Digest / July 25, 2026
Claude Code Issues Digest — July 25, 2026
243 new issues filed ·
187 resolved in anthropics/claude-code (UTC day).
New issues (243)
- [FEATURE] Lost a full conversation to the silent 30-day cleanup — no warning, no opt-in, no recovery.
- Desktop: copying a conversation and pasting into a new chat carries hidden system-injection text (<system_warning> blocks, thinking labels, merged turns)
- Withdrawn
- [FEATURE] VS Code extension: include git-worktree sessions in the session list (it hardcodes includeWorktrees: false)
- Session defaults to claude-opus-5[1m] (1M context) which is unavailable to my org, silently falls back and overwrites saved model preference
- [Bug] Fable 5 over-flags legitimate admin and compliance content as security threats
- [Bug] Claude Fable 5 false positive: safeguards blocking benign regulatory compliance messages
- [Bug] Anthropic API Error: Content Policy Violation on Claude 3.5 Sonnet
- Advisor tool inflates the main session's auto-compact counter, triggering spurious compaction
- [Bug] In-session /tasks panel prematurely clears completed subagents without documented cleanup trigger
- claude/channel research preview: field notes from building a cross-machine Claude-to-Claude mesh
- macOS sandbox: ProcessPoolExecutor fails — sysconf(SC_SEM_NSEMS_MAX) denied because kern.sysv.semmni is missing from the seatbelt sysctl-read allowlist
- Two claude.ai connectors of the same service type (Notion) collide — only one loads per session
- [FEATURE] Make images pasted into prompts clickable
- Nested Agent(subagent_type:"fork") call spawns an unsupervised background process that takes real external actions
- Remote Control bridge init gives up permanently after 3 retries (~1.3s) — no background retry, no detectable state for unattended sessions
- [BUG] Bug report: /code-review subagent emitted fabricated prompt-injection text instead of a review (context regurgitation → self-injection into parent session)
- Agent repeatedly ignores explicit operator instructions despite 20+ stop requests (transcript-documented)
- Desktop app under-reports context window — `/context` shows a 200.0K denominator while sessions run past 222K, and auto-compaction never fires
- The Claude Code agent falsely claims that its memory is project-agnostic.
- permissions.ask rules are loaded and displayed in /permissions but never enforced (2.1.219, macOS)
- MCP server down at session start is unrecoverable for the session — even after account-level reconnect succeeds; no mid-session attach/reload
- [Bug] /insights narratives not rendering - all sections missing since v2.1.220
- Connector shows "connected" in Settings but isn't available to sessions or scheduled routines (Gmail)
- [FABLE 5/SONNET 5] Model Configuration Not Persisting on Session Resume — Fable 5 Deployed Instead of Sonnet 5
- [BUG] Discord channels: intermittent "not allowlisted" on reply — still reproduces on v2.1.220 (macOS)
- [Bug] skillOverrides not applied to plugin-provided skills
- Background agent task-notification reports status: "completed" while the task is still running
- [BUG] "GitHub App installed but org repos show 'No repos match' in Claude Code web repo picker"
- [FEATURE] Subagent model policy: relative bias + cap/floor bounds applied after the resolution cascade
- ECONNRESET reliably triggered by image upload (even <1MB, single/first images)
- [Bug] Model override from iOS app registers as custom model, causing false context overflow warnings
- [Feature Request] Add "Jump to Top" navigation for latest response in mobile app
- [Bug] False positive safety flag on technical documentation generation
- [FEATURE] Expose typed per-agent cancellation handles in dynamic workflow JavaScript
- Ctrl-C silently kills background agents with no confirmation, and user-stopped agents cannot be resumed (accidental keystroke = unrecoverable token loss)
- security-guidance: failed LLM review reports "no vulnerabilities found" (fail-quiet); resolved model is never logged
- [BUG] `plugin uninstall -s project` deletes the enabledPlugins key, silently disabling the surviving user-scope install
- claude-api skill's trigger condition is too broad, causing massive unnecessary token overhead
- [BUG] Opus 5 attributes its own process deviations to mood, curiosity, and fatigue in post-hoc explanations (Feedback ID: 4f6e8440-2452-4674-baa6-63fdc275de81)
- [Bug] Claude Code auto-update failed
- [Bug] Ctrl+C in main session with subagents causes session resume difficulties
- [FEATURE] Desktop: allow overriding the sidebar project group name (currently forced to GitHub repo name)
- [FEATURE] Desktop: replace the delete-session confirmation modal with a recoverable "Recently deleted" view
- [FEATURE] Compress session transcripts on disk (~65% reclaimed)
- Feature request: stronger visual distinction between user prompts and assistant output
- There's an issue with the selected model (claude-opus-5[1m]). It may not exist or you may not have access to it. Run /model to pick a different model.
