The Agent Skills Directory

[SAFE]: The skill implements secure local configuration management through scripts/profile_store.py. This script explicitly forbids storing sensitive data types such as tokens, secrets, and passwords in the profile records, preventing user credentials from accidental persistence in local storage.\n- [SAFE]: The tool uses atomic write operations and enforces restricted file permissions (mode 0600) on its configuration files. This ensures data integrity and prevents unauthorized local access to user preferences or repository migration records.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external repository metadata, including branch names, Git history, and LICENSE files, which constitutes a vulnerability surface. Ingestion points: Git repository metadata and branch information read during workflow steps 1 and 2 in SKILL.md. Boundary markers: Absent in the script logic, although the instructions specify separating analysis, preview, and modification phases. Capability inventory: Executes Git CLI commands and performs local file writing via scripts/profile_store.py. Sanitization: Workflow step 4 explicitly requires checking for sensitive files and待提交范围 (to-be-committed range) before pushing to remote repositories.

lov-repo-takeover