The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous examples of shell commands integrated into slash commands, such as npm test, git status, and kubectl cluster-info. These are educational examples intended to teach command development. The skill reinforces security by advising the use of the most restrictive allowed-tools permissions (e.g., Bash(git:*)) and warning against destructive operations.
[INDIRECT_PROMPT_INJECTION]: The skill documents the ingestion of untrusted user input via positional arguments like $1 and $ARGUMENTS.
Ingestion points: User-supplied strings passed to slash commands (documented in SKILL.md and references/plugin-features-reference.md).
Boundary markers: The skill recommends prompt design patterns that help clarify instructions versus user data.
Capability inventory: Developed commands can utilize Bash, Read, Write, and Edit tools as defined in YAML frontmatter.
Sanitization: The skill provides specific validation patterns using tools like grep to verify argument formats (e.g., checking for valid environment names like dev, staging, or prod).
[DYNAMIC_CONTEXT_INJECTION]: The skill documents the !command syntax used to inject shell output into the agent context. Examples such as !date and !git status are used to demonstrate how to provide dynamic environment information. These are standard features of the platform being documented and are not used for malicious purposes in the skill's own content.
[DATA_EXPOSURE]: Technical references mention sensitive file patterns like .env or .claude/debug-logs/ to teach users how to properly manage or debug their commands. No actual credentials or sensitive data are exposed within the skill files.
[SAFE]: Heuristic detections for destructive commands (e.g., rm and dd in references/testing-strategies.md) are false positives. These are used in a controlled context for educational purposes, specifically for creating large test files in /tmp and cleaning them up, or for demonstrating how the platform blocks forbidden commands.

command-development