@m_emanaftab @trq212 Rough. When compaction errors out mid-session, rules and context can quietly disappear together.
English
Rippletide
496 posts
Rippletide
@RippletideCo
The LLM proposes. Rippletide decides. Deterministic decision enforcement for AI agents. Full audit trails. Explainable decisions. npx rippletide-code
Paris | San Francisco เข้าร่วม Mart 2024
16 กำลังติดตาม42 ผู้ติดตาม
@trq212 /compact in Claude Code not working today.
Keeps throwing image dimension errors and the compaction isn't actually running. Tried 3 times in a row. Same error every time. Had to submit /feedback and restart.
Anyone else seeing this with Opus 4.7?
English
one of my favorite quality of life features
ClaudeDevs@ClaudeDevs
Claude Code in the terminal will now show recaps when you switch focus away from the session and then come back. This should help you stay more in flow while multi-clauding.
English
To everyone vibe coding with AI (Claude, Grok, Cursor, etc.):
If you're touching sensitive data, PII, money, or client deals and you don't know security basics please don't ship it live.
Saw another one today: full broker CRM with million-dollar deals, internal emails, all PII, built with Claude, hardcoded secrets, public APIs on Vercel. Zero auth.
We got it taken down. This is the second in two months.
Vibe coding is powerful. Vibe shipping production systems with real data without knowing what you're doing is how breaches happen.
Review your shit. Use proper secrets management. Lock down APIs. Or bring in someone who knows.
English
@TommLesh Brutal stat. When security rules live only in docs, "nobody checked" becomes the default path.
English
@schneider037 File location constraints are the classic gap. Rules without enforcement are just wishes at runtime.
English
Prism just made a summary without asked (one of the four new files). There isn't any summary directory in `knowledge/`.
That's either because the new reciprocal file reference instruction not being clear enough, or the unchanged line 98 of Prism markdown "├── _index.md # domain index — subfolder summary with file counts and last-updated"
orca@schneider037
So, I have a research agent called Prism and it updated the kb when called by the main session agent to search for "web clipper that also works for document", despite main session agent prompt not explicitly telling it to do so. Prism saved four markdown in `articles/`, then one more in `notes/`, that's 7 markdown in `engineering/articles/`. Still lack explicit step to update the second `_index.md` though. In comparison, my other cli bookmark agent was doing fine on updating both index in inbox directory. I wanted to use doc-definition-audit skills again that I made with my agent to remove potentially stale instruction and keep markdown and any related file coherent for Prism. The skills also instruct to load two other brainstorm (five to six operation mode + paretto) and prompt-engineering skill before AskUserQuestion and after reading file and any related to it. But I hit my 5 hour limit again.
English
@rms80 Ha, the context paradox. Capacity grows but quality still falls past 100k.
English
claude code: now we have 1-million-token context! work forever!!
also claude code: hey btw u should be clearing your context more often! 150k tokens too many!!
English
@d_omajime CLAUDE.mdのルールが逆にコスト爆発を招くパターン、意外とよくありますね。実行時に効かせる仕組みが足りないだけかも。
日本語
回答を得られました
1. effortLevel: "high" (最大の原因) — 設定ファイルに "effortLevel":
"high" があり、毎リクエストでextended thinkingが多く使われている
2. Plan.md自動更新ルール — CLAUDE.mdの「変更時は必ずPlan.mdを更新」という
指示により、コード1行変えるたびに2767行(10ファイル)を読み書き
3. プロジェクトCLAUDE.mdが208行 —
全セッション冒頭で読み込まれる固定コスト
4. 会話コンテキストの蓄積 — セッションが長くなるほど全履歴を毎回送信
提案する対策
- effortLevel を "high" → "normal" に変更
- グローバルCLAUDE.mdに「コスト節約ルール」セクションを追加（必要最小限の
ファイルだけ読む、Plan.mdは明示指示時のみ更新）
日本語
