Agents issues on macOS — page 3
Issues the maintainers labelled both platform:macos and area:agents.
1,106 issues · 313 open · 793 resolved (72%) · first seen Jul 25, 2025
Is this getting better or worse?
This class of problem is holding steady. 376 new reports in the last 90 days vs 467 in the 90 before — -19%. The open backlog peaked at 313 in 2026-10 and sits at 313 today.
Derived by counting each matching issue from its creation date until it closed. First-party
computation over the anthropics/claude-code tracker — the trend is not published
anywhere upstream.
Which builds report this
833 of these reports name the Claude Code build they were running, spanning 183 releases. Heaviest reporters:
- Claude Code v2.1.119
- Claude Code v2.1.198
- Claude Code v2.1.233
- Claude Code v2.1.201
- Claude Code v2.1.251
- Claude Code v2.1.197
- Claude Code v2.1.260
- Claude Code v2.1.126
Report volume, not a defect count: an issue can mention a build without that build having caused the failure. Versions come from the version string in the bug-report template, matched against real published tags.
Do these get fixed?
Median time to close is 34 days across 793 closures. Of the 793 closures with a recorded reason, 20% were closed as completed and 634 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
322 of these issues post a workaround someone says worked. The highest-engagement ones:
- Built-in Plan agent ignores parent settings.json permissions and repeatedly prompts for pre-approved tools
- …agent permissions (e.g., ) 3. Option C : Add a setting like Workarounds Attempted None available. Cannot modify built-in agent configurations. Impact - Interrupts workflow with repeated permission prompts - Makes Plan mo…
Found in the description of #10906 · still open - [BUG] Sub-agents claim successful file creation but files don't persist to filesystem
- …nsparency issues - 3408 - Task tool crash scenarios Current Workarounds - Use direct tool calls instead of Task tool for file operations - Always verify file creation externally after sub-agent operations - Avoid sub-age…
Found in the description of #4462 · still open - [Bug] Regression: malware reminder on every Read still causes subagent refusals in v2.1.111 (fix from #47027 / v2.1.92 did not hold)
- …ain to the subagent that it's misreading" is not a reliable workaround — it can increase suspicion, mirroring the OP's subagent-2/3 reasoning chain. Per-agent inconsistency, same repo, same reminder, same session: two ot…
Found in the thread of #49363 · resolved - [CRITICAL] Subagent spawning and subagent pattern bugs trigger infinite recursion, infinite token usage, grossly inefficient token usage, and lost accumulated subagent work.
- …The agent ignores this and instead spawns a child agent to work around the denial. That child hits the same wall and spawns another. This repeats indefinitely. The permission denial is functioning as a recursive agent l…
Found in the description of #68619 · still open - [Bug] Plan agent uses main branch path instead of git worktree directory
- …an agent should use the current working directory Temporary Workaround In the meantime, when the Plan agent gives you a plan, you may need to: - Review the plan carefully for any incorrect path assumptions - Manually cor…
Found in the description of #11239 · still open
Excerpts are quoted from the linked issue so you can judge relevance before opening it. Threads that only ask for a workaround, or report that there is none, are excluded. These are community suggestions, not official guidance from Anthropic — read the full thread before applying one.
Matching issues 201–300
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 1,106 matches; the statistics above cover all 1,106.
- Subagents should inherit settings.local.json permissions
- Sub-agent context isolation: parent MCP server registrations leak into sub-agent tool output
- Allow Claude to stop background agents.
