Agents issues on Linux
Issues the maintainers labelled both platform:linux and area:agents.
356 issues · 126 open · 230 resolved (65%) · first seen Feb 7, 2026
Is this getting better or worse?
This class of problem is holding steady. 158 new reports in the last 90 days vs 136 in the 90 before — +16%. The open backlog peaked at 126 in 2026-10 and sits at 126 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
300 of these reports name the Claude Code build they were running, spanning 134 releases. Heaviest reporters:
- Claude Code v2.1.119
- Claude Code v2.1.281
- Claude Code v2.1.234
- Claude Code v2.1.198
- Claude Code v2.1.251
- Claude Code v2.1.260
- Claude Code v2.1.233
- Claude Code v2.1.296
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 230 closures. Of the 230 closures with a recorded reason, 16% were closed as completed and 193 as not-planned or duplicate. The most recent completed fix landed in 2026-09.
Workarounds reported by the community
106 of these issues post a workaround someone says worked. The highest-engagement ones:
- [BUG] Subagent Write tool rejects .md files named "report"/"summary"/"findings"/"analysis" — no opt-out
- …all users — including those with legitimate workflows — to work around it. The fix could be as simple as a key: . 2. The fallback degrades the multi-agent pipeline. When the Write tool is blocked, the subagent falls bac…
Found in the thread of #44657 · still open - PermissionRequest hooks not triggered for subagent permission requests in Agent Teams
- …The current behavior makes these two features incompatible. Workaround Pre-approve common subagent commands in permissions so they don't need to ask: This lets subagents work autonomously but removes the ability to revie…
Found in the description of #23983 · still open - [BUG] Agent tool isolation: "worktree" has no effect for team agents — agent runs in main repo
- …rktree so the agent works on an isolated copy of the repo." As a workaround, we manually create git worktrees before spawning agents: ...and set in the agent prompt to the worktree path. This works correctly, but the par…
Found in the description of #33045 · resolved - 2.1.178: native team-management tools TeamCreate / TeamDelete no longer surfaced (regression from 2.1.177)
- …in/claude), Linux 6.17.0-14-generic, tool search enabled. - Workaround in place: pinned to 2.1.177 and moved the 2.1.178 binary out of the version pool so the launcher can't re-select it (editing the symlink alone wasn't…
Found in the description of #68721 · resolved - 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 1–100
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 356 matches; the statistics above cover all 356.
- [BUG] Subagent Write tool rejects .md files named "report"/"summary"/"findings"/"analysis" — no opt-out
- PermissionRequest hooks not triggered for subagent permission requests in Agent Teams
- [BUG] Agent tool isolation: "worktree" has no effect for team agents — agent runs in main repo
- 2.1.178: native team-management tools TeamCreate / TeamDelete no longer surfaced (regression from 2.1.177)
- advisor() tool inflates reported input tokens by forwarding full transcript, triggering premature auto-compaction on extended context models
- Agent Teams: subagent permission request crashes the Bun process (2.1.111)
- [BUG] Rate limiter not functioning correctly "API Error: Rate limit reached" when spawning more that one agent
- PreToolUse/PostToolUse hooks do not fire for subagent (Agent tool) tool calls
- [BUG] SendMessage tool missing/unregistered in desktop app engine 2.1.260 (cross-session messaging broken)
- Subagents prompt for permission on every tool call starting v2.1.56
- Background agents drop queued SendMessages and skip completion notifications
- [BUG] Agents view does not work on Linux with Claude 2.1.139
- Team-lead Claude Code crashes with stack overflow in rQ7/permission-explainer after teammate permission request
- [BUG] Workflow harness relays user chat into subagents and demotes the computed task
- [BUG] /deep-research — default workflow frequently hits "API Error: Server is temporarily limiting requests"
- [Bug] Subagent results route to root teammate instead of spawning teammate
- [BUG] Plugin-namespaced agent definitions not injected for teammates (works for subagents and .claude/agents/)
- Parent session deadlocks when subagent tool execution hangs — no timeout or recovery
- You've hit your limit · resets 1:40pm (America/New_York)
