Remove the move-killers first
An Agent J speed run is a chain of clean exits: each scene clears when every active threat is gone and you reach the marked safe position. The input changes per scene, so the runner's job is to name the immediate threat, act once, and let the scene update before the next decision.
Clear one target, then read the new scene
Clear the scene, don't farm the targets
Winning is the clean exit, not the kill count. Leaving a scene with a threat still able to interrupt you is a fail, so the fastest run is the shortest safe chain to the marked safe position rather than the busiest shooting. Extra actions are extra time.
The opening read decides the run
On the first pass, name the single most immediate threat and any object that changes the environment, then act on the one that makes the next step safest. The enemy nearest to you is not always the one that will interrupt first, and the wrong first pick costs the whole scene.
One completed action before the next cue
Let one action fully finish, check how the scene shifted, then take the next. Committing before the update lands usually means treating two threats as one, which is the most common way a clean run becomes a reset, so the pause after each clear is part of the speed.
The prompt swap saves a retry
Each scene shows its input (mouse aim, a drawing gesture, or touch) when it opens. Checking that cue at load costs a second, while acting with the previous scene's gesture costs a full attempt, so the fastest habit is to let the prompt name the control before you commit.
Recover by splitting, not stacking
After a failed scene, list the threats again and deal with the one that can interrupt your next move, one at a time. Most failures come from combining two threats into one response, so splitting them and acting in order restores the chain faster than a big redo.
Exit at the safe position, never mid-threat
The safe position is both the win state and the cheapest pause: progress is banked the moment the scene closes. Leaving mid-scene leaves a threat active and the next read stale, so always clear the room before you leave and the next visit starts at a clean boundary.
Name the threat, then name the exit
Two things decide a scene: what can stop you on the way out, and where the marked safe position sits. Name both before you act, and the order of your actions becomes obvious, because anything standing between you and the exit has to be handled first. Acting without the second half of that read is how a run ends in a busy scene that looked handled.
Each scene is a new control, so check the prompt
The gesture that cleared the last scene may not be the one this scene wants, and the prompt shown at the start names the right input. Reading it takes a moment, while using the previous scene's gesture costs a whole attempt. Making that check a fixed habit at the top of every scene is one of the cheapest ways to keep a chain alive.
Why rushing the scene backfires
How do I pick the first action in a scene?
Find the threat that can interrupt your next move and any object that changes the environment, then act on what makes the next step safest. The closest enemy is not always the right first target. The interrupt is.
What is the fastest recovery from a failed scene?
Relist the threats and take the one that can interrupt your next move, one at a time. The failure almost always came from stacking two threats into one response, so splitting them is the shortest road back to a clean chain.
Why does the input keep swapping between scenes?
Every scene is built around a different interaction, so the prompt shows a mouse aim, a drawing gesture, or a touch cue as needed. Matching the displayed input is part of the run, and the check at load is faster than paying for a wrong gesture with an attempt.
When should I restart a run?
Keep going as long as you know the next threat. A scene only truly costs time when you repeat the same wrong response twice; a clean re-read and one action at a time recovers almost every run, so restarting early rarely pays.