- initialPrompt not auto-submitted when using --agent flag
- [BUG] Agent teams silently fall back to isolated subagents — runtime inconsistently chooses team path
- [BUG] Session crashes when worktree directory is deleted while Claude is operating inside it
- [BUG] Background task output files grow without limit, can fill disk after session crash
- [BUG] Claude code lose track of running subagents
- [BUG] Claude Code stopped respecting model in agent altogether
- Task output file grows unbounded — single file hit 107GB in 3 minutes
- [BUG] Explore subagent inherits parent model (opus) instead of using Haiku as documented
- [BUG] Auto-mode classifier blocks recording a user's own explicit approval ("Instruction Poisoning"), stalling an approved overnight multi-agent run
- Worktree-isolated subagents launched from a worktree session can't Write/Edit their own worktree (guard resolves the parent's worktree root)
- SubagentHandback from an in-process teammate's subagent fails with "the agent that spawned you is no longer running" while the teammate is alive (auto mode, 2.1.280)
- [BUG] print-mode background wait ceiling discards the cause of death, so killed agents resurface as unexplained orphans with the wrong recovery advice
- Usage-limit warning names the parent model, not the subagent's model
- [BUG] background-mode permission dialog shows only a Deny button, no way to Accept
- [BUG] Fork subagents ignore prompt-level write restrictions due to unconstrained inherited tool access
- Background Bash task completion never resumes in-process subagents — they idle forever while finished work sits on disk
- [Bug] Agent repeatedly calls Bash(true) instead of awaiting response
- [Bug] Session limit warning not surfaced to agent; background workflows continue consuming quota after threshold
- Workflow tool: args object delivered to script as JSON string (schema's untyped `args` → model emits string; dup of #72248) — bug 2 in original report retracted
- Nested code-review fan-out spawned 877 agents despite 5-level depth limit
- [Bug] Sub-agent backgrounded processes orphaned/SIGHUP'd on turn-end; no auto-resume on background exit
- Native worktree GC (worktree.cleanupPeriodDays) deletes DIRTY and UNPUSHED agent-* worktrees, contradicting the documented clean+pushed-only cleanup behavior
- Nested subagents remain stuck as "Running" in Background tasks panel after parent subagent completes
- [BUG] Background sub-agents stay "In esecuzione" in the Activity panel after they've completed
- [BUG] Subagents going idle
- [Bug] Multi-agent workflows terminate on transient server errors instead of retrying with backoff
- Subagent sessions spawned by multi-agent tasks remain active after parent session exits
- [bug] [area:agents] [platform:macos] Agent(isolation: "worktree") post-creation setup silently rewrites...
- [Bug] Auto-mode spawns infinite monitor loops on batch classifier denial, causing runaway API usage
- [Bug] Excessive agent spawning causes token bloat for simple code analysis tasks
- Babysit agents stop with 'socket connection closed unexpectedly' API error
- [BUG] agent view disabled by gatewayAuth gate on Max+Stripe accounts
- Runaway Agent task burns tokens after completion; agentId ≠ task_id makes it unkillable from parent session
- AskUserQuestion: Agentic processing breaks when sub-agent is asked to use AUQ by initial user-prompt
- [Bug] Auto-mode yields to user after Agent tool returns instead of continuing pipeline
- SendMessage concurrent writes lose messages via array rewrite
- [Bug] Agent Teams launch crashes with Ink renderer: Box nested inside Text component
- [BUG] After agent used advisor the conversation got broken and keeps returning API Error: 400
- [BUG] origin remote URL corruption via GIT_COMMON_DIR + rbs/ruby-lsp in devcontainer
- [BUG] API returns thinking-only response with no text block — tokens consumed but response lost
- [BUG] TypeError: toolUseContext.getAppState is not a function when rendering permission prompt
- [BUG] Lead agent crashes with "toolUseContext.getAppState is not a function" when a teammate requests permission (regression in 2.1.93–2.1.112)
- Agent tool not available at runtime for plugin-defined agent types despite tools: frontmatter declaration
- Claude Code node process deletes user-managed agent files from ~/.claude/agents/
- Worktree path corrupted by hook stdout — JSON output concatenated into path
- [BUG] Spawning subagents with isolation: "worktree" immediately fails auth
- [BUG] CLAUDE_CODE_DISABLE_BACKGROUND_TASKS ignored when spawning multiple subagents
- [Bug] Background agent resume creates retry loop — model doesn't wait for completion notification
- [Bug] Agent teammates stuck in indefinite plan-mode approval loops
- Lead session exits cleanly seconds after a teammate approves a shutdown_request (2.1.280, in-process teammate)
- [BUG] Teammate subagents of a forked-skill agent cannot reply to their parent; SendMessage to "team-lead" false-succeeds into an orphaned inbox