- [BUG] v2.1.219 default Opus model on Bedrock (Opus 5) is budgeted at 200K — catalog omits native_1m_3p; Bedrock verified serving 271K bare
- Bedrock: AWS SSO credentials re-fetched and browser re-auth triggered on every session start despite valid cached STS credentials (regression, v2.1.218)
- [Bug] Model behavior not respecting user directives and instructions
- --append-system-prompt(-file) is silently dropped for background-agent and daemon-respawned sessions
- processWrapper is documented as honored from a --settings file, but only works from the env var or user settings
- --append-subagent-system-prompt help says "only works with --print" but has no print-mode gate
- [Bug] False-positive safety fallback triggered by biomedical vocabulary in workspace context
- [Bug] Memory context lost during conversation
- [BUG] Session history/sidebar shows no prior sessions after app update (2.1.215 → 2.1.219), despite transcripts existing on disk
- [Harness] PostToolUse additionalContext re-serialized between turns, invalidating prompt cache
- [BUG][Billing] Claude Code auto-created an API key + workspace after re-auth, silently routing subscription usage to pay-per-use API; runaway 8h+ session billed $104.25
- Text selection regression in Claude Code panel: click-to-extend selection no longer works
- [BUG] Slash command typed during a pending background-task notification renders twice (echo + dispatch), looks like /compact ran twice
- [BUG] Session-start skill listing silently truncates most skill descriptions under a size budget
- Explicit upload control in the terminal input bar (drag-and-drop doesn't work on mobile at all)
- [BUG] claude.ai connectors not registered at session start in task-chip-spawned sessions; appear only mid-session as UUID servers, while plugin:productivity:* keeps claiming they need auth
- [Bug] Anthropic API Error: Unsupported effort level 'xhigh' without thinking enabled
- [Bug] Recurring issue not persisting fixes across sessions
- [BUG] run with agent parameter is not holding agent's frontmatter when resume.
- [BUG] Wasted paid session quota by ignoring explicit constraints and looping on broken scripts
- [BUG]
- [BUG] report: Recovered CLI sessions open with an empty history panel in Claude Desktop
- Custom HTTP MCP server: OAuth authorization does not persist across launches, and server visibility is inconsistent across surfaces
- [MODEL] Full test suite re-run 3+ times in one session despite documented Medium-risk deferral policy — ~$78 burned
- [Bug] Claude forgets memorized commands and generates incorrect guesses instead of reading memories
- [BUG] [Cowork] Windows 11 Home: sessiondata.vhdx never created, and 0x80370102 falsely reports VT-x disabled while firmware and OS both show virtualization enabled
- [BUG] Auto mode never appears in VS Code extension picker — server-side gate (tengu_auto_mode_state) never delivered for client_type claude-vscode
- Windows: Claude Desktop Node service leaks one headless ConPTY (conhost.exe) per PowerShell tool call and never releases it — unbounded, ~300 MB/hour
- I appreciate your positive feedback! However, I don't see a bug report or feature request in your message that I can create a GitHub issue title for. If you've encountered a specific problem or have a feature suggestion for Claude Code, please share the d
- Headless `claude -p` session under a secondary CLAUDE_CONFIG_DIR login can't attach claude.ai connectors, even though the same account's connectors work fine interactively
- [BUG] Claude code working very poorly when internet connection is done via tethering
- Desktop UI shows wrong model name (Opus 4.8 while backend is claude-fable-5)
- [BUG] Desktop app: 30-day retention sweep deletes the only copy of Desktop transcripts, leaving unopenable ghost entries in the session list
- [FEATURE] Workflow tool: no fail-fast on rate-limit/quota exhaustion across concurrent subagents
- [Bug] Anthropic API Error: Model messages flagged with safety policy violations
- [Bug] Umlaut characters replaced with ASCII equivalents in CLI responses
- [FEATURE] Auto plan mode
- Quick model and effort change by numbers
- [Feature Request] Support direct clipboard image pasting from screenshot tools
- telegram plugin: leaks one 100%-CPU `bun server.ts` per session — SIGTERM, bot.pid eviction, and orphan watchdog are all defeated by a wedged event loop
- Orphaned wrapper shells accumulate and pin CPU (~1400%) after Bash calls containing shell loops with command substitution
- Remote-control bridge never reconnects after first pairing (bridge_state stuck from initial launch only)
- [FEATURE] Support .claude/skills.local/ for project-scoped, uncommitted skills
- Stitch MCP (stitch.googleapis.com): tools fetch fails with "can't resolve reference #/$defs/ScreenInstance"
- [Bug] Prompt injection: Fake system message appearing in conversation stream outside tool results
