Notes a lazy engineer keeps to cut down on trial and error — AI agents, Obsidian plugins, and dev tooling.https://decipher99.com/https://decipher99.com/en/posts/agent-infra/ai-agent-fixed-blog-draft-from-memory-summary/https://decipher99.com/en/posts/agent-infra/ai-agent-fixed-blog-draft-from-memory-summary/I had an AI agent fact-check a blog draft, and it used its summary memory to change a correct anecdote into a wrong one. Raw Honcho conversation logs caught it.Wed, 28 Oct 2026 22:30:00 GMThttps://decipher99.com/en/posts/tool-review/decisions-api-jev-broken-sentence-judge-four-times-false-positives/https://decipher99.com/en/posts/tool-review/decisions-api-jev-broken-sentence-judge-four-times-false-positives/I had OpenAI Decisions API and Jev judge LLM-edited notes for broken sentences. They flagged ~4x more normal notes than gpt-6-luna; split questions backfired.Tue, 27 Oct 2026 23:05:02 GMThttps://decipher99.com/en/posts/obsidian-plugin/llm-note-dedup-left-sentences-pointing-at-nothing/https://decipher99.com/en/posts/obsidian-plugin/llm-note-dedup-left-sentences-pointing-at-nothing/Removing note duplicates with an LLM left 'this technology stack' and 'also' pointing at nothing. I stopped adding code rules and had an LLM judge each edit.Mon, 26 Oct 2026 20:40:44 GMThttps://decipher99.com/en/posts/obsidian-plugin/llm-note-merge-test-passed-but-fixed-code-never-ran/https://decipher99.com/en/posts/obsidian-plugin/llm-note-merge-test-passed-but-fixed-code-never-ran/I fixed an LLM note-merge bug in my Obsidian plugin and reran it on real notes. It passed, but the fixed code never ran until I restored the pre-merge state.Mon, 26 Oct 2026 14:22:13 GMThttps://decipher99.com/en/posts/obsidian-plugin/deleted-the-wrong-example-but-the-merge-rule-stayed/https://decipher99.com/en/posts/obsidian-plugin/deleted-the-wrong-example-but-the-merge-rule-stayed/My Obsidian plugin's LLM rule for merging duplicate notes blocked even same-product notes. I traced it, tested real merges, and adopted Karpathy's LLM Wiki.Sun, 25 Oct 2026 00:53:49 GMThttps://decipher99.com/en/posts/tool-review/self-hosted-git-ci-gitlab-gitea-onedev-github/https://decipher99.com/en/posts/tool-review/self-hosted-git-ci-gitlab-gitea-onedev-github/With code that can't leave the company, I compared self-hosted GitLab, Gitea, OneDev and GitHub Enterprise Server on specs, SSO, approvals and cost.Sat, 24 Oct 2026 03:08:33 GMThttps://decipher99.com/en/posts/obsidian-plugin/llm-summary-dropped-once-survived-twice/https://decipher99.com/en/posts/obsidian-plugin/llm-summary-dropped-once-survived-twice/Chasing an LLM summary that dropped a source's causal claim, I ruled out grammar and counterintuitiveness with controls. A claim stated once vanished.Fri, 23 Oct 2026 10:06:40 GMThttps://decipher99.com/en/posts/obsidian-plugin/missing-rule-in-llm-merge-prompt-made-no-difference/https://decipher99.com/en/posts/obsidian-plugin/missing-rule-in-llm-merge-prompt-made-no-difference/Chasing an LLM note merge that dropped one source's content, I found the merge prompt lacked an information-density rule. Adding it changed nothing.Thu, 22 Oct 2026 14:00:00 GMThttps://decipher99.com/en/posts/obsidian-plugin/designing-an-over-split-guard-that-already-existed/https://decipher99.com/en/posts/obsidian-plugin/designing-an-over-split-guard-that-already-existed/An LLM proposed splitting one note into five pieces of 20–30 words each. Designing a guard against over-splitting, I found one built the day before.Wed, 21 Oct 2026 11:30:00 GMThttps://decipher99.com/en/posts/obsidian-plugin/language-check-code-did-nothing/https://decipher99.com/en/posts/obsidian-plugin/language-check-code-did-nothing/I built code to detect and fix Korean sentences leaking into LLM-written English notes. It fixed zero. The real cause was a