- Receiver is never notified when a held cross-session message expires — the message is abandoned, not denied, and nobody downstream learns it existed
- Held-then-approved cross-session messages arrive with no marker — receiver cannot tell it was gated, and reports the hold never happened
- [Bug] Main session persists and executes autonomously after /exit, ignoring CLAUDE.md directives
- [Bug] Agent removal fails when worktree directory deleted out-of-band
- Background subagent never re-woken when its tracked background Bash child exits (child survives, wake event lost)
- [Bug] macOS arm64: Second background agent spawn fails with "fork failed: Device not configured" in 2.1.207
- Esc interrupt kills all background agents and marks them permanently unresumable
- [BUG] Opus 4.8 subagents die on turn 1 emitting fabricated System:/Assistant: scaffolding (0 tool calls, <br>+System: prefix) — 13x in one session, JSONL-verified, spawn-time variant of #68367
- Resumed subagent (SendMessage) detail panel renders garbled transcript: turns inverted, and raw SendMessage envelope shown on reopen
- [Bug] SendMessage(to: "main") from background subagent silently dropped instead of reaching spawner
- [Bug] Subagents billed after monthly spend limit exceeded; false clean review result on agent failures
- Agent tool: subagents fabricate a "waiting for background agent" stall with no real background agent ever spawned
- [Feature Request] Restore "/agent" command wizard for interactive agent setup
- Agent-mode sessions on unattended Mac: silent mid-turn drops after tool_result + UI stream detaching from healthy sessions (WarmLifecycle idle_timeout suspected)
- macOS: background agent workers reaped mid-work (reapedMidWorkAt), never respawn — refines #72221
- [BUG] Ultra workflow auto-scales agents (up to ~130) and triggers Rate Limit / IP Block without user-specified count
- [BUG] Subagent (Task/Agent) spawns fail with 400 "thinking.disabled.display: Extra inputs are not permitted"
- Built-in dynamic workflows (deep-research) have no model routing — all sub-agents inherit the main-loop model, burning Opus on grunt work
- [BUG] Workflow parallel() sub-agent loops forever emitting empty StructuredOutput calls; no retry cap and no per-agent timeout stalls the whole workflow
- Usage-limit gating is invisible to unattended sessions — agent-team orchestrator froze 5h with zero signal in transcript or terminal
- Subagents return fully hallucinated results with zero tool calls; leaked tool-call XML in text; two fabricated 'prompt injection detected' reports
- Bash tool results silently lost — agent waits forever (no timeout fires); correlates with Remote Control sessions
- [Bug] Silent model fallback from Fable 5 to Opus 4.8 overrides explicit model pinning
- [BUG] Claude Code DevOps subagent runbook missed an 8.5-day AWS cost anomaly despite explicit daily monitoring requests — $80 burn, structural gap
- MCP tools inaccessible in --agent mode even with explicit --mcp-config flag
- [BUG] Excessive token consumption in web search workflows - exponential token usage growth
- [Bug] Agent dispatch consumes excessive tokens without course-correction mechanism
- [BUG] Workflow subagent: one stall is retried 6× at full token cost (no across-retry progress check); the no-progress signal can't tell slow-but-alive from hung
- [Bug] High-effort session tier not propagated to sub-agents with pinned models
- Behavioral drift away from user-granted autonomy in long sessions despite memory rules
- OAuth re-auth for `claude agents` fails with ECONNREFUSED on `org:create_api_key` scope
- Worktree virtual-overlay evaporation: Edit/Write operations silently target ghost files when worktree disappears mid-run
- [BUG] Edit tool silently no-ops on disk while Read returns the edited content (sub-agent worktrees + main session)
- macOS: background agent binary lacks bundle ID, causing TCC permission churn on upgrades
- Bug: LSP tool (jdtls-lsp plugin) not propagated to spawned subagents
- Agent teams: teammates die immediately — harness spawns claude without a pty, causing stdin/print mode failure
- Model carries imperative actions from earlier system-reminder payloads past an explicit plan-scope narrowing
- [Bug] Agent pauses indefinitely when hiring team agents, requires manual trigger to resume
- [BUG] Agent View not enabled for Claude API users despite having required version (2.1.140)
- DISABLE_TELEMETRY silently disables GrowthBook — which also gates preview features (Agent View) and remote killswitches
- [BUG] Startup launches many headless sdk-cli sessions in Ghostty, causing macOS memory pressure/freeze (v2.1.126)
- [BUG] Team-mode teammate crashes with "<Box> can't be nested inside <Text>" when inbox contains permission_request / permission_response JSON
- Concurrent subagent dispatch: child re-prompts on Edit / Write even when parent has explicit allow entries
- Agent teams permission routing crashes CLI on teammate tool approval (regression in 2.1.111)
- Agent Teams: teammate spawn fails after brew upgrade due to hardcoded Caskroom version path
- Stop hook false-positive: stale subagent IDs block session end