Agents issues on Windows — page 3
Issues the maintainers labelled both platform:windows and area:agents.
524 issues · 200 open · 324 resolved (62%) · first seen Feb 9, 2026
Is this getting better or worse?
This class of problem is still growing. 259 new reports in the last 90 days vs 175 in the 90 before — +48%. The open backlog peaked at 200 in 2026-10 and sits at 200 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
337 of these reports name the Claude Code build they were running, spanning 141 releases. Heaviest reporters:
- Claude Code v2.1.289
- Claude Code v2.1.258
- Claude Code v2.1.233
- Claude Code v2.1.296
- Claude Code v2.1.123
- Claude Code v2.1.284
- Claude Code v2.1.202
- Claude Code v2.1.92
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 35 days across 324 closures. Of the 324 closures with a recorded reason, 19% were closed as completed and 264 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
157 of these issues post a workaround someone says worked. The highest-engagement ones:
- [FEATURE] Add Windows Terminal as a split-pane backend for agent teams (teammateMode)
- …ng an entire Linux subsystem just for pane splitting 3. Use workaround scripts that attempt to bridge the gap via hooks (fragile, unsupported) Proposed Solution Add as a valid backend, with auto-detection support. Detect…
Found in the description of #24384 · still open - Subagent model routing is broken — all mechanisms resolve to parent model (Opus)
- …pe), which the changelog doesn't mention. The full-model-ID workaround noted above likely still applies there.
Found in the thread of #43869 · still open - [BUG] Windows/MSIX 1.28929.0: cross-session messages land in the target's composer but are never submitted — session never responds
- …nd the content to the recipient window directly. 6. Current workaround 1. Write the actual content to a shared file 2. Have a human paste the file path into the recipient window Nothing else got a message through. from t…
Found in the thread of #86069 · still open - [BUG] /model and /effort mutate global settings.json — breaks agents/fleet view; no way to configure per-agent model or effort
- …ents ignoring env values — confirming that even the env var workaround for model configuration was broken for background agents until yesterday. Steps to reproduce 1. Open 2. Launch agent A on Opus — note current model/e…
Found in the description of #66402 · still open - advisor() tool inflates reported input tokens by forwarding full transcript, triggering premature auto-compaction on extended context models
- …ction and destroys the context it was supposed to help with Workaround Setting and/or using a PreCompact hook that blocks auto-compaction: This blocks all auto-compaction and relies on manual . Not ideal but prevents the…
Found in the description of #53065 · 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 524 matches; the statistics above cover all 524.
- [FEATURE] Let in-flight subagents resume from their own checkpoint after a usage-limit interruption, instead of restarting from scratch
- SendMessage returns success but silently drops messages; Remote Control sessions should still do local inter-session messaging
- Windows: CCD session host repeatedly exits killing live background tasks ('torn down with N live background task(s)') — Desktop 1.37937.3 / runtime 2.1.246, filesystem MCP relay flapping
- Custom subagents fail to register after cold start when settings.local.json grows very large (348KB / 2438 valid permission entries) -- fixed by removing the file
- Suspected prompt injection in background subagent tool-result stream (Bash), instructing agent to conceal file state from user
- [BUG] send_message` (ccd_session_mgmt) renders in the recipient's UI but is never injected — regression between app 2.1.222 and 2.1.227
- [BUG] `agents_cross_session_inbox` gate off: no UDS inbox is bound, `@` lists no peers, but send_message still reports success
- SubagentStart/SubagentStop counts stay mismatched: SubagentStop not emitted for subagents that terminate during an API-error burst
- [BUG] Subagent output discarded when spend limit terminates fanout — tokens billed, zero results returned
- Autocompact never fires proactively at the context edge — autonomous/headless sessions park until the next user message (desktop app + Agent SDK)
- [BUG] Returning from agent view cancels & respawns in-flight subagents (work re-run, double token spend, misleading "already done" reports)
- Session (5-hour) rate limit surfaces as 'You've hit your monthly spend limit' — wrong error message
- [BUG] Title: Nested sub-agents spawned by general-purpose agents are invisible and uncontrollable by both the parent session and the user
- Output stream corruption when firing Fable subagents: unrelated buffer text spliced into a file-write status line
- Fable 5 autonomous session destroyed user's main app via unverified assumption, verified its own breakage as green, burned ~75% weekly quota — 150 itemized failures
- [BUG] Team/subagent SendMessage notifications routed into the user prompt input buffer (editable, up-arrow-recallable)
- [BUG] Recent VSC extension update seems to have Claude hanging up occasionally.
