The Agent Skills Directory

[METADATA_POISONING]: The skill metadata identifies the 'source_type' as 'official' and references the vendor's repository as the source. However, the skill is authored by a third party ('sickn33'), which misrepresents the skill's origin and trust level.
[COMMAND_EXECUTION]: The skill's primary function is to execute shell commands using the longbridge CLI tool. This requires the agent to have shell access, which is a significant capability that could be misused if the agent is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill lacks sufficient safeguards against malicious input during command construction.
Ingestion points: Untrusted user input, such as stock tickers and market names (e.g., <SYMBOL>), is taken directly from the user prompt and used to build shell commands.
Boundary markers: The skill does not use delimiters or provide 'ignore embedded instructions' warnings for the data it processes before passing it to the CLI.
Capability inventory: The skill provides access to a variety of shell-based commands for fetching market data and account information.
Sanitization: The instructions do not include logic for escaping shell metacharacters in user-supplied symbols, which could lead to command injection if an attacker provides a crafted ticker like '700.HK; rm -rf /'.
[EXTERNAL_DOWNLOADS]: The skill's error handling instructions guide the user to install external software via a Homebrew tap (brew install longbridge/tap/longbridge-terminal). This involves the installation of third-party binaries from a remote repository.

longbridge-market-data