The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including news articles, SEC filings, and community discussion topics, creating a surface for indirect prompt injection.
Ingestion points: Data enters the agent context through the output of longbridge news, longbridge filing, and longbridge topic CLI commands as defined in references/news.md, references/topic.md, and references/sec-filings.md.
Boundary markers: The instructions in the reference files do not prescribe the use of strict delimiters or explicit 'ignore embedded instructions' warnings when processing the fetched content.
Capability inventory: The skill has active write capabilities, including longbridge topic create and longbridge topic create-reply (documented in references/topic.md), which could be misused if the agent is influenced by malicious instructions embedded in ingested community content.
Sanitization: Although the skill includes a mechanism to filter 'hype vocabulary' in references/news.md, it lacks robust sanitization to prevent the interpretation of data as instructions.
[COMMAND_EXECUTION]: The skill heavily relies on the execution of the longbridge CLI tool to perform its functions.
Evidence: SKILL.md and the reference files list numerous shell commands (e.g., longbridge news, longbridge filing detail, longbridge topic search) that the agent is expected to execute in the user's environment to retrieve and manipulate data.

longbridge-content