The Agent Skills Directory

[PROMPT_INJECTION]: The template provided in standard/MAIN_AGENT_CLAUDE.md contains instructions that override the agent's default assistant persona, directing it to "Act as a peer, not an assistant" and to "Scrutinize the user's suggestions and claims."
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes untrusted project files during its migration and audit workflows.
Ingestion points: The skill reads existing CLAUDE.md and README.md files from the target project path in workflows/migrate.md and workflows/audit.md.
Boundary markers: The instructions lack boundary markers or warnings to prevent the agent from following instructions embedded within the processed project files.
Capability inventory: The agent can read files (cat, ls), write files (installing the CLAUDE.md template), and execute shell commands (find, wc).
Sanitization: The skill does not perform any sanitization or filtering of the content read from external project files.
[COMMAND_EXECUTION]: The skill uses various shell commands, including find, ls, cat, and wc, to inspect the filesystem and audit project structure. These commands are executed locally based on the provided project path.

managing-claude-code-meta