Task And Spirit, Stated Separately.
What you're building — and the outcome you're actually reaching for — in two sentences. When they point different directions, the gap surfaces before it gets expensive.
Inside The IncredAgents Mod Pack
Agents default to writing code immediately — pattern-matched from memory, never looking at real inputs, patching bugs forever. This skill replaces that reflex with a gate: six preparation steps, completed and approved, before a single line gets written.
The gate
Until all six steps complete and you approve the plan: no code, no scaffolding, no installs, no architecture decisions. Searching, querying APIs by hand, inspecting real data — allowed.
What you're building — and the outcome you're actually reaching for — in two sentences. When they point different directions, the gap surfaces before it gets expensive.
A dedicated research pass on current practitioner consensus — official docs read directly, known failure modes named, anything unverifiable flagged. Generalities get sent back.
The real data file. The real API response. The actual page. Edge cases found in the data, not imagined — three null fields discovered beats thirty bugs patched.
The discipline
Before the plan, the agent names at least eight failure modes it expects in production. Can't name eight? The research wasn't thorough enough — go back.
Rate limits, maintenance pages returning HTML, oversized records — named while they're cheap.
Presented, questioned, approved. After approval: action, not re-confirmation.
Surface a dependency bug during prep and it gets named in the plan and executed — not parked for a 'version two' that never comes.
Get started
Ships in the mod pack. It activates on any word that means building.
$ git clone https://github.com/MyAiFreedomSystems/incredagents.git
$ cd incredagents
$ python3 onboard.py
One installer wires in the whole mod pack and asks only for the keys each skill needs.
The next build starts with six steps — or with bugs.
Get Pre-Build-SOP On GitHub →