The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/find_project.py executes system commands including ripgrep (rg), macOS Spotlight (mdfind), and find. These executions use the subprocess.run method with arguments passed as a list, which effectively prevents shell injection vulnerabilities by avoiding the use of a shell to interpret the command string.
[DATA_EXPOSURE]: The skill accesses local file system metadata and AI chat logs stored in ~/.claude/projects and ~/.gemini. This access is necessary for its primary purpose of finding project paths based on historical context. The logic is strictly local, with no evidence of data being transmitted to external servers or non-whitelisted domains.
[SAFE]: The scripts/profile_store.py utility handles persistent user settings using security-conscious patterns. It implements a blacklist of sensitive keywords (e.g., 'secret', 'token', 'password', 'credential') to prevent the storage of secrets in the user profile. Furthermore, it performs atomic writes using temporary files and applies restricted file permissions (0600) to ensure data integrity and confidentiality.
[INDIRECT_PROMPT_INJECTION]: The skill processes AI chat logs, which are external data sources that could theoretically contain malicious instructions. However, the skill limits its processing to path extraction and keyword matching, and returns the findings in a structured JSON format for the agent to interpret, which serves as a mitigation against accidental instruction execution.

lov-search-project