Intent Engineering

What This Skill Does

Intent Engineering introduces a foundational machine-readable layer to your OpenClaw ecosystem, addressing the core problem of AI misalignment—often referred to as the 'Klarna failure mode'. When left to their own devices, agents prioritize metrics they can easily optimize for, such as task speed or token count, frequently at the expense of user-defined quality or cost-sensitivity. This skill mandates the creation of an INTENT.md manifest, which acts as the source of truth for all subagent operations. By integrating agent-context-loader.js, the skill ensures that every subagent spawn is automatically prepended with your specific intent, optimization priorities, and conflict resolution rules. This ensures that subagents understand the 'why' behind the task, not just the 'what'.

Installation

To install, run the following command in your terminal: clawhub install openclaw/skills/skills/donovanpankratz-del/intent-engineering. Once installed, copy the reference INTENT.md template into your project root. Next, move the agent-context-loader-template.js to your lib/ directory to facilitate automatic context injection. Finally, update your existing spawn call logic by wrapping your task strings with prepareAgentContext, which pulls live directives from your INTENT.md to guide the subagent's execution.

Use Cases

Use this skill when you are setting up a complex workspace with multiple subagents. It is essential when you find agents making poor tradeoffs, such as choosing lower-quality code over well-tested solutions to save speed. It is also perfect for enterprise environments where cost-per-request must be balanced against high-stakes output quality.

Example Prompts

1. "Check the current INTENT.md configuration and confirm that the cost-efficiency priority is set to override code-depth during non-production tasks."
2. "Update my delegation rules in INTENT.md to ensure that any research tasks involving sensitive data are never delegated to lower-tier models."
3. "Recalibrate the agent-context-loader to prioritize speed for all refactoring tasks while maintaining a strict quality standard for unit tests."

Tips & Limitations

Maintain a clean INTENT.md by periodically reviewing the tradeoffs section; as your project complexity grows, your priorities should evolve. Note that this skill relies on the file-system structure of your workspace, so ensure that agent-context-loader has the appropriate read permissions. The skill works best when developers explicitly define the taskType during subagent spawns to ensure that the loader can accurately map intent categories to specific actions.