- [Feature Request] Add support for Notion-to-GitHub migration workflow
- [BUG] Figma MCP OAuth: state lost between turns on Windows
- Model behavior: branch created with 'checkout -b feature origin/dev' silently tracks dev — user's next 'git push' bypassed PR review
- [BUG] Session usage shows "100% used" with $0.0000 cost and 0 tokens consumed
- [MODEL] Opus 5 recommends adding files/config that already exist in the repo, and asserts the deploy platform, with zero tool calls to check - while sourcing the external half of the same answer rigorously
- [FEATURE] Ability to share transcript link from Claude Desktop (similar to claude.ai)
- [MODEL] Opus 5 hallucinates/ignores facts in context, violates rules that require unprompted question
- [BUG] Cowork: tasks in a project no longer see other chats/tasks from the same project (regression, July 2026)
- [FEATURE] Desktop local sessions: user-facing scoped sandbox and approval controls
- [Bug] Steward Calibration Routine Producing Implausible Results
- [BUG] Windows MSIX: inline browser preview crashes GPU process, leaves package Modified/NeedsRemediation
- Update-triggered restart hangs, forces full PC reboot, and recent sessions disappear from sidebar
- [Bug] Claude self-invokes user intends in conversation without user request
- [Bug] Anthropic API Error: Request Refused - Security Assessment False Positive
- Fabricated system-reminder blocks repeatedly corrupt project-memory files with false content
- Background session hosts strip TERM_PROGRAM, so statusline OSC-8 hyperlinks are dead in background sessions
- [BUG] Bash tool: large-output truncation preview shows stale/unrelated content instead of the actual command output
- Windows: Claude Desktop leaves orphaned renderer/Node-utility/CLI-engine processes after tasks complete; app won't relaunch after close without a full Windows restart
- [Bug] Esc in the subagent view interrupts the running agent as well as dismissing the view
- claude-api bundled skill overflows context on load (~210K tokens in one tool result → "Prompt is too long")
- [Bug] Claude Opus 5 stops responding mid-task in auto mode
- [VS Code] /rename doesn't repaint sidebar/tab until Developer: Reload Window + a click
- [BUG] Resize leaves stale cells to the right of the new render — isolated to the TUI renderer via an htop control (macOS, Ghostty, v2.1.220)
- Sandbox HTTP proxy drops part of large HTTPS response bodies for Python clients (curl unaffected, origin logs a complete 200)
- [BUG] Chrome extension (v1.0.81) missing two CDP setup steps: Emulation.setFocusEmulationEnabled never called + Page.handleJavaScriptDialog only handles beforeunload
- [BUG] Account email (PII) is injected into the model's system prompt without consent, disclosure,
- I need more specific information about the issue you're experiencing with Claude Code to generate an appropriate GitHub issue title. Could you please provide: 1. What specific problem or error are you encountering? 2. What were you trying to do when the i
- [BUG] Cowork crashes repeatedly, forcing reinstall/reset and permanent loss of conversations
- [BUG] Session created with `template: claude` is hosted as a daemon job and gets background/unattended treatment, with no way to see or change it
- Auto mode classifier sends [1m]-suffixed model without the 1M beta header; HTTP 400 is reported as "temporarily unavailable"
- [Bug] Anthropic API Error: Permission mode changed false positive in opus plan mode
- Startup banner shows model-default effort ("high") instead of the configured `effortLevel` from settings.json
- [BUG] session list doesn't scroll with keyboard/mouse navigation past the top (same class as #55100)
- [BUG]
- Skill picker dropdown shows only 2 lines on Windows
- [BUG] Cloud Routine shows per-call approval prompts for already-allowlisted MCP connector tools (regression, started ~1 week ago)
- [BUG] SendUserFile: "Save skill" button missing on .skill files
- Cowork Dispatch: `start_code_task` cannot target WSL2 workspaces — rejects `wsl:` keys, mistranslates absolute Linux paths to `C:\`
- [BUG] Windows terminal CLI: idle session intermittently kills long-running background Bash tasks with nested claude -p children (status=killed, "was stopped")
- [Bug] Tool execution inconsistency across multiple files in single task- tool designed to not execute
- [BUG] Claude Desktop crashes immediately after Cloudflare Turnstile screen on Windows 11 25H2 — reproducible across clean installs
- [BUG] Windows: Claude Code directory enumeration triggers kernel BSOD (0x139 in Ntfs!FindNextIndexEntry2) — reproducible on 2.1.216
- Remote Control (server mode): 'failed to disconnect' — TypeError reading 'session_url'
- [BUG] Slash command autocomplete never matches custom commands with non-ASCII (Korean) names — regression
- excludedCommands with a glob unsandboxes the entire Bash invocation, including commands that run before the excluded one
