The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, specifically Jira issue IDs and URLs, to generate task cards that drive agent behavior. This creates a surface for indirect instructions to influence the execution of child engines.
Ingestion points: External Jira issue IDs, URLs, and user-provided requirements are parsed in Stage 1 to generate backlog cards (SKILL.md).
Boundary markers: The skill implements "frozen decisions," "Decisions-I-made-for-you" blocks, and mandatory human approval events that separate the intake phase from the execution phase (SKILL.md).
Capability inventory: The orchestrator writes to the local filesystem (.goal-driven/), manages git branches, and invokes high-privilege child engines such as jira-fix-workflow and goal-driven-workflow (SKILL.md).
Sanitization: The skill relies on a "Deep Intake Interview" involving "Presence Tiers" to resolve ambiguity and validate task goals before they are recorded in persistent storage (SKILL.md).
[DYNAMIC_EXECUTION]: The skill dynamically generates execution contexts by creating task cards at runtime. It also performs dynamic dispatching, where child engines are invoked based on the Engine field value parsed from these generated cards. While this is the primary function of the orchestrator, it represents a pattern of executing code/skills based on dynamically computed paths (SKILL.md).

goal-driven-queue