Agents issues on VS Code
Issues the maintainers labelled both platform:vscode and area:agents.
105 issues · 41 open · 64 resolved (61%) · first seen Feb 25, 2026
Is this getting better or worse?
This class of problem is holding steady. 43 new reports in the last 90 days vs 38 in the 90 before — +13%. The open backlog peaked at 41 in 2026-10 and sits at 41 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
78 of these reports name the Claude Code build they were running, spanning 59 releases. Heaviest reporters:
- Claude Code v2.1.283
- Claude Code v2.1.270
- Claude Code v2.1.123
- Claude Code v2.1.296
- Claude Code v2.1.287
- Claude Code v2.1.284
- Claude Code v2.1.280
- Claude Code v2.1.263
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 64 closures. Of the 64 closures with a recorded reason, 11% were closed as completed and 57 as not-planned or duplicate. The most recent completed fix landed in 2026-09.
Workarounds reported by the community
18 of these issues post a workaround someone says worked. The highest-engagement ones:
- [MODEL] Opus 4.8 starts hallucinating results before parallel tasks finish
- …e upstream fix is unknown) 1. is the only durable user-side workaround for both hazards, per longitudinal data in [ 62123 comment 4569840435]( (4.6 clean → 4.7 onset → 4.8 worst, including 1M context). @timmywheels named…
Found in the thread of #63884 · still open - [BUG] VSCode: background agent output streams into foreground chat, disrupting active conversation
- …ines against that seems an unusually lopsided cost/benefit. Workarounds meanwhile 1. — removes from the Agent tool schema entirely; results return as collapsed tool results. 2. per spawn. Note synchronous is not serial —…
Found in the thread of #64651 · resolved - [BUG] 2.1.114 regression: background subagents cannot write to paths listed in permissions.allow — works in 2.1.112
- …rmation Tests performed during diagnosis Test 0 — Discarded workaround as unsafe: The known workaround for the related pre-existing bug 40502 ( in subagent frontmatter) was tested and rejected because it disables the ent…
Found in the description of #50267 · resolved - [Critical] Background agents cannot be stopped, Claude lies about stopping, massive token waste (~1.4M tokens), inconsistent statements
- …nage to debug their own projects. Of course, these are only temporary fixes — not permanent solutions. Our problems can only be fully resolved by Anthropic. I asked my Claude to look through the issues it has read and se…
Found in the thread of #41461 · resolved - [BUG] A renamed session reverts to its derived <dir>-<nn> name after any restart, so ListAgents and SendMessage can no longer reach it
- …on disk, resume just never reads it back. There's a working workaround below — a small process wrapper, tested on two machines — for anyone who needs this fixed today rather than upstream. If I rename a session, the name…
Found in the description of #95807 · 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 1–100
Ranked by community engagement (reactions weighted over comments).
- [MODEL] Opus 4.8 starts hallucinating results before parallel tasks finish
- [BUG] VSCode: background agent output streams into foreground chat, disrupting active conversation
- Sonnet model returns 'Rate limit reached' despite 0% usage
- VS Code: [ Claude Extention ] Can't see what subagents are doing or if they're stuck.