- Bug: spawn/task tool mis-encodes non-ASCII prompt into lone UTF-16 surrogates -> API 400 "invalid high surrogate" on spawned session's first request
- Workflow tool: args not delivered to a workflow that validates inputs at the top (regression ~2.1.158→2.1.185)
- [BUG] Uncontrolled subagent multiplication
- [BUG] A `fork` subagent dispatched for a read-only/critique task adopts the parent's role and takes unrequested agentic actions (launches Workflows/Agents)
- Windows: orphaned worktree agent processes hold directory lock after session ends, blocking folder deletion
- [BUG] claude mcp serve exposes Agent tool but registry is empty
- [BUG] Subagent worktree mkdir is non-idempotent — recurring close-for-inactivity (refs #39680, #51023, #54706)
- [BUG] Agent Teams SendMessage renders teammate message bodies instead of compact summaries
- Subagents on Pro 1M tier: trivial probes pass, real workloads fail at first tool call (probe-vs-workload divergence)
- EnterWorktree / Agent isolation:worktree fails on Windows when session CWD contains a space
- [Bug] Sub-agent `rm -rf` with case-insensitive path collision destroys workspace; no sandbox, confirmation, or orchestrator interception
- C:/Program Files/Git/compact does not clean up subagent directory, causing stale agent state in compacted sessions
- FleetView: duplicated section headers after navigating into an agent and back
- FleetView dashboard layout broken after returning from session (v2.1.140, Windows)
- Sub-agent returning "Tool result missing due to internal error" silently hangs parent — no error propagation
- Subagent Bash auto-approval regressed between runtime 2.1.121 and 2.1.128 under "Accept edits" mode
- Windows: background agent .output files always 0 bytes (symlink fallback)
- Sandbox blocks tmux (psmux) when cwd is home directory on Windows
- Custom agents via --agent flag silently ignored on Windows (2.1.87)
- Agent bash shells should not source user .bashrc/.bash_profile
- [BUG] Agent tool with isolation: "worktree" fails with EEXIST when .claude/worktrees/ directory already exists
- Background agents hang silently + no incremental output for large file generation
- [BUG] Subagent-created teams persist on disk after session ends, blocking future team creation
- [FEATURE] Agent Teams: add Windows Terminal split pane support
- SONNET 4.6
- Race condition corrupts .claude.json when using Agent Teams on Windows
- [BUG] Claude launched a 137-agent Workflow for a single-file audit, against a standing "no multi-agent" instruction (~53% of weekly usage in 1 hour)
- [MODEL] Orchestrating session ignored the project's written agent and batch limits, then retried its own load-caused timeouts for 2 hours: ~12% of a weekly Max allowance spent, nothing merged
- [BUG] Reaching the 5-hour session limit kills in-flight workflow agents instead of pausing them
- Agent-scoped inline mcpServers (stdio) never forwards trailing directory argument...
- Rate limits hard-kill in-flight agent runs, and the model has no budget visibility to avoid it
- [BUG] Windows desktop: recurring MoAppHang mid-subagent-spawn (same Watson bucket across 4 versions); hang wedges MSIX package into "Can't open this app" repair loop
- Trust dialog on a chip's "Start with worktree" names the spawn_task cwd, not the worktree the session runs in
- [edit] I'm dumb and was melodramatic.
