The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the generation of shell commands intended for the Claude Code CLI. These commands can include file modifications, bash execution, and repository analysis. The instructions guide the user toward restrictive flags like --tools and --allowedTools to limit the scope of execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes natural language tasks to be delegated to an external agent (Claude Code).\n
Ingestion points: User-provided strings passed to the claude -p flag in SKILL.md and references/examples.md.\n
Boundary markers: The instructions recommend using --tools to define hard boundaries and specific permission modes to reduce accidental obedience to embedded instructions.\n
Capability inventory: The delegated agent (Claude Code) has broad capabilities including Bash, Read, Edit, and Write as noted in the Safe Edit Run example in SKILL.md.\n
Sanitization: The skill relies on the internal safety guardrails of the Claude Code CLI and explicitly advises the user to choose the 'least-permissive suitable permissions' to mitigate risks.\n- [SAFE]: The installation guidance references @anthropic-ai/claude-code via NPM, which is an official package from a trusted organization. The documentation also points toward official setup guides.

claude-skill