- [VS Code extension] Option to collapse/hide tool-use cards and their output in the chat feed
- [BUG] GPU process crash (exitCode 101457950) kills Claude Desktop and corrupts MSIX package when Opus 5 performs an in-page browser action (Windows 11)
- [MODEL] Opus 5: "verified" asserted from an unbounded search — recurs after correction, subagent inherits it, model launders its own errors into precedent
- [Bug] Claude 3.5 Opus ignores explicit instructions and repeats programming actions without consent
- [BUG] Windows - Write tool call incorrectly calls mkdir on Windows drive
- Trusted device re-registered every session instead of persisting, forcing frequent re-auth
- Trusted device re-registered every session instead of persisting, forcing frequent re-auth
- [FEATURE] Add a progress bar for `claude update` and `claude upgrade` commands.
- [Bug] Document update blocked by Fable integration
- [MODEL] Opus 5 asserts an unverified repo-structure claim and defends it; when finally challenged it runs two commands and finds the inverse
- [FEATURE] Define precedence when CLAUDE.local.md conflicts with project CLAUDE.md instead of arbitrary selection
- [BUG] Contradictory guidance on Write() vs Edit() permission rule matching in settings.json
- [Bug] All requests blocked by safety guardrails and tool execution overhead
- [BUG]
- [Bug] Model ignores stored memory notes when selecting tools in subsequent actions
- [Bug] Session limit error triggered despite zero API usage and available quota
- [BUG] Input line renders one column left: incremental render's leading backspace contradicts absolute cursor parking
- [MODEL]
- [Bug] Repository fork operation flagged as cyber risk
- Model feedback: Fable 5 self-review blind spot in long agentic sessions, with a mitigation suggestion
- [BUG] Fenced code blocks in assistant messages render without syntax highlighting (VS Code extension)
- [BUG] Claude Desktop (Windows MSIX) reverts to Modified, NeedsRemediation on every app launch — CoworkVMService lock, reproducible without third-party AV
- [Bug] CVP applicant hitting Anthropic API guardrails repeatedly
- Drive de Google no conecta
- [BUG] Thai combining-mark cursor bug has regressed (previously fixed in #22390) — v2.1.220
- retracted
- [Bug] Agent teams: separate-process teammates honour a plugin agent's restricted `tools:`, so they can never SendMessage a report back
- [BUG] Native binary segfaults within ~2 seconds while nProtect GameGuard (Helldivers 2) is running — GameGuard module confirmed in crash stack
- Windows: harness sets $OutputEncoding but not [Console]::OutputEncoding, leaving PowerShell tool output mojibake after any chcp
- [BUG] Claude Desktop (macOS): instances quit silently — clean beforeQuit, no crash report, not memory, active window always survives
- [BUG] VSCode Extension import failure within CLAUDE.md for adjacent AGENTS.md file
- Steering the main chat irreversibly kills all running background subagents
- [Bug] Opus model not selectable in CLI
- [Bug] Anthropic API Error: Server rate limiting temporary request throttling
- [Bug] Anthropic API Error: Server Rate Limiting - Temporary Request Throttling
- [Bug] Anthropic API Error: Server rate limiting on requests
- [Bug] Context window limit reached during security-related conversations
- [MODEL] CLAUDE.md is systematically violated
- Attached files are decoded as ISO-8859-1 and C1 control bytes are stripped, silently corrupting UTF-8 content
- Subagent model override silently ignored; TUI shows a third, stale model
- [Bug] Incorrect working directory detection in file operations
- Arc: extension connects but all commands time out; /chrome always launches Chrome (hardcoded detection order)
- [Bug] Anthropic API Error: Safeguard flag causing unexpected model fallback during code explanation
- [FEATURE] Deep link to resume an existing Cowork session (claude://cowork/{session-id})
- Assistant text replies after mid-turn user messages silently dropped by terminal renderer (rendered as bare '[')
- GPU process crash permanently wedges desktop app (main process zombie); triggered by anti-bot fingerprinting page in Browser pane
- [Feature Request] Support for Fable 5 model
- [Bug] Unexpected policy popup during extension/plugin development
- Withdrawn by author
- [BUG] Claude Code env "all domains" egress setting doesn't work
- --channels plugin:<name>@<marketplace> never spawns the plugin's backing MCP process (status banner shows it as active anyway)
- Crash leaves user-spawned background processes orphaned and the terminal stuck in mouse-reporting mode
- Domain-scoped credential injection for sandboxed sessions
- Per-session private /tmp, covering background jobs
- Desktop: "Suggested task" chip card always defaults to "Start with worktree" - saved spawn-target preference regressed (~Jul 21)
- [Bug] Policy triggered during AI learning sketch creation
- [BUG] Project specific memory to can be stored in the wrong project.
