Agently Prompt Management

Use this skill when the core problem is how prompt state should be structured before one request or request family runs.

Native-First Rules

• prefer input(...), instruct(...), info(...), and output(...) over concatenated prompt strings
• move reusable prompt structure into prompt config or YAML instead of ad hoc literals
• keep runtime variables as ${...} placeholders in prompt files and inject them through mappings at load time
• keep task-specific request contracts in prompt config, and keep only widely reused persona setup in small code-side factories
• when the output contract is stable and shared across a request family, keep it in prompt config such as .request.output instead of rebuilding it ad hoc in Python
• keep prompt composition separate from transport and orchestration
• use config files as an editable bridge when UI or product teams need to adjust prompt-driven behavior without rewriting workflow code

Anti-Patterns

• do not flatten business context into one opaque string unless the task is trivial
• do not rebuild prompt templates through ad hoc .format(...) or string concatenation when prompt mappings already fit
• do not scatter stable prompt or output contracts across multiple Python helpers when one prompt config can own them
• do not use prompt config files as a substitute for workflow state

Read Next

• references/overview.md