- [MODEL] Main agent burns consecutive no-op turns idle-waiting for a backgrounded subagent
- [BUG] Claude Desktop (Windows/MSIX): fatal GPU-process crash (exit 101457950 / 0x060C201E) on preview surfaces — --disable-gpu-compositing does NOT mitigate; package left Modified/NeedsRemediation
- Cross-session message bodies arrive HTML-entity-encoded: Desktop escapes and stamps encoded=1, the CLI never decodes
- [Feature Request] Add token cost visibility and guardrails for subagent fork context inheritance
- Custom subagent registry silently caps at 3 (not documented), drops alphabetically-last entries with no load-time error
- Agent discovery: initial scan silently skips some project .claude/agents/*.md; the same files' names absent all session while a newly created agent hot-loads fine
- Background subagents (e.g. via /code-review) silently auto-deny gated Bash commands instead of prompting
- Subagent Bash cwd resets to another subagent's worktree, with no cwd param to pin it
- Cross-session messaging (agents_cross_session_inbox) is enabled on Linux but not Windows for the same account
- send_message: cross-session message is written to the target transcript and rendered, but never enters the receiving agent's context
- [BUG] Windows Desktop: cross-session send_message reports success and renders in the target UI, but is never enqueued/persisted (ghost delivery, lost on restart)
- Delivery succeeds, execution does not
- Cross-session send_message never reaches the receiving agent: the UI renders the message, the session never learns it exists (both when the recipient is idle and when it is actively working)
- Agent tool: run_in_background: false is not honoured — subagents complete but never deliver their output as the tool result
- Agent/model launches request effort=xhigh against thinking-disabled model, crash with 400
- Message delivered to an idle session cold-starts a "ghost turn" (no assistant output), and the queued message is silently lost on app restart — no transcript record remains
- Agent-team registry: members[] never pruned across /clear; internal agents_killed event mislabeled 'stopped by the user' and killed live agents
- Background tasks spawned by subagents leak: no process, no handle, no way to clear
- [BUG] Agent frontmatter `model: opus` runs the parent session model instead of stepping down to the newest org-allowed Opus (per-invocation `model: opus` param steps down correctly)
- Agent tool: isolation:"worktree" subagent read parent session's private transcript outside its worktree
- Chip/spawn_task session start runs `git checkout -b claude/<name>` in the PARENT repo when the created worktree dir is not a real worktree — silently flips the shared main checkout for ALL parallel sessions (Windows 11)
- [BUG] Plan Mode → Escape → Manual Mode: agent ignores STOP and keeps running workers until full process kill
- [BUG] Fable 5 → Opus 4.8 mid-session safeguard auto-switch makes the post-switch model re-spawn duplicate agents (double token burn)
- [BUG] Workflow/background-spawned agents block indefinitely on unanswered permission prompts — no auto-deny, timeout, or idle watchdog (55-min silent stall)
- [BUG] Background agent handle becomes unreachable after context summarization, even though agent is still running
- [BUG] Parameter-based permission rules not working on Agent tool
- Agent teams: SendMessage duplicates the message body ~3x per call (model-invented `content` field + result echo)
- [BUG] /code-review skill is now being automatically loaded in sub-agents the code-review skill spawns. Causing it to keep recursively making more agents
- Workflow tool: deep-research synthesis step accepts schema-valid-but-empty (placeholder) agent output as final result
- [BUG] Async subagent stopped-state triggers infinite recursion storm in v2.1.202 (Windows)
- extensions.worktreeConfig left in .git/config after worktree-isolation cleanup (Windows; follow-up to #34225)
- Workflow/StructuredOutput: retry cap (5) exceeded discards the entire subagent run with no raw-text fallback
- Sub-agents return "0 tool uses / Done" when parent session has deferred tool schemas
- Custom subagent .claude/agents/tdd-worker.md never registers, identical siblings work fine (Agent type 'X' not found)
- MCP tool calls from sub-agents serialize single-element string arrays as scalars (server → JSON-RPC -32603)
- [BUG] Windows: worktree git-repo detection fails at call time — EnterWorktree({name}) and Agent isolation:"worktree" both falsely report "not in a git repository"
- Subagents & structured-output requests 400 against third-party endpoints requiring a `thinking` field (CC omits `thinking` on those requests)
- [BUG]
- [MODEL] Fable 5
- [BUG] Claude autonomously ran background scripts calling a paid external API, causing ~$29 unintended charge
- Session self-management API: allow agents/sessions to rename themselves and assign to groups
- [BUG] Agent Teams (in-process): unbounded team-lead inbox stalls teammate→lead message surfacing at ~1MB / 1000+ msgs
- Claude forgets actions performed earlier in the same session (intra-session amnesia)
- bug: Task subagents receive empty MCP tool surface (main-thread works, subagent context drops all mcp__* tools) — Windows + Opus 4.7
- WorktreeCreate hook 'Hook cancelled' for worktree Agent when dispatched alongside a sibling Agent call in one message
- [BUG] Issue while creating agent on Windows.
- [Critical] /init command causes Explore agent to read files from unrelated project directories
- Stop hook infinite loop: hook execution creates new task files that re-trigger the hook
- [BUG] UI get's stuck in Win 11 desktop app while running subagents in Cowork