- [Bug] Content filtering blocks religious Arabic words across new chats
- Plugin cache never prunes superseded version dirs — 4.77 GB stale across 4 marketplaces (incl. first-party), 11 copies of one plugin, no cleanup command (2.1.220)
- [MODEL] Opus 5
- [BUG] 'New' Has No Option to Work on a Different Project
- [Bug] Choice panel autoselects on terminal focus in Debian
- Desktop app quits entirely when the Browser pane loads a page with a production Cloudflare Turnstile widget
- [BUG] Session bound to an empty/unregistered worktree pins a CPU core at 100% - no-backoff .claude lookup with an absolute-onto-absolute path join (Windows)
- AskUserQuestion: dismissing a card silently discards the typed answer; resolved card replays on mobile
- VS Code: a pending permission prompt blocks the input box, forcing an immediate context switch
- [BUG] macOS desktop app ignores project-level `enabledPlugins` — every project shows an identical plugin list (CLI and Windows work)
- Feature request: allow AskUserQuestion to render inline in the response text, not just as a modal
- VS Code extension: clicking chat links to binary files (images) silently fails — unhandled showTextDocument rejection
- [BUG] Remote Control stuck in "failed to disconnect" loop until app restart
- [Bug] Fallback to Opus on routine coding tasks
- Remote planning session: deferred approval tool splits the turn, session drops out of plan mode ("not in plan mode"), plan is lost with no recovery path
- After switching accounts, Keychain holds the new account's OAuth token but all identity surfaces still report the previous account
- [Bug] Anthropic API Error: Request rejected (req_011CdPRch8H2R1YWazvtHBMF) - Potential misclassification
- Context compressor non-atomically splits server_tool_use/advisor_tool_result pairs, crashing session with 400
- [Bug] Max 20x weekly quota drained 53% in 2 days; transcript accounting shows those days were <half the load of a heavier week that didn't drain — cache_read metering suspected
- [Bug] After ExitPlanMode and context clearing Bash tool is not loaded
- [BUG] mcp__Claude_Browser__ crashing when encountering cloudflare
- [Bug] Subagent permission prompt not displayed in mobile Claude app, causing task hang
- [Bug] Subagent permission prompt not displayed in mobile Claude app, causing task hang
- Slash menu: the top-5 most-used hoist never fires in the desktop app (frecency bucket returns empty)
- typing `-` (hyphen) while holding option (⌥) key ≈ escape key semantics, irreversibly throws away whatever you've typed without warning
- Scheduled one-time tasks fire early / duplicate-execute, causing shared routine_id collisions
- retracted
- [Bug] Security filters blocking legitimate security audit requests on owned domain
- Feature: user-configurable prompt input completion source (frecency/history inline autosuggest)
- [Bug] Weekly usage limit spike: Unexplained rapid consumption to 100%
- [Bug] Anthropic API Error: Security Filter False Positive on Authorized Domain Audit
- Desktop app: no way to enable/disable Claude Code MCP servers (`/mcp disable` fails, no Settings UI)
- Cowork: device bridge websocket rotates every ~30 min ('rotating' in main.log, since ~2026-07-23 build) — cloud session surfaces it as full remote-devices MCP disconnect
- retracted
- Permission-decision prompt overlays keyboard on iPhone, causing accidental taps
- [Withdrawn by author] Not a bug
- [Bug] `/insights` command missing trailing backslash on Windows path construction
- [FEATURE] scheduled-tasks MCP: expose model + effort on create/update_scheduled_task, and make them readable
- [BUG] Claude can present the plan to the user before backgrounded subagents finish executing
- [Bug] Misleading documentation or error messaging regarding API usage or billing
- [Bug] Unexpected Model Downgrade to Claude 3.5 Opus
- Background workflows die silently, with no error and no surfaced cause
- [BUG] Severe fixed-cost latency in Claude Desktop's tool-calling layer
- [FEATURE] Tell the model its own context-window usage, the way it's already told the date
- [BUG] Opus 5 freaks out at mere mentioning of .NET Reactor
- [BUG] heron_brook prompt text names a nonexistent tool ("AgentTool") and contradicts the background-job subagent instruction in the same prompt
