Mashang Huasheng
A Bilibili technology knowledge content creator specializing in AI tool tutorials. The channel focuses on systematic instructional content about AI programming tools such as Claude Code, delivered in a whiteboard-style presentation format.
Timeline (last 90 days)
In the next episode, the instructor will bring the acceptance evidence into the PR and continuous integration stage to check the current commit and the required check items.
In the course, the candidate fix passed 12 assertions on the instructor's local machine, covering the original bug, negative numbers, 0, duplicate values, and the empty array, with each case verifying the three items of order, unchanged original input, and a returned new reference.
The numeric sorting bug deliberately written in this lesson is not a Grok Bot product issue, but a teaching case.
Whether review suggestions apply must also be judged; one must not blindly change everything, but rather explain the reason for adopting them against the requirements, impact, and scope.
CI failures can be divided into three types: results not meeting requirements, code behavior not matching the use case commands, and environment failures (dependencies inaccessible or tools reporting missing permissions), with different types corresponding to different handling approaches.
Deliverability judgment framework: when the current version B has a core regression failure and the duplicate value has been removed, only fix the relevant functions and tests, keep the finite-number premise, return a new commit and corresponding logs, and no longer reference A's passing screenshot; a maximum of two rounds of correction, and if it remains the same or goes out of bounds, escalate to manual handling.
Disputed or out-of-scope changes should be decided by the responsible person, and the executor should not be allowed to expand the task boundary on its own.
An all-green CI status is only a necessary condition for delivery, not a sufficient one; what really determines whether something can enter the review state is whether the evidence chain is complete, whether versions correspond, and whether the checks cover the original requirements.
Skipping a test is not necessarily wrong in itself, but the reason for skipping and the unverified items must be explained, and color must not be allowed to substitute for test content.
In lesson 16 of the Grok Bot series, Bilibili creator "马上花生" used a mock requirements progress table and mock emails to recreate a typical weekly report data conflict scenario: the progress table says 8 items are complete, while the email says 2 items are pending acceptance.
1 more timeline events
All Facts (11)
The numeric sorting bug deliberately written in this lesson is not a Grok Bot product issue, but a teaching case.
50%UnverifiedIn the course, the candidate fix passed 12 assertions on the instructor's local machine, covering the original bug, negative numbers, 0, duplicate values, and the empty array, with each case verifying the three items of order, unchanged original input, and a returned new reference.
50%UnverifiedAn all-green CI status is only a necessary condition for delivery, not a sufficient one; what really determines whether something can enter the review state is whether the evidence chain is complete, whether versions correspond, and whether the checks cover the original requirements.
50%UnverifiedSkipping a test is not necessarily wrong in itself, but the reason for skipping and the unverified items must be explained, and color must not be allowed to substitute for test content.
50%UnverifiedWhether review suggestions apply must also be judged; one must not blindly change everything, but rather explain the reason for adopting them against the requirements, impact, and scope.
50%UnverifiedDisputed or out-of-scope changes should be decided by the responsible person, and the executor should not be allowed to expand the task boundary on its own.
50%UnverifiedCI failures can be divided into three types: results not meeting requirements, code behavior not matching the use case commands, and environment failures (dependencies inaccessible or tools reporting missing permissions), with different types corresponding to different handling approaches.
50%UnverifiedDeliverability judgment framework: when the current version B has a core regression failure and the duplicate value has been removed, only fix the relevant functions and tests, keep the finite-number premise, return a new commit and corresponding logs, and no longer reference A's passing screenshot; a maximum of two rounds of correction, and if it remains the same or goes out of bounds, escalate to manual handling.
50%UnverifiedWhen a long-running Bot behaves abnormally, the first troubleshooting step is not to restart or reset, but to confirm the current actual state item by item.
50%UnverifiedIn the next episode, the instructor will bring the acceptance evidence into the PR and continuous integration stage to check the current commit and the required check items.
50%UnverifiedIn lesson 16 of the Grok Bot series, Bilibili creator "马上花生" used a mock requirements progress table and mock emails to recreate a typical weekly report data conflict scenario: the progress table says 8 items are complete, while the email says 2 items are pending acceptance.
50%