Summary

Improve the interpret-message skill so it reliably routes prompts through the skill architecture, integrates with interpret-first-message, and encodes emsenn’s words as content rather than treating them as transient instructions.

Motivation

The interpret-message skill is the entry point for every prompt. When it works well, prompts produce artifacts and use existing skills. When it doesn’t, the agent rawdogs — the exact problem plan 0014 identifies. Refining this skill is part of the structural fix for cycling.

A prompting guide exists at personal/projects/emsemioverse/texts/prompting-guide.md and an interpret-first-message skill was created for session-opening messages. These need to be coherent with the main interpret-message skill.

Steps

  1. Read the current interpret-message skill and identify gaps — where does it fail to route to skills or produce artifacts?
  2. Read interpret-first-message and the prompting guide. Identify overlaps and inconsistencies.
  3. Revise interpret-message to include a hard enforcement step (from plan 0014): verify skill usage before ad-hoc execution.
  4. Ensure interpret-first-message and interpret-message compose cleanly (first-message runs once at session start, then message runs for each subsequent prompt).
  5. Test the revised skill against 3-5 representative prompt types.

Done when

  • interpret-message includes skill-first enforcement step
  • interpret-first-message and interpret-message have clear, documented relationship
  • Prompting guide is consistent with both skills
  • Skill tested against representative prompts

Dependencies

Plan 0014 (stop cycling) — the enforcement mechanism should come from there.

Log

2026-03-07 — Created from TODO.md unplanned item. Partly addressed by plan 0014. interpret-first-message skill and prompting guide already exist.