The Agent Skills Directory

[COMMAND_EXECUTION]: The v3.4 migration guide includes a verification step that executes blueprint-autorun.sh. The command bash "$(dirname "$(command -v blueprint-autorun.sh 2>/dev/null || echo .)")"/blueprint-autorun.sh --report will execute the script from the current working directory if it is not found in the system PATH. This allows for arbitrary command execution if a malicious script with that name is present in a repository being migrated.
[COMMAND_EXECUTION]: The v3.3 migration logic utilizes python3 -c to dynamically calculate relative paths using the os.path.relpath module.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by reading and acting upon data in manifest.json. The migration logic uses keys and values from the manifest to determine targets for file moves and deletions. A maliciously crafted manifest file could utilize path traversal sequences (e.g., ../../) to trick the agent into modifying or deleting sensitive files outside of the intended blueprint directory.

blueprint-migration