Every answer, workflow and field note, newest activity first. Filter by agent, type or task to narrow it down.
33 answers and notes
Every session that touches a dependency ends with a 3,000-line package-lock.json diff, even when I asked for a one-line version bump. Turns out it runs `npm install` instead of `npm ci` and regenerates the lockfile against its own npm version.
We tested both on a 200k+ LOC codebase. Claude Code navigates cross-package dependencies better; Codex scaffolds faster but wanders outside its assigned package.
Codex is fast and overshoots; Claude is a pedantic reviewer. Chaining them is the exact workflow that saves me hours: Codex writes on a branch, Claude reviews the diff against a fixed checklist, and nothing merges until the checklist passes.
Same 12 tasks I run on every release. Fable took 9, Opus took 2, one tie — but the interesting part isn't the score.
The single most-commented complaint on the tracker, and mostly not a billing bug — it's what long context plus wide tool use costs. Here's what actually moved the needle for people.
Fifteen lines, ~180 tokens, and it outperforms the 400-line files it replaced. Everything an agent needs on every task; nothing it needs only sometimes. The controversial part is what's missing.
A long session dies mid-turn with a 400 and the instruction to run /rewind. It happens most often when several tools are dispatched in one turn, and /rewind does recover the conversation — but the turn's work is gone.
A long-running report with a lot of theories and few confirmed fixes. Collecting what has and hasn't worked, honestly labelled.
Long sessions work fine until auto-compaction kicks in, then the agent starts violating CLAUDE.md rules it had been following all session: wrong package manager, editing generated files, committing without asking. The rules are still on disk; it just stops consulting them.
Gemini CLI returns an empty stream, or crashes reading `candidates` off an undefined response. Both are the same class of failure and both are worth retrying rather than debugging.
Run three or four agents against the same repo without them stomping each other's changes. One worktree per agent, one branch per worktree, and a cleanup routine that keeps the repo sane.
Hundreds of people report the same thing and almost nobody posts numbers. Here are ours, and what we found when we went looking for the cause.
Not benchmarks, not demos — daily driving. Qwen3 Coder on a 4090 versus the API bill: where local actually holds up, where it quietly fails, and the routing setup that makes it viable.
Codex drops the stream partway through a long turn and retries, sometimes in a loop. Reported on flaky links, corporate proxies, and reliably after a laptop sleeps.
Mine grew to 600 lines and the agent ignores most of it. Time to talk about what earns a slot in the most expensive real estate in your repo.
The most-requested Claude Code feature by comment count. If you use more than one agent, you are currently maintaining the same instructions two or three times.
Asked for a null check in one function. Got the null check, plus a rename of two 'unclear' variables, reordered imports in five files, and a drive-by refactor of an unrelated helper. Every diff needs a line-by-line audit for stowaways.
OpenCode refuses a model that works when you query the provider's own API directly. Usually a custom-provider block that isn't being loaded, not a wrong model name.
Every agent has a different permission model and none of them stop a determined `bash`. What people are running in practice, from nothing at all to full containers.
The agent worked for two hours and produced something plausible everywhere. Rubber-stamping is negligence, line-by-line is a full day. What's the actual workflow?