Add reusable single-entry Obsidian upsert helper, Add "Reset AI Breaker" button, Fix AI Studio response parsing for nested JSON payloads, Add Polyceph compatability hooks

[Ryah]

I had my own personal fork to work on my companion extension, and genuinely forgot pull requests were a thing. So I decided to submit one. Sorry if it's a lot in one PR lol.

Summary

• Add reusable upsertConvertedEntry(...) helper in the import bridge for single-entry conversion/upsert workflows.
  • Future lorebook automation support, added example for MemoryBooks in my companion extension.
  • My example is pretty hardcoded for specifically MemoryBooks, so I didn't want to include it in this PR to avoid any future conflicts.
• Add Polyceph compatibility hooks for retrieval/publish integration:
  • deepLoreEnhanced_polycephRetrieve
  • deepLoreEnhanced_polycephPublish
• Improve AI response parsing to handle additional provider envelope shapes and nested response payloads.
• Fix AI Studio JSON API parsing fallback so nested JSON response envelopes resolve to usable model text/results.
• Fix extractAiResponseClient object-wrapper guard to only unwrap objects with known array-valued keys, preventing silent mis-parse on arbitrary object shapes.
• Add AI search trace logging ([DLE][TRACE] aiSearch_call_result / aiSearch_parse_failure_context) to surface raw result shape and parse failures without changing production behavior.
• Add a drawer footer Reset AI Breaker control for manual breaker recovery
• Improve AI breaker UX with countdown visibility and explicit manual reset control.
  • Breaker kept firing when I was testing the auto summary + keyword part of my MemoryBooks -> Obsidian pipeline failing (which led me to figure out the consistent JSON enveloping bug in AI Studio) so I added that in as a fallback to not have to refresh the page).

Why

• Enables cleaner extension-to-extension integration without duplicating import/write logic.
• Improves resilience and usability around AI service degradation/recovery.
• Increases compatibility with provider-specific response payload formats.

[pixelnull]

Hey! @Ryah

Don't apologize for doing a PR or the bundle. You said you forgot PRs were a thing while working on your fork; that's how most useful contributions actually happen, and the volume of useful stuff in this PR is real. Thanks for sending it. The Reset AI Breaker UX, the upsertConvertedEntry helper, the AI Studio response-parsing work, and the Polyceph integration are all things I want to land... BUUUT, you're right, just not in one bundle.

The reviewability blockers, concretely:

Bundled scope - Four genuinely independent changes (breaker UX, import helper, AI parser, Polyceph pipeline-replay). Reviewing them together means a problem in any one blocks the others, and rolling back later means surgical reverts across 14 files. I want to merge the breaker button this week and still be asking questions about Polyceph in two weeks without holding the breaker hostage.

Reformatter pollution. I diffed the base and head of each large file with whitespace stripped:

• settings-popup.html: 1163 add / 927 del at the line level - but identical content character-for-character once whitespace is removed. 100% reformatter pass, zero behavior change.
• style.css: 1827 add / 377 del at the line level - but only a few hundred bytes of real content change (the new .dle-breaker-reset-btn, .dle-footer-breaker-row, .dle-ai-breaker-countdown rules). The other ~99% is rule-body expansion / indentation.
• drawer.html: 151 add / 80 del - ~380 bytes (Reset AI Breaker button markup).

That's an enormous amount of noise for a small amount of real change, and it's hiding the real change. I'd guess your editor has format-on-save with different rules than this repo uses. Maybe a attribute-per-line HTML wrapping at ~120 cols, and one-property-per-line CSS expansion. If you can either turn that off for these files, or revert the formatter pass before opening the smaller PRs, the actual review goes from "scan 3000 lines for behavior changes" to "review 30 lines".

Polyceph surface is bigger than the PR body implies. retrieveLoreForPolyceph in index.js is basically a full parallel entry point that re-runs the pipeline (matchEntries → runPipeline → applyPinBlock → contextualGating → reinjectionCooldown → requiresExcludes → stripDedup → formatAndGroup) and publishes injection state to the drawer.

That's ~109 new lines plus two globalThis.deepLoreEnhanced_polyceph* surfaces. Adding a hook is a soft API commitment; this is a parallel pipeline entry point, which is a much bigger commitment because the drawer/sidebar state has to stay consistent across both call paths forever. I want to evaluate that in isolation, not bundled with three other things.

What I'd like

Four PRs against staging:

1. feat: Reset AI Breaker drawer button + cooldown countdown - src/state.js (resetAiCircuitBreaker, getAiCircuitCooldownRemainingMs), src/drawer/drawer-render-footer.js, src/drawer/drawer-events.js, the ~3 new selectors in style.css, the breaker button markup in drawer.html, the test/unit.mjs cases that cover it. Drop the formatter-only chunks from the CSS and HTML. Smallest PR, lowest risk, highest immediate utility. I'll merge this first.

2. feat: upsertConvertedEntry helper for single-entry Obsidian writes - src/vault/import.js only, plus tests. Clean addition, no behavior change to existing import flow.

3. fix: handle Google AI Studio nested response envelopes - src/ai/ai.js, src/helpers.js (cmrsResultToText + extractAiResponseClient changes), trace logging, tests. Looks good. One question I want to dig into before merge: extractInspectStyleEnvelopeText is regex-parsing util.inspect-formatted output (single-quoted strings concatenated with +). That's weird, it implies the AI Studio response is reaching cmrsResultToText already converted to inspect-string form rather than as an object. Could you point me at where in the call stack the inspect-stringification happens? I'd rather fix the upstream stringify than parse the symptom downstream, if we can. The Gemini candidates[].content.parts[].text extraction itself is clearly correct and welcome.

4. feat: Polyceph pipeline-replay surface (deepLoreEnhanced_polycephRetrieve / Publish) - the index.js additions and the two globals. I want to look at this one with fresh eyes because it's effectively a second public entry point to the retrieval pipeline, and the drawer/sidebar injection-state contract has to hold across both. Things I'll want to verify: does publishPolycephTrace racing with a real onGenerate cause the injection panel to show wrong data? does setLastInjectionEpoch(chatEpoch) after Polyceph desync the swipe-rollback tracking? Worth slowing down on alone.

Workflow suggestion:

git checkout staging
git pull
git checkout -b breaker-reset
git cherry-pick <commits relevant to breaker>
# remove formatter-only hunks from style.css / drawer.html / settings-popup.html
git push
gh pr create --base staging
# repeat for the other three branches

If you'd rather not split it yourself, fair. Say the word and I'll cherry-pick the breaker UX and upsertConvertedEntry directly off your branch into focused commits on my end, credit you in the commit messages, and close this PR with thanks. The AI Studio parser and Polyceph pieces I'd really rather you author standalone, since I want to ask the upstream-stringify question on one and the pipeline-state-contract questions on the other, and you're best positioned to answer.

Thanks again: the forgot-PRs-existed energy is exactly why I left issue #2 open. Rather get a 14-file dump and ask for a split than not see the contribution.

🖤

[Ryah]

Sorry, I just saw this. Honestly my plate has gotten very full now that I've been brought onto the Freaky Frankenstein team and with personal things in life, so I don't have much time to pick the commits apart. Instead, if you would like, you are free to do whatever with my fork. I probably won't be updating it for a while, so absolutely feel free to take whatever you want from it! I still use DeepLore all the time so I'm happy to contribute in any way I can, I just won't have the mental energy to look at a line of code for a while lol.

There's technically 2 branches. The main one (this PR) and another branch called "fork" where I threw more things into. Feel free to search through the commits and cherrypick whatever you would like added.