The Agent Skills Directory

[SAFE]: The skill implements a structured 'plan before code' process. It includes a <HARD-GATE> directive that strictly forbids the agent from writing code or modifying files until the user has reviewed and approved a proposed approach, serving as a significant safeguard against unauthorized modifications.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied messages to understand task requirements and generate project proposals.
Ingestion points: User messages are ingested to determine the task 'what' (SKILL.md).
Boundary markers: The workflow relies on a manual user confirmation step (Step 5) as the primary control before any generated plan is executed.
Capability inventory: The skill can invoke internal context-discovery tools, perform task breakdowns, and execute code changes upon approval (SKILL.md).
Sanitization: No specific input sanitization or filtering logic is defined for the user-provided data.
[COMMAND_EXECUTION]: The workflow incorporates the use of specialized tools such as oac:context-discovery and oac:task-breakdown. These commands are used to find project standards and manage complex features within the context of the planning process, and their outputs are reviewed by the user as part of the proposed plan.

oac-approach