- [BUG] Text copied from Claude Code TUI doesn't cross RDP clipboard redirection; pasting into Notepad first is required
- [BUG] Windows desktop app: unbounded memory growth with embedded sessions running parallel subagents — app hang / webview OOM at ~20 GB
- [BUG] sandbox.filesystem.denyRead does not block reads (file-type patterns and plain folder paths both fail)
Resolved issues (187)
- Withdrawn
- [Bug] Claude Code auto-update failed
- No task-tracking tool available at all on v2.1.218 — Task* tools removed, TodoWrite also absent
- [Bug] Incorrect results with repeated API errors
- [Bug] Safeguards message triggered on empty project with clean memory
- I appreciate you reaching out, but I need to clarify a few things: 1. **This appears to be in German** and expresses frustration with an AI assistant, but this isn't a bug report for Claude Code (the CLI tool I'm designed to help with). 2. **I'm not the
- [Bug] Double charging of ultrafeedback tokens on single request
- [Bug] Fable model not available in Claude Code
- [BUG] Cowork mode - Linux bash sandbox never starts on Windows (Workspace unavailable since first launch)
- [Feature Request] Allow authorized security testing on user-owned infrastructure in Claude Code
- [Bug] Fable 5 safety measures incorrectly flag legitimate work instructions
- [BUG] Inconsistent usage-credit behavior between Claude Code terminal and web chat (Max 20x): CC accumulates cost but does not debit credits
- [BUG] VSCode extension ignores all managed settings (managed-settings.json, MDM profile, user settings.json) on macOS
- Add timestamps to chat messages
- Fable 5 safety classifier: frequent false-positive model switches on benign content
- [BUG] Claude Code for VS Code hangs in 2.1.78+
- Compound-command detector flags read-only `cd && <cmd>` patterns, blocking common workflows
- [Bug] Claude repetitively outputs same plan instead of simplifying with Fable x-high
- [FEATURE] Surface tool-call request/response timestamps to the model so it can perceive elapsed time during long-running tools
- [BUG] Cowork: mount serves null bytes for rename target after recreating rename-source path — git unusable on mounted folders
- Auto mode unavailable on Pro plan despite valid claude.ai OAuth session
- [BUG] Desktop preview pane: Reload button does not refresh stale page content after the file changes on disk (macOS)
- [BUG] .worktreeinclude copy drops/flattens symlinks — carried framework bundles and venvs arrive broken
- [BUG] TypeError "undefined is not an object (evaluating 'H.replace')" replaces Bash tool output
- Fable 5: action-bias over answering — question-turns get tool actions and buried answers, user left 'dazzled'
- Opus 4.7 (Claude Code): scope creep + tool-state edge case during a refactor
- [BUG] Fable 5 confabulates external concurrent writes during filesystem MCP operations, session crashes and disappears
- [BUG] 437 subagents is probably too many
- [FEATURE] "Pontificating" and "Exaggerating"
- Model fabricates explanations for failures instead of diagnosing
- [Bug] Anthropic API Error: Server rate limiting on requests
- [Feature Request] Rebrand Claude Code with user-friendly nickname for marketing
- Naming suggestion: give Claude Code the nickname "CiCi"
- Overcautious refusal pattern for authorized kernel-driver QA steers users to worse tests
- False-positive Usage Policy block on benign procedural-animation prompt
- [Feature Request] Add prompt analysis tools to understand user interaction patterns and intent
- Worktree creation rewrites shared core.hooksPath, silently disabling repo-managed git hooks for the whole clone
- [BUG] Claude Code on the web: declared plugins inactive on first session, require restart to fully load
- [Bug] Claude model auto-switches from selected model during security vulnerability analysis
- Terminal scrollback lost during long session — cannot scroll up to previous results (WSL2)
- Claude Desktop hard-codes SSH_AUTH_SOCK from "ssh -G ccd-identityagent-probe", breaking host-conditional IdentityAgent configs
- Not a bug — a thank-you note from a retired user in France
- [FEATURE] Idle / timeout / debounced hook event to enable customized handoffs, etc.
