Review and testing Context in practice
Turn an incident lesson into the next task’s test cases
A practical way to carry a failure sequence, invariant, and unknowns into focused regression checks.
THE SHIFU BLOG
Build on what
you already know.
Practical guides and lessons from building Shifu. Carry decisions, constraints, and reasoning into your next change.
Team delivery Building with context
The context handoff between planning, implementation, and review
Make decisions, exceptions, and verification state travel across the engineering workflow without turning every handoff into a document project.
Before coding Context in practice
What belongs in AGENTS.md—and what should be retrieved per task?
A practical sorting method for stable project guidance, scoped rules, and changing task evidence.
Behind Shifu Building with context
The retrieval optimization we decided not to ship
Why successful lookup and better precision did not settle one Shifu release decision.
Review and testing Context in practice
Give your coding agent a review brief it can use
A copyable review brief connecting the problem, invariants, evidence, failure cases, and unresolved decisions.
Beyond coding Building with context
How we use the context of building Shifu to write about Shifu
Follow an actual editorial workflow from product decisions and evaluation evidence to article choices and qualified claims.
Before coding Context in practice
Recover architecture decisions before your next AI-assisted change
A practical method for finding rationale, exceptions, and current applicability before turning history into a plan.
Behind Shifu Building with context
A good summary can lose the most important detail
What a context-retention experiment taught us about pending work, provisional choices, and the limits of aggregate scores.
Team delivery Context in practice
Measure the cost of rebuilding context before measuring AI productivity
A bounded worksheet for separating preparation, clarification, review, and rework around AI-assisted changes.
Review and testing Building with context
A code review needs the problem statement as well as the diff
Why intent, current behavior, and failure history change what an AI-assisted reviewer should inspect.
Before coding Context in practice
Claude Code memory and context: explain less between tasks
Choose what belongs in project instructions, native memory, the current session, and a task-specific evidence brief.
Behind Shifu Building with context
Why we are building a context engine for AI coding
The reasoning behind Shifu: preserve useful engineering understanding and bring it into the next task.