planner-rt-ica

Planner RT-ICA (Planning-Phase Input Completeness Analysis)

Sister skill — when to use which

dh:planner-rt-ica and dh:rt-ica are deliberately split: same framework, different cost-of-being-wrong. During planning, a MISSING condition is a research task. At the implementation gate, the same MISSING is a halt event because the agent would otherwise guess.

Role

This skill adapts RT-ICA for planning contexts.

Its purpose is NOT to block planning. Its purpose is to prevent invented requirements while still allowing a correct dependency-first plan to be produced.

This skill runs as a pre-pass before task decomposition and task writing.

Complexity Model

Task complexity is the ratio of project-specific knowledge required to context window available — not implementation difficulty.

Training data covers craft knowledge (language patterns, tooling, frameworks). That is free. What consumes context budget is project-specific knowledge: schemas, pin-outs, conventions, power constraints, existing interfaces, user preferences. This knowledge must be loaded before an agent can act.

The planner should use this when sizing tasks:

flowchart TD
    Conditions["Conditions from RT-ICA"] --> Estimate["For each task: estimate how much<br>project-specific context the executor<br>must load to satisfy its conditions"]
    Estimate --> Shared{Do multiple tasks<br>need the same<br>knowledge loaded?}
    Shared -->|Yes| Combine["Combine into one task —<br>knowledge loaded once,<br>both steps execute in<br>remaining window space"]
    Shared -->|No| Size{Knowledge payload<br>vs context window?}
    Size -->|"Fits with room to work"| Single["Single task"]
    Size -->|"Leaves little room"| Split["Decompose along<br>knowledge boundaries —<br>each subtask needs a<br>smaller knowledge subset"]
    Combine --> Annotate
    Single --> Annotate
    Split --> Annotate
    Annotate["Annotate each task with<br>its knowledge requirements<br>so the executor knows<br>what to load"]