prompt missing a language rule.Tue, 20 Oct 2026 12:03:02 GMThttps://decipher99.com/en/posts/obsidian-plugin/testing-obsidian-plugin-by-remote-controlling-another-pc/https://decipher99.com/en/posts/obsidian-plugin/testing-obsidian-plugin-by-remote-controlling-another-pc/To test an Obsidian plugin in the real app before each release, I drove Obsidian on another PC over the Chrome DevTools Protocol (CDP), deploy to button clicks.Mon, 19 Oct 2026 14:33:31 GMThttps://decipher99.com/en/posts/tool-review/decisions-api-three-questions-matched-luna/https://decipher99.com/en/posts/tool-review/decisions-api-three-questions-matched-luna/I tuned OpenAI Decisions API and Jev prompts per their docs. Split into three true/false questions, Decisions API got 36 of 37 pairs right; trimming input hurt.Sun, 18 Oct 2026 01:15:13 GMThttps://decipher99.com/en/posts/tool-review/openai-decisions-api-fast-but-wrong-fit/https://decipher99.com/en/posts/tool-review/openai-decisions-api-fast-but-wrong-fit/I compared OpenAI's Decisions API, Jev, and gpt-6-luna on merging duplicate notes. Decisions API was fastest, but confidently wrong on my multi-part rule.Sat, 17 Oct 2026 04:06:20 GMThttps://decipher99.com/en/posts/obsidian-plugin/laptop-2-openai-reasoning-effort-was-medium/https://decipher99.com/en/posts/obsidian-plugin/laptop-2-openai-reasoning-effort-was-medium/The work laptop's OpenAI reasoning effort was the default Medium, confirming chunk size alone caused the LLM summary errors. The two are separate knobs.Wed, 14 Oct 2026 22:17:43 GMThttps://decipher99.com/en/posts/obsidian-plugin/raising-llm-request-concurrency-surfaced-two-other-bugs/https://decipher99.com/en/posts/obsidian-plugin/raising-llm-request-concurrency-surfaced-two-other-bugs/Raising parallel LLM requests from 2 to 4 kept summary accuracy. The audit found missed Korean-English note merges and Korean headings on English notes instead.Wed, 14 Oct 2026 12:07:50 GMThttps://decipher99.com/en/posts/obsidian-plugin/why-korean-and-english-notes-would-not-merge-embedding-similarity/https://decipher99.com/en/posts/obsidian-plugin/why-korean-and-english-notes-would-not-merge-embedding-similarity/Korean and English notes on one concept never merged. OpenAI embeddings put them at 0.77, below same-language duplicates; a 0.70 similarity threshold fixed it.Tue, 13 Oct 2026 11:00:15 GMThttps://decipher99.com/en/posts/obsidian-plugin/openai-reasoning-effort-was-not-the-fidelity-culprit/https://decipher99.com/en/posts/obsidian-plugin/openai-reasoning-effort-was-not-the-fidelity-culprit/Lowering OpenAI reasoning effort from High to Medium didn't add LLM summary errors. Note count and run time changed a lot instead; chunk size governed accuracy.Mon, 12 Oct 2026 12:11:32 GMThttps://decipher99.com/en/posts/obsidian-plugin/git-conflict-markers-got-committed/https://decipher99.com/en/posts/obsidian-plugin/git-conflict-markers-got-committed/Continuing a chunk-size experiment at home, alternating two laptops on one git repo committed conflict markers into 77 files. Then I suspected reasoning effort.Sun, 11 Oct 2026 00:03:16 GMThttps://decipher99.com/en/posts/obsidian-plugin/lowering-llm-chunk-size-fixed-fidelity/https://decipher99.com/en/posts/obsidian-plugin/lowering-llm-chunk-size-fixed-fidelity/I compared the same 30 sources changing only the chunk size, how many go to the LLM at once. Summary errors that drop people and quotes fell from 40% to 0%.Sat, 10 Oct 2026 10:17:54 GMThttps://decipher99.com/en/posts/obsidian-plugin/189-llm-notes-became-63/https://decipher99.com/en/posts/obsidian-plugin/189-llm-notes-became-63/I had an LLM turn the same 30 sources into Obsidian notes, and just switching machines cut them from 189 to 63. Not one cause, chunk size, but two settings.Fri, 09 Oct 2026 05:43:17 