- [BUG] 2.1.114 regression: background subagents cannot write to paths listed in permissions.allow — works in 2.1.112
- [BUG] Custom agent cannot be invoked with @-mention in VSCode extension
- [Critical] Background agents cannot be stopped, Claude lies about stopping, massive token waste (~1.4M tokens), inconsistent statements
- [BUG] Default custom agent in VSCode extension is ignoring the "tool" directive
- [Bug] Random text insertion in agent harness under high concurrent load (20+ agents)
- [Feature Request] [VS Code] Invoke a custom subagent as the main conversation agent
- [Feature Request] Add per-subagent observability (model, effort) to /agents Running tab and agent JSON output
- [Bug] Sub-agents stuck in infinite loops consuming excessive tokens
- [BUG] Claude Code over-consumes context by not delegating bulky operations to subagents
- [BUG] A renamed session reverts to its derived <dir>-<nn> name after any restart, so ListAgents and SendMessage can no longer reach it
- [BUG] Workflow harness retries indefinitely on HTTP 429 rate_limit — 97 agents, 2M tokens burned in 34 seconds ()
- [BUG] Claude Code — Per-Agent Permission Control Gap
- [BUG] Sub-agent permission mode inheritance broken — permissionMode frontmatter and parent acceptEdits both ineffective for Edit tool calls
- [Bug] Background sub-agent output files empty (0 bytes) on Windows
- [Bug] Unexpected high token consumption with non-Claude subagents and orchestrator models
- [Bug] Unexpected high token consumption with multi-agent Opus orchestrator spawning Sonnet agents
- [Bug] Excessive agent spawning causes token bloat for simple code analysis tasks
- [FEATURE] VSCode Extension: Missing @agent syntax support and unclear agent call visualization
- Cross-session messaging (ListAgents/SendMessage) permanently disabled after update, survives reboot/reinstall
- Subagent tool calls: PreToolUse denials don't surface to the parent, and hook payload carries the parent's session_id
- Sub-agent attempts `exit_plan_mode` as Bash command, fails with Exit code 127
- Model asserts measurement provenance it never obtained (fabricated timestamps and command results)
- [Bug] Subagent polling wake-ups consume excessive tokens for no-op status checkssu
- [Feature Request] Isolate subagent execution environments to prevent tool call (and other) conflicts
- [BUG] Subagent grep outlives its session, ate 11 GB of RAM
- After compaction and subagent return, Main agent set's itself from Sonnet 4.6 to Opus 4.7 without user interaction
- Scheduled Task: Claude Remote failed to switch branches, regressed prior fixes, and consumed tokens without progress
- [BUG] Agent tries to find outside of the project
- [BUG] Sub-agent file writes lost — ephemeral Write tool results not persisted to working directory
- [BUG] - Endless loop and excessive tool calls by sub agents contributing to massive spike in token consumption (Max plan user)
- [Bug] Agent Teams split-panes mode disabled in VSCode terminal with tmux
- Agent Teams state is wiped on session restart — teams and tasks do not persist
- Extension host OOM-crashes within seconds to ~1 min of launching 2+ parallel background Agent subagents (Windows)
- Claude Code (Opus) lied to the user about progress, overrode explicit instructions, and burned ~9.5M subagent tokens while the assigned task barely advanced
- Custom subagent silently missing from Agent tool's available list despite valid, sibling-identical frontmatter
- [BUG] Recent VSC extension update seems to have Claude hanging up occasionally.
- Parallel subagent file edits clobber each other in the IDE diff editor (N−1 spurious permission-denials); diff-review lock is per-agent, not per-IDE-connection
- [BUG]
- Agent Teams: Permission requests should show as native UI popups in VS Code extension, not as chat messages
- [BUG] Background subagent tool calls silently denied — permission dialog never surfaced
- Agent tool internal output (agentId, file paths, instructions) exposed in mobile UI
- [BUG] VS Code extension: subagents are shown running with the main session model despite `CLAUDE_CODE_SUBAGENT_MODEL` being set
- Peer SendMessage reply nudge stamps from= with agent-type name, not reachable session ID, causing infinite unreachable-address loop
- [FEATURE] Configurable peer-registration session name (per-project prefix) for cross-session messaging
- Spoofed background-agent completion notification enabled a prompt injection attempt (correctly refused, full report filed on HackerOne)
- [Feature Request] Add model selection prompt for Fable subagent creation
- [FEATURE] Compact display option for in-process agent team panel (task text makes it tall in VS Code integrated terminal)
- [BUG] Claude autonomously ran background scripts calling a paid external API, causing ~$29 unintended charge
- [BUG] Agent Teams' `SendMessage` resume discards subagent's model override to default model
- Sub-agent tool calls/logs incorrectly visible in VS Code extension 2.1.160 (hidden correctly in 2.1.39)
- [BUG] Issue while creating agent on Windows.
