The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents how to execute system commands using the platform's bash execution feature (! syntax) to gather repository and environment context. Examples include using git status, npm test, and custom plugin scripts. The documentation emphasizes security best practices, such as using command filters in the allowed-tools frontmatter field (e.g., Bash(git:*)) to adhere to the principle of least privilege.
[DYNAMIC_CONTEXT_INJECTION]: The skill provides instructions on using the !command`` syntax to inject dynamic content into an agent's context at load time. This is presented as a core platform feature for building context-aware slash commands. The documentation includes examples of using this for git repository state and version checks.
[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for interpolating user-supplied arguments ($1, $2, $ARGUMENTS) into prompts and shell commands. This creates a potential injection surface if the inputs are not sanitized. The skill explicitly addresses this risk in the "Validation Patterns" section of SKILL.md and through practical examples in plugin-commands.md, providing remediation guidance such as using grep or script-based validation to verify inputs against an allowlist before processing.
Ingestion points: User-supplied positional arguments ($1, $2) and the $ARGUMENTS variable defined in SKILL.md and several examples.
Boundary markers: Absent in basic prompt examples, but recommended in the validation documentation.
Capability inventory: The skill utilizes allowed-tools: Bash(*), Read, Write, and the AskUserQuestion tool across various scripts and examples.
Sanitization: The skill provides detailed patterns for sanitizing and validating external content using bash-based regex checks and script-driven validation loops.

command-development