How do you stop stale repo instructions from pulling a coding agent away from the current task?
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
This covers a coding tool or code-capability update — useful for developers assessing workflow changes and reusable value.
Stale documentation can quietly become a second task description for a coding agent. The agent may have the current request in front of it, but an old README, instruction file, or saved explanation can still pull the result toward a goal that no longer applies.
The failure is subtle because nothing has to crash. The output can look coherent while solving the wrong version of the task.
A reasonable safeguard might be to make repository instructions prove that they are still current. If a document cannot point to a file, command, or test that supports it now, the agent should treat it as context rather than authority.
How are you handling this in practice? Generated docs, explicit source-of-truth files, freshness checks, or a rule that current code and tests override prose?