Repository navigation
Conversation
A debug script that failed to compile never ran its wrapper, so the run waited out its whole timeout and the only clue was an app-level javascript-error event - which WebKit sanitizes in sandboxed frames to "Script error. (details withheld ...)". The bridge now checks whether the injected script started. If it did not, the error event raised during insertion becomes the run's result: the parser message plus line/column within the debug script on Chromium and Firefox, or an explanation that details were withheld on WebKit. The run finishes straight away instead of timing out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
Newer Datasette releases add "private" to responses for signed-in actors. Check for the no-store directive instead of the exact header so the test passes against older and newer Datasette alike. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
…rapping Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
WebKit reports every uncaught error from a classic script in a sandboxed (opaque origin) frame as a bare "Script error.", so app_debug runs in Safari told the agent nothing about the app's own errors. Probing WebKit in CI showed module scripts and same-origin external scripts keep their error details, and that a MutationObserver sees each parser-inserted script before it runs. On engines where a probe error comes back sanitized, the debug bridge now: - swaps each inline classic <script>, just before it runs, for a same-text blob: script - still a classic script run in document order, so globals, document.write and document.currentScript are unchanged. Its errors are reported against "app-script-N" (the app's Nth <script> element) with the failing line as sourceLine - wraps callbacks the app registers (timers, animation frames, microtasks, listeners, on* handler properties, inline on* attributes, observers) to capture what they throw before it is sanitized - names the script whose top-level code threw when neither applies The debug frame CSP additionally allows blob: scripts (granting nothing 'unsafe-inline' does not) and sets worker-src 'none' to keep workers blocked. The production app view policy is unchanged. Chromium and Firefox report full details already, so none of this activates there. Debug scripts now run after DOMContentLoaded, so they always see the page after its own parser-inserted scripts have run. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
WebKit drains microtasks between a callback invoked from the event loop (a timer, a real user click) throwing and reporting the exception, so a remembered error cleared by a microtask was gone before its sanitized "Script error." event arrived. Remember caught errors on a stack cleared by a later task instead - the report always lands within the same task, and a callback throwing during that microtask drain is reported first. The recovery test now reports every captured error in full on failure. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A debug script that failed to compile never ran its wrapper, so the run
waited out its whole timeout and the only clue was an app-level
javascript-error event - which WebKit sanitizes in sandboxed frames to
"Script error. (details withheld ...)".
The bridge now checks whether the injected script started. If it did
not, the error event raised during insertion becomes the run's result:
the parser message plus line/column within the debug script on
Chromium and Firefox, or an explanation that details were withheld on
WebKit. The run finishes straight away instead of timing out.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01VX4speFtsjrJCtLtVaFB85