The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local YAML specification files (swarm-specs/*.yaml) and repository contents (e.g., workflow files) to drive its logic.
Ingestion Points: Configuration is loaded from swarm-specs/*.yaml (Phase 1). Transformation and verification steps read from the target repository's files (Phase 3).
Boundary Markers: The skill lacks explicit instructions or markers to distinguish between trusted control logic and untrusted data contents within the processed repositories.
Capability Inventory: The skill has access to powerful tools including Bash (shell execution), Agent (spawning subagents), and Write (file system and PR creation).
Sanitization: There is no evidence of sanitization or filtering applied to the commands or data interpolated from external sources.
[COMMAND_EXECUTION]: The core functionality of the skill involves executing arbitrary shell commands provided in the YAML specification.
Evidence: The transform.command and verify.command fields in the spec are directly executed using the Bash tool during the migration process (Phase 3).
Context: While this is the intended purpose of the tool, it allows for arbitrary code execution if a malicious or compromised specification file is provided to the agent.
[DYNAMIC_EXECUTION]: The skill dynamically executes logic based on run-time configurations and subagent feedback.
Evidence: The skill uses the Agent tool to spawn git-operations-engineer or backend-system-architect subagents to perform repository transformations.
Logic: The coordinator-subagent DAG (Directed Acyclic Graph) is built dynamically from the upstream dependencies defined in the YAML file, which dictates the flow of execution and command dispatch.

swarm-migrate