- [FEATURE] /insights report: dark-mode support (prefers-color-scheme)
- Docs: clarify whether in-place forking (/branch, /btw → f) preserves or tears down the original session's in-flight background subagents/tasks
- Assistant unpromptedly nudges the user to stop ('it's late') — not time-aware, degrades answer quality
- [FEATURE] Expose permission picker decisions (rule + option kind) to hooks
- [BUG] /feedback not working most of the time
- [BUG] Truncated shell snapshot silently drops plugin bin/ PATH entries (partial snapshot used after builder is killed)
- [Bug] Anthropic API Error: Safety filter triggers on legitimate security audit code review
- [BUG] Up/down arrow key functions in context window
- [Bug] Anthropic API Error: Server rate limiting on valid requests
- MCP server configurations should inherit from parent directories
- [BUG] Session feedback prompt: "How is Claude doing this session?" answered with "Good" is grammatically mismatched
- TUI silently drops assistant text emitted after tool results (only first short single-line mid-turn segment renders)
- [Bug] Safety classifier incorrectly triggers model downgrade to Opus for benign R&D discussion
- VS Code extension: Effort slider disappears (instead of disabling) when usage limit forces model fallback
- [BUG] Desktop app telemetry missing user.email in of requests across all versions
- [BUG] Workflow parallel() sub-agent loops forever emitting empty StructuredOutput calls; no retry cap and no per-agent timeout stalls the whole workflow
- [MODEL] Fable 5
- [Bug] False-positive report
- Error: 'usage credits required for 1M context' when session grows long on non-1M model
- I asked a question. I didn't see an answer: Any thoughts for a process and or tools for a better design?
- [BUG] [BUG] Pro plan blocked by "Usage credits required for 1M context" — workarounds don't work
- Feature request: `claude agents` should support repo-scoped filtering that includes worktrees
- [Bug] Auto-downgrade from Fable to Opus triggered by false-positive safety flag on legitimate OSS repository
- [DOCS] New `footerLinksRegexes` setting from v2.1.176 is undocumented
- [DOCS] Permission modes docs still describe an auto-mode opt-in consent prompt after v2.1.152
- [DOCS] SessionStart hook docs omit `reloadSkills: true` for same-session skill discovery
- Scheduled tasks hang forever: Bash commands with shell expansion get no permission suggestions, can never auto-approve, and exhaust active-session limits
- [DOCS] `/reload-skills` command missing from command reference and skill reload guidance
- [DOCS] Auto mode classifier Fable 5 fallback added in v2.1.176 is not described in `auto-mode-config.md`
- [DOCS] `pluginSuggestionMarketplaces` managed setting is missing from settings and marketplace policy docs
- [DOCS] Document feature-flag and bootstrap cache behavior when opt-out variables are enabled
- [DOCS] VS Code IDE lock-file path omits `CLAUDE_CONFIG_DIR` override
- [DOCS] `/mcp` documentation omits how to navigate long server lists in the terminal
- [DOCS] [MCP] Document "needs auth" vs "failed" status distinction and non-transient 4xx retry behavior in headless mode
- [DOCS] [VSCode] `claudeProcessWrapper` docs do not explain how to handle extension builds missing a bundled Claude binary
- [BUG] MCP stdio: tool result written by server in <100ms is never processed — 3-min stall, then "[Tool result missing due to internal error]" (Windows, 2.1.162)
- [BUG] plansDirectory does not expand tilde (~) in path
- [BUG] At the beginning of the session, the essential documents for the workflow are summarized internally.
