Rebuilding the /plan Meta-Skill for Mythos-Class Models

The Philosophy of Intentional Planning

IndyDevDan argues that the current industry trend of 'vibe coding'—relying on agents to infer intent from sparse prompts—is a degradation of engineering talent. He posits that 'great planning is great engineering' and that the emergence of Mythos-class models (like Fable 5) provides a higher intelligence ceiling that can only be unlocked through rigorous, structured upfront planning. By treating the plan as a living, versioned artifact, the engineer can reduce the 'review' constraint, which is the primary bottleneck in agentic workflows.

The Trade-off Trifecta: Performance Over Speed

When building the /planF3 (Plans for Fable 5) meta-skill, the author explicitly rejects cost and speed optimization. The goal is maximum performance. This involves spending more tokens on rich context, embedded image generation (via GPT Image 2), and verbose metadata. The rationale is that by spending compute upfront to create a highly detailed, machine-readable plan, the agent is less likely to hallucinate or deviate, ultimately resulting in a more reliable and higher-quality output.

Designing for the 'Agent Trifecta'

The new plan template is designed for three distinct consumers: the human engineer, the engineering team, and the AI agents themselves. To satisfy these, the output is HTML-first, allowing for visual clarity and structured data. The template includes:

• Updatable Header Metadata: Tracking created/modified dates, commit hashes, agent IDs, and cross-references.
• Closed-Loop Validation: Embedded checklists and testing sections that block task completion until specific criteria are met.
• Interactive Q&A: A toggleable section that allows for human-in-the-loop intervention only when the complexity of the task demands it.
• Visual Context: Integration of focused images to provide spatial or architectural context that text alone cannot convey.

Implementation Strategy

The process begins with a 'raw' markdown brain dump to prime the context, followed by using a meta-skill to generate the template. The author emphasizes that the plan is a 'meta-skill'—a tool that generates other tools. By templating the engineering process, the developer effectively teaches the agent their specific engineering style, ensuring consistent results across different tasks and projects.