Outcomes, not loose instructions
A task starts with an objective and its acceptance criteria, not with a list of commands. AIdrasil defines scope, identifies risks and proposes how to verify the result. If the objective is ambiguous and the ambiguity blocks progress, it asks before writing code.
Stage 1 · Connect
It connects to the authorized repository and creates an isolated working environment. Before changing anything it maps architecture, dependencies, conventions and project constraints, so changes respect what already exists.
Stage 2 · Build
It implements the change on its own branch, inside an environment separated from production. It respects the stack, the code style and the limits set by the organization. Secrets stay outside the repository and database queries are parameterized.
Stage 3 · Verify
It runs functional tests for the happy path and edge cases, quality analysis, dependency review and security checks over code and secrets. If a control fails the change does not advance: it goes back to build.
Stage 4 · Document
It records what was decided and why, which risks remain open, which tests ran and with what result. That documentation travels with the change and is what makes later audits possible without rebuilding context from memory.
Stage 5 · Deliver
It delivers a reviewable Pull Request with attached evidence and requests explicit human authorization before merging, deploying or working with production data. The decision on merit stays with the team.
Project memory and context
Each project keeps its decisions, pending items and evidence. That reduces context loss between sessions and prevents every task from starting by re-explaining how the system works.