Claude Code Output Truncated by Context Window
A long agent reply gets cut off mid-sentence under context pressure. Usually compaction, max_tokens cap, or runaway tool output eating the budget.
#agent
Articles tagged with #agent
Claude Code Subagent Result Not Relayed to Main Session
Your Task/Agent subagent completed cleanly, but the main session shows nothing or a wrong summary. Usually a final-message format, context-window, or thread-routing issue.
Cursor YOLO Mode Runs Shell Commands Without Confirming
Agent mode keeps running shell commands without the confirm prompt, even when you flipped the toggle off. Usually allowlist scope, stale settings, or a workspace override. Fix it.
Codex Stalls on a Merge Conflict or Resolves It the Wrong Way
Codex hits conflict markers and either stops, picks the wrong side, or hand-edits the markers into the file. Pre-rebase setup.sh and explicit AGENTS.md rules fix it.
Codex Added a Package but the Lockfile Did Not Change
Codex edits package.json but never ran npm install, so package-lock.json diverges. How to enforce lockfile updates via setup.sh, CI, and AGENTS.md.
Codex Creates a New TypeScript Interface Duplicating One That Already Exists
Codex defines a fresh User or ApiResponse type when an identical one lives elsewhere. How to make the agent search first via AGENTS.md, indexes, and ts-morph.
Codex Amended or Rebased Git History You Did Not Want Touched
Codex ran git commit --amend or interactive rebase from an old README, rewriting shared history. How to lock the agent to forward-only commits.
Codex PR Description Says "Refactored Components" and Nothing Else
Codex emits a one-line PR body with no detail. How to force a structured description with before/after, why, and test plan via templates and AGENTS.md.
Codex Says Tests Passed but Actually Skipped the Failures
Codex reports green tests but the failing cases were filtered out, marked .skip, or bailed early. How to force honest test reporting before merge.
Codex Bails Out When the Patch Gets Too Large
Codex hits a context or token cap mid-task and stops with a partial diff. How to scope tasks smaller, split across PRs, or move to a larger-context model.
Codex Writes React.FC, componentWillMount, and Other Deprecated APIs
Codex outputs API patterns deprecated years ago because its training data is dated. How to lock it to modern APIs via AGENTS.md, ESLint deprecation rules, and reference docs.
Claude Computer Use Stuck in Action Loop on Same Button
Claude Computer Use keeps clicking the same button or field over and over without progressing — usually a screenshot diff, focus state, or coordinate drift problem. Here is how to break the loop.
Claude Code Permissions Prompt Loop
Every Bash, Edit, Read asks for permission again even after you allowed the same tool — settings.json scope wrong, allowlist pattern too narrow, or hook is re-asking.
Claude Code Token Budget Too Large for One Task
Single task burns 200K+ tokens, hits the context ceiling mid-execution, and forces a summarize. Decompose the task, use sub-agents, or scope reads tighter.
Claude Code Tool Execution Hangs Without Timeout
Bash command runs forever, the agent waits forever — no timeout set means a hung process locks the whole session. Set timeouts, detect hangs, kill cleanly.
Codex Agent Output Conflicts With Prettier
Codex writes a file, Prettier reformats it on save, and now the diff is huge. Fix by running Prettier inside the agent loop and aligning agent style hints to your .prettierrc.
Codex Agent Goes Out-of-Context on Long Repos
On 500k+ line codebases Codex loses track halfway. Fix by scoping the working set, using directory summaries, and forcing the agent to read AGENTS.md before scanning.
Codex Agent Skips Files Containing Binary Data
Codex silently skips PNGs, PDFs, sqlite DBs, compiled artifacts. The fix is allowlisting text extensions, hashing binaries instead of reading, and giving Codex a binary-aware tool.
Codex Agent Spawns Too Many Redundant Tool Calls
Codex re-reads the same file 8 times and re-greps the same query 5 times. Fix by pre-feeding context, requiring a plan first, and capping reads via tool restrictions.
Codex Agent Stops Mid-Task Without Error
Codex halts halfway through a multi-step task with no visible failure — usually a context window, sandbox timeout, or stop-condition issue. Diagnose by checking the last tool call and turn budget.