomberg, and Seeking Alpha (Step 3b).
Boundary markers: Absent. There are no explicit instructions or delimiters used to isolate the news content from the rest of the prompt or to instruct the agent to ignore commands within the data.
Capability inventory: The skill possesses the ability to execute shell commands (bash), perform web searches (WebSearch), and fetch content from URLs (WebFetch).
Sanitization: The skill does not implement any sanitization or validation of the fetched news text before scoring and computing sentiment.
[COMMAND_EXECUTION]: The skill uses a shell command to read a local configuration file.
Evidence: Step 2 instructs the agent to run cat ${CLAUDE_PLUGIN_ROOT}/config.json to retrieve the newsapi.key. While intended for configuration management, this loads the raw API key directly into the agent's interaction context.

news-sentiment