Workflow on Vibe Coding

Doc Co-Authoring

Mon, 01 Jan 0001 00:00:00 +0000

Doc Co-Authoring packages a structured process for writing technical specs, PRDs, decision docs, and proposals: gathering context up front, iteratively building sections through brainstorming and editing, then stress-testing the draft by having a fresh, context-free Claude read it as a real reader would.

Get Shit Done

Mon, 01 Jan 0001 00:00:00 +0000

Get Shit Done is a lightweight meta-prompting and context-engineering system for Claude Code, built around spec-driven development rather than ad hoc prompting.

Why a skill for this

Spec-first workflows tend to produce more predictable output than freeform prompting, especially on larger features. Packaging that discipline as a skill makes it the default instead of something re-explained per task.

n8n Skills

Mon, 01 Jan 0001 00:00:00 +0000

n8n Skills lets Claude directly understand and operate n8n automation workflows — reading, editing, and reasoning about workflow graphs rather than treating them as opaque JSON.

Why a skill for this

n8n workflows are graphs of connected nodes that are awkward to reason about as raw exported JSON. A skill that understands the workflow model can edit and debug workflows the way a human building them in the n8n UI would.

Breaking Big Tasks Into Agent-Sized Steps

Mon, 09 Feb 2026 00:00:00 +0000

Vague, large prompts produce vague, large diffs — and vague diffs are the hardest kind to review. The fix isn’t a cleverer prompt; it’s decomposition.

Signs a task is too big

You can’t describe what “done” looks like in one sentence
The task touches more than two or three unrelated parts of the system
You’d need to check in multiple times even if a human were doing it

If any of these are true, break it down before prompting.

What Vibecoding Actually Means

Mon, 12 Jan 2026 00:00:00 +0000

The term “vibecoding” gets used two very different ways. One camp treats it as a punchline — typing vague prompts and accepting whatever an AI spits out, bugs and all. The other treats it as a real discipline: directing an AI collaborator with enough precision and judgment that the result is actually production-worthy.