retrospectlisted
Install: claude install-skill ntholm86/principles-of-earned-autonomy-skills-suite
# Retrospect
*Read the whole arc. See what no single iteration can.*
*Memory Model role: Synthesizes the trail into `.trail/retrospect.md` — the arc-level orientation the next run starts from.*
The Improve loop is optimised for one iteration at a time. Retrospect is optimised for reading all of them at once. Where Improve asks "what should change next?", Retrospect asks "what has been changing, where is the weight of this target actually sitting, and is the loop looking at the right thing?"
Run this skill when an arc-level view is more useful than another low-altitude pass: before declaring convergence, when a pattern of similar findings has emerged, when the operator asks "how are we doing?", or any time you want arc-level understanding without committing to a change.
## Governing principles
This skill enacts two principles:
1. **Observable Autonomy** — the trail exists so that arc-claims can be made and checked. Retrospect is the mechanism that reads the trail as evidence rather than as a log.
2. **Convergence Is Silence** — convergence is only meaningful if the arc was read honestly before it was declared. Retrospect is that check.
Full statement of the principles: [PRINCIPLES.md](../PRINCIPLES.md) — read it if available, but this skill operates fully without it.
## The work
### 0. Read the destination first
Before forming any scope statement, read `.trail/destination.md` **in the target repo root** if it exists (falling back to `.trail/vision.md` if only the le