The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on several bash scripts, primarily scope-actions.sh, to manage the task's alignment data. These scripts perform all file reads and writes using jq for structured data manipulation. Additional scripts like surfaces-actions.sh and task-actions.sh are invoked to handle project-wide configuration and task state transitions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from untrusted or user-controlled sources within a project's task folder.
Ingestion points: The skill reads task.md, alignment.v5.md (legacy contracts), and existing alignment.json files to draft or update the task scope.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing the content of these files, potentially allowing embedded instructions to influence the agent's drafting logic.
Capability inventory: The skill can execute local bash scripts, modify project JSON files, and dispatch a 'distiller' agent to review the contract.
Sanitization: The implementation uses jq with proper argument passing (--arg) for all JSON updates, effectively preventing command or JSON injection at the data storage layer.
[DYNAMIC_EXECUTION]: The scope-actions.sh script executes alignment-render.sh at runtime to generate updated Markdown documentation for the user. This follows a standard pattern using static local paths.

scope