# Show HN: A simple decision layer for coding agents

> Source: <https://smarchone.github.io/signals/>
> Published: 2026-11-22 19:14:59+00:00

**Key takeaway:** Signals is a Claude Code skill, installed with `npx skills add smarchone/signals -g`, that persists approved architectural decisions and their rejected alternatives across sessions so coding agents stop re-arguing settled choices.

# [~/ smarchone/signals](https://github.com/smarchone/signals)

tl;drA Claude Code skill. It captures the decisions you make as you talk and build, across sessions, and asks you only about new ones that are hard to undo or felt by users.

What are you steering?

[the thesis →](blog.html)
- **Sessions** → none of them knows what the others decided.
- **Prompts** → prompt hell. You re-steer the same calls, prompt after prompt.
- **Specs** → spec-driven dev. The spec is stale by build two. Back to prompting.
- **Decisions** → signals. Decided once, it steers every build after.

Big calls only

## Decisions, not diffs.

Only the ones hard to undo, or felt by users.

Decided stays decided

## Each keeps its why.

And what was ruled out, so agents stop re‑arguing it.

Exec level

## You stay at the top.

Never variable names. Single player, for now.

you talk

*→*agent builds*→*you approve new decisions*→*remembered
Get started · **~1 min** · needs Claude Code

``` bash
$ npx skills add smarchone/signals -g   # then, in Claude Code:
> /signals why sqlite?                  # answers from your decisions
> /signals build the first slice        # builds what you discussed
> /signals exec drop sqlite for pg      # the panel argues, you decide
```

made with ♥ · [@smarchone/signals](https://github.com/smarchone/signals)
