The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute shell commands retrieved from a local configuration file (~/.bonfire/source.json). These commands, such as node ~/.bonfire/skills/clickup/query.mjs, are executed with interpolated task IDs extracted from user-provided URLs or text.
[REMOTE_CODE_EXECUTION]: A 'Discovery' feature uses npx skills find <keyword> to search for and potentially recommend the installation of third-party tools or scripts based on input text. This allows for the introduction of external code into the execution environment at runtime.
[CREDENTIALS_UNSAFE]: The skill configuration explicitly references and checks for .env files (e.g., ~/.bonfire/skills/clickup/.env) to verify the environment. Accessing these files is a high-risk activity as they typically contain sensitive API tokens or credentials required to access task platforms.
[EXTERNAL_DOWNLOADS]: Through the use of npx, the skill facilitates the fetching and execution of packages from the NPM registry during the discovery phase.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (task links and descriptions) and interpolates them into shell commands. Although it specifies alphanumeric-only sanitization for IDs, this ingestion surface represents a risk for command injection if sanitization logic is bypassed or if match patterns are overly broad.

github-workflow