The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary purpose is to automate a development workflow by orchestrating local CLI tools including git, bd (Beads), gh (GitHub), and mdbook. These interactions are implemented through subprocess.run without using a shell, significantly reducing the risk of command injection while construction arguments programmatically.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from Git repository history and Beads issue descriptions.
Ingestion points: Logic in dstack_delivery.py and dstack_feature.py reads commit messages, issue titles, and documentation files into the agent's context.
Boundary markers: The controller enforces strict JSON schemas for adoption and alignment plans and validates Markdown structures to ensure data integrity.
Capability inventory: The tool has the ability to perform file system writes and execute local development binaries.
Sanitization: It implements extensive validation logic (e.g., FORBIDDEN_DOC_PATTERNS in dstack_commands.py) to prevent transient workflow state or malicious instructions from persisting in durable documentation.
[SAFE]: The skill includes a dedicated security mechanism to prevent the accidental commitment of sensitive Beads runtime and credential files (such as .beads-credential-key and .env) to the code repository, ensuring that machine-local state remains isolated from the shared repository history.

dstack-beads-core