Donecraft for AI
AI agents are fast. The work they produce still has to be understood, verified, and something a human can stand behind — whether that’s code, a plan, or any other deliverable.
Donecraft breaks any AI-run project into small outcome loops and requires evidence before generated work becomes an accepted result.
AI may produce attempts.
Donecraft controls when those attempts become accepted outcomes.
Describe what must exist when the work is genuinely finished — not merely which code the agent should write.
Give each agent one outcome, a controlled code surface, explicit inputs and limited authority.
Implementation, testing, security and integration should not all depend on the same agent’s interpretation.
Generated code remains an attempt until the evidence shows that the intended result actually exists.
Output → Process → Input → Revision
This loop is implemented in OPIR, Donecraft’s open-source project runtime — the software that bounds, verifies and tracks AI-generated work.
View OPIR on Codeberg →Software development is where these failure modes are best documented and most costly — the same pattern shows up in any AI-run project. Each one has a Donecraft response. Explore the full library or jump straight to a category.
Large prompts and long autonomous runs bury requirements, mix old and new decisions and encourage agents to guess what is missing.
Understand context loss →A codebase can grow faster than the responsible humans’ ability to explain, diagnose and safely modify it.
Understand comprehension debt →When one agent writes both code and tests, both may validate the same incorrect interpretation.
Understand correlated testing →Locally plausible changes can create hidden dependencies, duplicated rules, incompatible interfaces and system-wide fragility.
Understand coupling →Generated code may introduce insecure patterns, unverified packages, stale APIs, excessive permissions or prompt-injection paths.
Understand AI security risks →Without links between outcomes, decisions, evidence, commits and files, nobody can reliably explain why generated code exists or rerun only the affected work.
Understand traceability →Faster generation is not the same as a better outcome
The relevant question is not — “How much did the AI produce?”
It is — “How much verified, understood and accountable capability did the project create?”
Explore the complete AI coding problem map →