The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a series of shell commands, including grep, ls, and sed, to analyze the local repository. These commands are used solely for read-only codebase discovery to identify domain structures and project conventions.
[INDIRECT_PROMPT_INJECTION]: The skill generates new instructions by processing content from the user's local files, which could theoretically contain untrusted data.
Ingestion points: The agent is directed to scan files such as Gemfile, CLAUDE.md, and directory structures within app/ and spec/ to gather context.
Boundary markers: The generated instructions do not include specific delimiters or warnings to ignore embedded content within the analyzed files.
Capability inventory: The skill utilizes shell execution for discovery and file-writing capabilities to create new SKILL.md files in the .claude/skills/ directory.
Sanitization: The workflow mitigates risks by requiring the agent to present all drafts to the user for manual review and correction before they are finalized.
[DYNAMIC_EXECUTION]: The skill involves the dynamic generation of new instruction files (SKILL.md) based on the analysis of the local codebase and user input. These files are intended to be loaded into the agent's context for future project-specific tasks.

scaffold-project-skills