- Hooks: PreCompact/PostCompact/SessionStart(compact) fire for a subagent's compaction without agent fields; SubagentStop fires for internal summarizer calls with a never-created agent_transcript_path
- [BUG] Transient server errors MUST NOT kill subagents
- [BUG] Agent team subagents do not honor 1-hour prompt cache TTL — main sessions show 100% 1h, every agent session shows 0% 1h
- Team-spawned agents: agent frontmatter hooks don't fire, MCP servers not inherited, agent_type missing from hook payloads
- [BUG] Every background Agent-tool completion delivers two separate parent-turn events (SubagentHandback + redundant task-notification)
- Auto mode's Bash-preference directive silently overrides CLAUDE.md file-editing rules, and propagates into subagents
- /resume should include background sessions (or add a separate picker for them)
- [BUG] Resuming a fork forfeits its prompt cache: deterministic miss after any tool loop, fork 5m vs parent 1h TTL, full-context rewrites while the prefix is live
- All sessions appear as background agents; subagents stall 100% of the time
- OOM crash under multi-subagent load: subagent CLI processes grow to ~6.5 GB RSS each; CliGovernor logs critical memory pressure for 8 min but evicts nothing (Linux, Desktop 1.17377.0 / CLI 2.1.187)
- 2.1.198: headless -p run exits ~50s after model launches Agent subagent with run_in_background=true; bg wait ceiling not applied (regression from 2.1.197)
- Worker fork / subagent is not woken by the completion of its LAST pending background task
- [BUG] Ink rendering crash "<Box> can't be nested inside <Text>" when teammate requests sandbox bypass permission
- [BUG] TaskOutput deprecation note causes agents to Read full sub-agent conversation history, triggering autocompact thrashing
- [Bug] Sub-agents stuck in infinite loops consuming excessive tokens
- [Bug] Parallel agent spawning causes excessive token consumption before crashing or hitting limits
- [BUG] Sub-agent dispatched via Agent tool stalls silently on MCP permission gate — no surface to parent CLI UI
- [BUG] '--bg' is not enabled. If this is unexpected, retry in a moment.
- Agent-level hooks defined in .claude/agents/*.md frontmatter don't fire when run from CLI
- [BUG] Resuming a background subagent misses its prompt cache: messages_changed, no thinking blocks
- In-process teammate/subagent completion notifications frequently delayed by tens of minutes, require manual nudge
- [BUG] Agent Teams: tmux/auto teammates crash on spawn — non-TTY stdin triggers --print with no prompt (2.1.195)
- Forged `<system-reminder>` markup in subagent output is relayed unsanitized to the parent agent, impersonating a genuine system-reminder
- CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS auto-distributor injects orchestrator/team task descriptions into specialist contexts as fake teammate messages
- [BUG] Subagents / Explore use stale isolated worktree instead of parent repo HEAD
- [Bug] Multi-agent batch launches cause session context/history to disappear from UI
- [BUG] general-purpose subagents don't receive deferred tools (WebSearch, WebFetch)
- Feature request: --agents-file flag to read agents JSON from a file
- [BUG] --append-system-prompt-file blocks /fork while the inline --append-system-prompt doesn't; the refusal never names the blocking flag
- [Feature Request] Add configurable concurrency limiter for parallel agents with lower default
- Workflow harness: memory-aware throttling when fanning out subagents
- [BUG] Resumed background subagent/teammate loses its spawn-time model override and reverts to parent model
- [BUG] Rate limits blocking multi-agent Claude Code workflows even at highest paid tier
- [BUG] Opus 4.7 subagents enter generation loop on post-task finalization/return text (work completes, agent doesn't exit)
- [BUG] Confident hallucinations and gaslighting on user-described systems — getting worse since Opus 4.7
- Sub-agent dispatch denies Bash with tengu_harbor_permissions=true (regression 2026-05-03)
- [BUG] Duplicate compaction subagents spawned (5x identical work, 65% of session quota) — changelog fix incomplete as of v2.1.87
- Bug: Agent selects read-only subagent types for write tasks, then fails to self-correct on rejection
- Worktree isolation: likely branch name collision across sessions causes stale results
- [BUG] SendMessage returns {"success":true} for messages never delivered; a long-lived session instance goes deaf both ways; stale bridge-pointer leaves host "Connected" with 0 workers (2.1.234)
