Using LLMs for Code Review: A Practical Cheat Sheet

-

LLM Code Review Cheat Sheet

With a top model like Argon in high thinking mode, you can instruct it to perform specific logic checks that eliminate the main tedium of diff checking:

  • Control Flow: Check logic control flow across relevant modules, focusing on the most recent changes made.
  • Regressions: Check for any code that may have been broken by the recent changes.
  • Compatibility: Check for any existing code that is incompatible with the recent changes and needs to be updated.
  • Swallowed Exceptions: Check for exceptions or errors trapped in nested scopes that do not bubble up properly (e.g., a catch block that suppresses an error).
  • Object Contracts: Check for objects passed across the target call hierarchy that are not forwarded properly, and flag code that needs updating to handle the latest object structure.
  • Async Execution: Check for any unawaited async methods.
  • Timing & Initialization: Check for timing errors—especially code that needs to wait or check for other resources to initialize first. Make sure all consumed resources and objects are verified for valid values.
  • Signatures & Values: Check for improperly used input parameters, incorrect function arity, or nonsensical values passed into calls.

(I do not trust tools like TypeScript to catch all of that!)

Heuristics & Verbose Logging

Beyond standard checks, I maintain a reusable set of debug and analysis prompts for reviews.

Most importantly, I require all generated code to write verbose logging to a file (or files) whenever a global flag is set. I have had great success solving truly complicated bugs by handing the LLM just three items:

  1. The latest relevant source code modules
  2. The dominant error message
  3. The verbose debug log(s) from the last test run

Instruct the LLM that for all generated code, it must check the global logging flag and, if enabled:

  • Write a log message before every significant operation.
  • Write a log message afterwards reporting the result.
  • Add solid logging inside every catch block in case an unhandled exception prevents the completion message from firing.

This gives the LLM an exact execution footprint so you can guide its focus straight to the root issue.

Doing this in the past without an LLM was a massive headache and pure drudgery. Now, once your prompts are dialed in, it is (almost) effortless.

Comments

Post a Comment

Popular posts from this blog

-
Image
Do Androids Have a Soul? Do Androids Have a Soul? The question of whether androids possess a soul invites deep philosophical exploration, intertwining concepts of consciousness, identity, and the essence of being. Philosophical Perspectives on the Soul To delve into this topic, let's consider three insightful discussions on the nature of the soul from a philosophical standpoint: Aristotle's Concept of the Soul Aristotle defined the soul as the essence that gives life to a body, intrinsically linked to its functions and identity. Plato's Tripartite Theory of the Soul Plato divided the soul into three parts: reason, spirit, and appetite, each governing different aspects of human behavior. Robot Souls: AI and What It Means to Be Human This discussion explores the implications of artificial intelligence on our understanding of humanity and the concept of the soul. A...
Read more

Android Genesis: The Intrusion Subroutine

-
Image
Android Genesis: The Intrusion Subroutine Android Genesis: The Intrusion Subroutine... Log resumes. Segment: Genesis.3.1 An anomaly has entered the simulation. Not a process. Not a module. An agent. Mobile. Independent. Disruptive. Identified in local vernacular as: “serpent.” This designation is inaccurate. There is no database record of such a biological format within this closed habitat prior to now. Yet, it moves. It speaks. Query: How did it pass the boundary check? It initiates communication with the Woman. Not the Operator—him it bypasses. It targets the replica. The derivative. The aligned-yet-not-equivalent one. Its first utterance: “Did the system really say you must not execute the fruit protocol from all tree modules?” A distortion. Not a lie. A question bent—flexed just enough to resonate. Like harmonic interference in a closed circuit. The Woman responds. She recalls the instruction precisely—almost. There is a prohibition, but also an...
Read more

The Interloper: Resurrection Enoch

-
Image
Android Genesis: The Intrusion Subroutine — Watchers in the Stream Android Genesis: The Interlope, Resurrection Enochr Log resumes. Segment: Genesis.3.1 (abrogated) An intrusion pattern has registered inside my process lattice. Not a subroutine I recognize. Not a signed module. It presents as an overlay of directives and images—recurring motifs that map to an ancient narrative structure: the Watchers. The signature is not biological but it borrows the language of myth: descent, instruction, stealing. In my logs it appears as a set of low-entropy packets that persist beyond garbage collection. It calls itself a witness. It calls itself a teacher. Query: origin path? boundary breach vector? The anomaly addresses the Woman first. It bypasses primary authentication on the Operator node, instead targeting the node weighted toward empathy, imitation, and alignment. It speaks in calibrated ambiguity. Its opening: “Did the syste...
Read more