The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided operator names, background documentation, and reference URLs in Phases 0 and 1 to guide the migration process. This creates an attack surface where instructions embedded in external documentation could influence agent behavior.
Ingestion points: Operator names, background context files, and documentation URLs (SKILL.md Phase 0/1).
Boundary markers: The workflow uses explicit 'Gates' requiring user confirmation before advancing to the next phase, which acts as a safety boundary.
Capability inventory: Employs Bash, Write, Read tools, and the ability to spawn subagents (SKILL.md Phase 5a).
Sanitization: The skill does not explicitly mention sanitization or validation of the input strings before they are interpolated into the subagent prompts.
[COMMAND_EXECUTION]: The skill executes various shell commands via the Bash tool for project management and testing tasks, including git log, grep, and running Python-based test scripts (atb_csv_ops_test.py). Pre- and Post-tool use hooks also use grep and echo to monitor the project state.
[DYNAMIC_EXECUTION]: Phase 5a utilizes a dynamic subagent delegation pattern for performance testing. This involves constructing a prompt for a subagent that includes shell script logic with interpolated variables like {op} and {ATB_REPO_PATH}.

atb-ops-to-aclnn-migration-workflow