- [Bug] Opus agent invocations charged despite available Fable usage credits
- [BUG] Parent resumes as if subagent completed while subagent is waiting in Monitor for a background process
- [BUG] Subagents stall waiting for MCP servers to be approved when merged into worktrees
- [Bug] Claude Code Orchestrator Mode: Insufficient Output Verification and Spec Definition Rigor
- [BUG] claude -p with active Agent Team: 'shut down before final response' system-reminder fires every assistant turn, persists after team cleanup, induces premature shutdown (regression introduced in 2.1.126)
- Subagent `model:` frontmatter pin silently ignored in 2.1.117
- [BUG] Agent spawning steals tmux focus from the user's pane
- Subagent commits silently delete unrelated files via broad git staging
- [BUG] Conversation data loss: `/export` truncates transcript and `/resume` delivers incomplete context when parallel Agent tool results arrive out-of-order
- [BUG] Agent teams: active pane border color invisible on tmux < 3.3 (tmux 3.2a, Ubuntu 22.04)
- Auto-continue at usage limit never arms when the limit kills only background subagents
- Subagent Bash tool calls are never executed; run ends with terminal_reason "tool_deferred" while reporting success
- Fleet-of-sessions in tmux: Agent tool spawns new pane subagents instead of delegating to existing sessions
- [BUG] Transient server errors MUST NOT kill subagents
- [Feature Request] Add advisor auto mode for autonomous agent operations
- Subagents (Agent tool) cannot reach the LSP tool, contradicting documented MCP/tool inheritance
- [BUG] Agent Teams: tmux teammate pane dies instantly under Nushell login shell (hardcoded POSIX && in spawn command)
- [Bug] BTW agent disappears after main agent completes
- Two concerning behaviors: assistant asserts CLI-tool behavior without verifying, then controls its own accountability review
- [Bug] Experimental CreateTeam: peer agents not inheriting workspace auto-approval for sandbox bash and repo filesystem operations
- Sub-agents dispatched via Task tool have empty MCP tool registry — parent's MCP server connections don't propagate to sub-agent process
- Workflow counts API-errored subagents as "completed"; 429/500/529 storm under multi-agent load
- [FEATURE] TeamDelete: --discard-stale flag to force-cleanup zombie teammates with isActive=true
- [BUG] CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 disables TUI scrollback in v2.1.119 (terminal can't scroll up to view earlier turns)
- [BUG] Agent definition with invalid frontmatter is silently ignored
- [BUG] Parallel subagent worktree cleanup destroys .git directory and entire working tree — catastrophic data loss
- [BUG] Subagents inherit server cwd instead of parent session worktree directory
- [BUG] /clear doesn't clear conversation history for members of agent team
- Claude Opus 4.6 violates explicit SCOPE LOCK directives, makes unauthorized production changes
- [Feature] Agent Teams: Set higher oom_score_adj on teammate agents than on the lead/orchestrator
- Desktop remote (SSH): relaunching the app replaces the running ccd-cli (SIGTERM + --resume) and silently kills in-flight background work
- [BUG] Waking an idle agent fork (`subagent_type: "fork"`) forfeits its inherited prompt cache on every wake — `messages_changed`, `cache_read` pinned to a fixed boundary, not TTL
- SendMessage (cross-session/teammate) silently discards queued messages to a busy session: `queue-operation: remove` instead of `dequeue`
- [Bug] Sibling subagents share ONE scratchpad dir (and a forked Skill inherits the PARENT's), so generic filenames silently overwrite each other
- [Bug] Long-running sessions idle at turn boundaries without auto-continuation or wake signals
- Agent(run_in_background: false) does not block — returns spawn metadata instead of the agent's result (v2.1.227)
- Inline --agents 'tools: []' no longer denies tools in headless -p mode (2.1.214); no flag combination behaviorally denies all
- [BUG] disallowedTools not inherited by subagents spawned via Agent tool
- [BUG] Agent fan-out pays ~47K uncached startup tokens per small task, causing multi-million-token usage
- [BUG] Subagent SendMessage-resume 400s: deferred tool (ReportFindings) referenced in transcript not rehydrated into tools[] on resume
- Fable-5 subagent refuses academic peer-review of consciousness / AI-architecture manuscripts (appears to be an unannounced content classifier)