The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to ingest and answer questions based on external, potentially untrusted sources, which could lead to manipulation of agent behavior if the source content contains malicious prompts.
Ingestion points: Data is brought into the environment via the btca add and btca ask commands, which fetch content from Git repositories, local directories, and NPM packages.
Boundary markers: The skill body lacks instructions or markers that tell the agent to disregard instructions found within the retrieved data sources.
Capability inventory: The skill facilitates the execution of the btca tool to process and query external data.
Sanitization: There is no evidence of content sanitization or validation for the hydrated resources mentioned in the instructions.
[EXTERNAL_DOWNLOADS]: The skill provides examples of downloading data from external sources via CLI commands.
Source: Instructions include fetching from Git repositories (e.g., github.com/sveltejs/svelte.dev) and NPM registries.
[DATA_EXPOSURE]: The skill documentation references global configuration files on the host system.
Sensitive file paths: References the location of global configuration at ~/.config/btca/btca.config.jsonc.

btca-cli