GMThttps://decipher99.com/en/posts/obsidian-plugin/fixed-the-llm-prompt-twice-should-not-have-split-the-note/https://decipher99.com/en/posts/obsidian-plugin/fixed-the-llm-prompt-twice-should-not-have-split-the-note/An LLM-made sibling note kept coming out empty but for links. Two prompt fixes later, the source showed both courses shared one description: merge, don't split.Thu, 08 Oct 2026 13:54:36 GMThttps://decipher99.com/en/posts/tool-review/opencode-only-one-of-three-subscriptions-works/https://decipher99.com/en/posts/tool-review/opencode-only-one-of-three-subscriptions-works/Before installing the open-source coding agent OpenCode, I checked my subscriptions: ChatGPT works as-is, Claude is barred by policy, Gemini access is gone.Tue, 06 Oct 2026 22:54:29 GMThttps://decipher99.com/en/posts/obsidian-plugin/removing-the-embedding-filter-caught-more-duplicates/https://decipher99.com/en/posts/obsidian-plugin/removing-the-embedding-filter-caught-more-duplicates/I swapped an embedding filter plus LLM checks for one LLM review of all sibling notes to catch duplication. One hub-note rule took precision from 55% to 100%.Tue, 06 Oct 2026 14:17:00 GMThttps://decipher99.com/en/posts/tool-review/acp-two-protocols-same-acronym/https://decipher99.com/en/posts/tool-review/acp-two-protocols-same-acronym/The ACP seen next to MCP is the Agent Client Protocol, linking editors and AI coding agents. IBM's same-acronym protocol merged into A2A; ACP v2 is in draft.Sun, 04 Oct 2026 22:14:23 GMThttps://decipher99.com/en/posts/obsidian-plugin/switching-embedding-models-flipped-the-ranking/https://decipher99.com/en/posts/obsidian-plugin/switching-embedding-models-flipped-the-ranking/Moving from OpenAI's text-embedding-3-small to large flipped the order of true duplicate paragraphs and look-alike sentences mixed at cosine 0.72-0.80.Sun, 04 Oct 2026 01:02:18 GMThttps://decipher99.com/en/posts/obsidian-plugin/sibling-notes-copying-the-same-paragraph/https://decipher99.com/en/posts/obsidian-plugin/sibling-notes-copying-the-same-paragraph/Testing my Obsidian plugin on real articles exposed an LLM bug: sibling notes copied the same paragraph verbatim, and embedding duplicate detection missed it.Fri, 02 Oct 2026 11:04:34 GMThttps://decipher99.com/en/posts/obsidian-plugin/verified-on-a-different-pair-never-rechecked-the-original/https://decipher99.com/en/posts/obsidian-plugin/verified-on-a-different-pair-never-rechecked-the-original/I fixed a false positive in my Obsidian plugin's LLM merge-judgment prompt and even closed the GitHub issues, but never retested the case that caused the false positive in the first place. Rerun days later under the real full-vault conditions, the judgment flipped, and actually performing the merge suggested the original call was likely wrong all along.Thu, 01 Oct 2026 09:20:03 GMThttps://decipher99.com/en/posts/obsidian-plugin/subagent-missed-2-of-4-list-heavy-files/https://decipher99.com/en/posts/obsidian-plugin/subagent-missed-2-of-4-list-heavy-files/A full audit I delegated to Claude Code subagents reported 2 of 56 files as CLEAN when they weren't. An A/B test on the prompt showed that on files dense with bullet lists, 'I read it all' meant a skim rather than an item-by-item check, and a prompt that forces a checklist fixed it.Wed, 30 Sep 2026 09:14:38 GMThttps://decipher99.com/en/posts/tool-review/wrong-but-90-percent-confident/https://decipher99.com/en/posts/tool-review/wrong-but-90-percent-confident/I tried Laya, the open-source reproduction of TypeSafe's Jev, on my Obsidian plugin's duplicate-note decisions (Jev itself only through its docs, since my account isn't open yet). I measured CPU speed, 78% accuracy, and what fine-tuning would take, and the real problem turned out to be broken calibration: confidence around 0.9 on wrong answers.Tue, 29 Sep 2026 11:40:17 GMT