The Agent Skills Directory

[DYNAMIC_EXECUTION]: The hook scripts apex-lsp-validate.py and post-tool-validate.py modify the Python system path (sys.path) at runtime to import modules from computed paths located in the user's home directory (~/.claude/). This practice of loading code from mutable external locations represents a security risk if the directory permissions are not strictly controlled.
[COMMAND_EXECUTION]: The skill invokes external command-line tools to perform its functions. Specifically, prettier-format.py uses the subprocess module to execute the Prettier formatter, and apex-lsp-validate.py relies on a shell wrapper (apex_wrapper.sh) to interface with the Apex Language Server. These executions interact with the underlying host system and depend on the integrity of the installed binaries.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an automated feedback loop where diagnostics generated from analyzing user-supplied Apex code are fed back into the agent's context to trigger "auto-fix" operations. This design creates an attack surface where maliciously crafted comments or code patterns within the files being reviewed could potentially influence the agent's instructions or lead to the generation of insecure code.

sf-apex