Claim
Your manager is an information problem
They are accountable for work they cannot see, to people who ask them hard questions on short notice. Anything you do that reduces the number of surprises they receive buys you credit, and credit is what gets spent on your behalf later.
This is not the same as being liked. Being liked is unreliable and slightly demeaning to chase. Being the person whose work never generates an unpleasant surprise is structural, and it survives your manager changing, which they will.
Action
Send one weekly update
Every Friday, six lines by email or message. Shipped this week, with a number if there is one. In progress, with the date you expect it. Blocked, naming what you need and from whom. One line of anything they might hear from someone else first.
That last line is the one that earns the trust, because it means they never get told something about your work by a peer at a meeting. Keep it to six lines. If it grows into a report nobody reads, it has stopped working. Write it Friday afternoon while it is fresh, and copy the shipped line into your evidence file while you are there.
Claim
Escalate early, with an option attached
Men hide problems hoping to solve them before anyone notices, and the arithmetic is bad: the cost of a problem grows with time, and the cost of reporting it stays flat.
The form matters. Do not arrive with a problem alone, which hands over the whole burden. Arrive with the problem, the two options you can see, which one you would choose, and what you need in order to do it. That is thirty seconds of their attention instead of an hour, and it is how you get treated as someone who could hold more.
Example
Ruairi ran a migration that slipped. On the Tuesday it became clear the date was gone, he sent four lines: the date is going to slip by about two weeks, here is why, we can either cut the reporting module and hit the date or take the two weeks and keep it, I would take the two weeks, I need you to tell the finance director today.
His manager later used almost that exact structure in front of her own boss. Ruairi got the two weeks. What he actually got was being the person she escalated with rather than about, which showed up in the scope he was handed the following year.
Caution
The courtier failure
There is a version of managing up that turns into managing appearances: agreeing in meetings and complaining afterwards, taking credit in the update, carefully never being the bearer of bad news.
It works briefly and is obvious to everyone except the person doing it. It also fails completely at the first reorganization, because your value was a relationship rather than a record. The distinction is simple to hold: reduce the surprises your manager receives, disagree with them directly and privately, and never let the update claim something the evidence file cannot support.
Claim
Ask what they are measured on
Once, in a one to one, ask what your manager is judged on this year and what would make their year go badly. Most people never ask, and most managers will tell you.
The answer reorders your priorities usefully, because it tells you which of your tasks are load bearing for someone above you and which are simply on your list. Aligning your visible work with those items is not politics. It is doing the part of your job that the organization will actually notice.
Ask once a year, not every quarter, because their targets change about that often and asking more reads as anxiety.
Aside
When the manager is the problem
All of this assumes a manager who is broadly competent and acting in good faith. Some are not. There are managers who take credit, withhold information deliberately, or punish escalation, and against those the techniques here do nothing except make you a more useful subordinate.
The diagnostic is behavior over time, not one bad week: do problems you raise get used against you, and do the people who leave the team say the same thing? If the answer is yes, stop optimizing the relationship and start on day 7. No amount of clean weekly updates fixes a bad manager, and staying two extra years to prove otherwise is expensive.