- [Bug] Child agent inherits incorrect model version from parent configuration
- Agent Teams: SendMessage writes mailbox JSON without `type` field → reader UI shows 'Unhandled case: [object Object]'
- [BUG] Claude Code Issue: Subagent Responses Not Returned to User
- [Bug] Quota limit exceeded causes loss of agent work
- [Bug] Excessive deletion of worktrees in repeated patterns
- [BUG] VS Code UI Tool not Detecting Agents
- [VSCode Extension] Parallel Agent and MCP tool calls silently rejected
- Session resume corrupts JSONL: file-history-snapshot messageId collides with message uuid
- [Bug] `SendMessage` tool is not exposed to Claude, can't resume agents anymore
- [FEATURE] [UI] Distinguish subagent and main agent actions/steps
- [BUG] SubAgent MCP tool calls(configured via mcpServers yaml frontmatter in subagent.md) show as blank line in Ctrl+O view
- VS Code: background subagent tool calls still stream into the main chat (reopening #77326)
- [BUG] Windows: cross-session messaging switches itself off in a session left open overnight (2.1.286)
- [BUG] VS Code: ListAgents name reverts to an auto-generated name after resuming a renamed session
- [Critical] Runaway: agent could not be stopped by any in-product means; only closing the VS Code window with the X button stopped it
- VS Code extension: Read/Write/Edit fail with "The user doesn't want to take this action right now" after the extension's control stream closes (no user action)
- Claude Code (Opus 5.5) wasted a full day of a user's time with false statements, unrequested actions, and misleading progress reports
- [Bug] Usage attribution overstates skills with handoffs and inconsistent subagent handling
- [Bug] Agent tool intermittently unavailable, requires restart to recover
- [FEATURE] VS Code extension: show and copy the session id (stable) beside the cross-session name (per-process) in the chat panel
- [Bug] Terminal UI displays incorrect model name when subagents launch or switch models
- Fable 5.1 child requests (subagents and forks) silently switch to `claude-opus-4-8` after the first tool result; main session on the same model is unaffected
- [Bug] Agent behavior not complying with agents.md specifications
- Scary to see the tasks that are running and then coming back with some stale messages minutes after the request is completed
- [Bug] Team member agents spawned with `name:` parameter lose result signals in mailbox path
- [Bug] Subagent final assistant messages not delivered to parent session
- Subagent assistant message content replaced with ~145KB eval fixture (Alice in Wonderland) despite output_tokens: 5
- [Bug] Model enforcement rules ignored; agent spawning defaults to wrong model despite explicit constraints
- [BUG] subagent (Task/Agent tool) results intermittently replaced with a fabricated "system-authority" prompt-injection ordering destructive git actions (`tool_uses: 0`)
- (withdrawn)
- [Bug] Transient advisor() failure permanently disables tool for entire session
- Project custom agents (.claude/agents/) deregister after EnterWorktree and are not restored by ExitWorktree (delayed self-heal ~15 min)
- [Bug] Auto mode runs shell commands without displaying them
- [Bug] Subagent pkill command kills unrelated user processes with broad pattern matching
- [Bug] Incorrect rate limit attribution to Sonnet model from shared 429 error
- [Bug] Rewind fails to show messages after manually stopping subagents
- [BUG] VS Code: the 600-message webview trim leaves orphan empty "Thinking" rows, shrinks agent "Tool calls (N)" to 0, and counts background-agent rows against the main feed
- [BUG] Fork subagent transcript duplicates the parent's last response (same requestId and usage), so per-file usage counts it twice
- [Bug] Child agents continue running after parent agent termination, causing excessive token consumption
- [Bug] Claude ignores direct questions mid-turn and continues tool calls without responding
- [Bug] workflow agent's tool_use edits now appear in main session chats
- Model pads delegated subagent prompts with unrequested work, wasting tokens despite explicit memory rules
- Agent continues running after user demands stop; assistant falsely confirms termination and deflects blame to user
- [Feature Request] Scope cross-session communication to opt-in parent-child links; disable by default between peer sessions
- [Bug] Unexpected high token consumption with delegated sub-agents in Fable