The Agent Skills Directory

[COMMAND_EXECUTION]: The migration scripts extensively utilize subprocess.run to invoke a variety of developer tools, such as git, the Beads CLI (bd), and several package managers (npm, cargo, uv, etc.). This is necessary for the migration process but represents a significant capability surface.\n- [EXTERNAL_DOWNLOADS]: The adopt-template.py script fetches project templates from external Git repositories. By default, it targets the vendor's GitHub repository, but it can be configured to use any remote source, which could lead to the ingestion of untrusted template code.\n- [PROMPT_INJECTION]: The skill processes untrusted project data, including Markdown roadmaps, design documents, and legacy task files. Malicious instructions embedded in these files could potentially influence the agent's actions during the migration, a form of indirect prompt injection.\n
Ingestion points: Ingests data from docs/src/planned-features.md, tasks.md, and other project files via migration_core.py and adopt-template.py.\n
Boundary markers: Uses specific markers like  to delimit managed content, though these do not prevent adversarial injection.\n
Capability inventory: The skill has broad capabilities including file writing, directory renaming, and shell command execution across its various scripts.\n
Sanitization: The scripts use regex and HTML unescaping to parse content, but lack comprehensive sanitization to filter out executable prompt instructions embedded in the project data.

migrate-workflow