Claude Code Hook Blocks Edit Unexpectedly
A PreToolUse hook keeps rejecting Edit calls with no clear reason, including safe edits. Usually exit-code logic, stdin parsing, or matcher scope at fault.
🤖 Claude Code & Agent Workflow Issues
Agent loops, wrong-file edits, lost context, multi-agent conflicts, rollbacks, Plan mode issues.
Claude Code / Codex agent workflow failures are rarely "the model is weak" — usually "wrong context" or "too-broad permission." Articles here group by real failure mode: agent loops, wrong-file edits, lost context, broken Plan mode, multi-agent conflicts, messy rollbacks, scattered changes. One symptom, shortest fix path.
Common problems
- Claude Code stuck in a loop — how to break out Context bloat + no done condition — reframe + restart.
- Claude Code edited the wrong file Git restore + explicit "do not touch" list.
- Claude Code lost project context mid-task Summarization dropped details — restart with 5-line summary.
- Claude Code plan mode diverges from plan Plans are guides not contracts — add "no scope creep" rule.
- Multi-agent conflict: two agents edited the same files Parallel agents need scope split — merge manually.
- Agent rollback was incomplete Rollback covers code only — handle artifacts / DB / config separately.
- Claude Code stops mid-task due to token / usage limits 5-hour / weekly / Sonnet-specific caps — handle separately.
- Claude Code ignores your audit report Reports must be numbered checklists, executed by item.
- Claude Code refactor scope becomes too broad Hard scope in prompt + "list, don`t fix" for adjacent issues.
- Claude Code overwrites existing changes Commit / stash before agent runs — check Local History if lost.
- Claude Code misunderstands your project architecture CLAUDE.md needs "do / don`t" examples, not just rules.
- Claude Code keeps asking unnecessary questions Give defaults — only stop for destructive operations.
- Claude Code stuck after partial execution Save state + new conversation — don`t fight the stuck session.
- Claude Code output does not match your original prompt Make the agent restate the task before any tool use.
- Claude Code skips failing tests Ban skip / xit; test edits need separate commits.
- Claude Code accidentally committed a secret Rotate immediately; ban git add .; add secret scan to CI.
- Claude Code creates a pile of unused helpers Delete uncalled exports in review; helpers need ≥ 2 callers.
In-depth articles in this hub
Claude Code MCP Call Times Out Repeatedly
MCP tool calls (filesystem, GitHub, custom) time out in Claude Code, but a direct curl to the server works. Usually transport, stdio framing, or response shape.
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.
Claude Code settings.json Not Loading After Edit
You edited ~/.claude/settings.json but Claude Code ignores hooks, permissions, or env on restart. Usually JSON syntax, wrong scope, or shadowed precedence.
Claude Code Skill Not Discovered or Auto-Triggered
You dropped a skill under ~/.claude/skills/ but /<name> says not found, or it never auto-fires. Usually a SKILL.md frontmatter, path, or trigger-description issue.
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.
Claude Code Bash Sandbox Blocks an Expected Command
Claude Code refuses to run a command you whitelisted — the Bash sandbox rejects it as unsafe or unapproved. Diagnose permission scope, pattern match, and settings.json wiring.
Claude Code Project CLAUDE.md Not Loading Into Context
Claude Code starts a session but ignores your project CLAUDE.md — conventions, commands, and instructions never reach the model. Diagnose path, scope, and merge precedence.
Claude Code Session Resume Loses Memory of Prior Work
You resume a Claude Code session and it acts like a fresh start — forgetting the plan, the decisions, the half-finished refactor. Diagnose snapshot scope, compaction, and memory layers.
Claude Code Statusline Custom Script Errors or Hangs
Your Claude Code statusline shows an error, stays blank, or freezes the prompt — the custom script is failing silently. Diagnose exit codes, output format, and timeout limits.
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.