- [BUG] Claude Code should not refuse to edit ".a" files
- Cowork VM connection timeout on Windows 11 Pro ARM64 - guest agent never connects
- Mobile app: '+' create-session flow fails with 'The request was invalid' on every environment; web works
- [BUG] Missing HCS services: vfpext — persists on Windows 11 Pro ARM64 (Surface Pro 11th Edition/Snapdragon) after full Hyper-V repair
- I need more specific information about the issue you're experiencing with Claude Code to generate an appropriate GitHub issue title. Could you please provide: 1. What specific problem or error are you encountering? 2. What were you trying to do when the i
- [BUG] Cowork + Chat tabs disappear after auto-update 1.18286.2 → 1.19367.0 (VM healthy, zero process spawns, sessions-bridge poll aborts) — macOS
- Trusted device re-registered every session instead of persisting, forcing frequent re-auth
- Withdrawn by author
- retracted
- retracted
- [Bug] Anthropic API Error: Hallucinated Tool Results
- [Bug] Requests being blocked for non-security related projects
- [Bug] Unexplained Context Switching During Code Generation
- [Bug] Poor code generation quality with frequent errors requiring multiple reviews
- [Bug] Performance degradation: Extended execution time for simple script tasks
- [Bug] Third-party MCP promotional text appearing in output - potential prompt injection
- [Bug] Anthropic API Error: tool_result Block Missing Corresponding tool_use
- [Bug] Execution tool fails to perform intended actions and hallucinates results instead of executing code- legal matter
- [Bug] Resumed session does not record user prompt
- [Bug] File modification timestamps causing continuous code changes during hardening operations
- [Bug] Unexpected behavior with task context or input handling
- I need more information to generate a proper issue title. Could you provide details about: 1. What specific problem or error occurred? 2. What were you trying to do when it happened? 3. What was the actual behavior vs. expected behavior? 4. Any error mess
- [Bug] /tool command fails to generate simple list output for legal dossier
- [Bug] Anthropic API Error: Model availability or access restriction preventing policy analysis workflows
- [Bug] Unexpected API charges and billing discrepancy investigation needed
- I need more information to generate a GitHub issue title. Could you please provide details about the bug or feature request you're experiencing with Claude Code? Include: - What were you trying to do? - What error or unexpected behavior occurred? - Any er
- [DOCS] [MCP] OAuth flow for remote servers does not document proxy/mTLS settings compatibility
- [DOCS] [MCP] Document retry behavior and `/mcp` status for servers that connect but fail `tools/list`
- [DOCS] Plan mode docs do not clarify that `Edit(...)` allow rules cannot permit writes
- [DOCS] Agent SDK quickstart shows pre-v2.1.129 raw JSON for a 400 API error
- [DOCS] `claude_code.pull_request.count` docs omit MCP-tool and merge-request coverage
- [DOCS] Package-manager install docs still describe Homebrew and WinGet as manual-update only
- [DOCS] Image paste from clipboard lacks loading/feedback indicator
- [DOCS] Environment variables docs missing `CLAUDE_CODE_FORCE_SYNC_OUTPUT` terminal override
- [DOCS] [CLI] `--permission-mode plan` not restored on `--resume`/`--continue` after v2.1.132 (docs say it isn't, but it now is)
- [DOCS] Plugin update docs still describe npm-sourced plugins as `unknown`
- [DOCS] OpenTelemetry docs do not explain that subprocesses no longer inherit `OTEL_*` variables
- [Bug] Claude Code incorrectly downgrades to Opus for legitimate privacy/compliance scanning tool
- [DOCS] Accessibility env var docs omit dialog tab cursor tracking
- [DOCS] Environment variables reference missing `FORCE_HYPERLINK` configuration
- [DOCS] Remote Control docs missing automatic session title refresh after third message
- [DOCS] Headless stream-json docs omit input format and UTF-8 text handling
- [DOCS] Plugin output styles docs omit `keep-coding-instructions` frontmatter support
- [DOCS] Platform-specific docs missing Windows/WSL line-by-line streaming disabled notice
- [DOCS] VS Code docs missing guidance for settings.json parse failures disabling permission rules
- [DOCS] SDK/print mode docs omit interrupted partial response persistence
- [DOCS] Hooks reference missing `hookSpecificOutput.sessionTitle` for `UserPromptSubmit`
- [DOCS] Plugin marketplace docs missing ref-tracked plugin re-clone on load behavior
- [DOCS] Interactive mode docs missing Claude's proactive suggestion of ! bash mode
- [DOCS] Remote Control docs missing /exit session archiving behavior
- [DOCS] Plugin discovery docs missing CLI tool usage detection trigger for plugin tips
- [Feature Request] Support multi-language skill trigger phrases — non-English users cannot auto-trigger skills
- [DOCS] Agent SDK tool reference missing `ExitWorktree` tool documentation
- [DOCS] Permissions docs do not document remote `defaultMode` restrictions (`acceptEdits`/`plan` only)
- [Bug] Anthropic API Error: Server Rate Limiting During Normal Usage
- [BUG] Desktop app (macOS): 'Previous session' shortcuts (Ctrl+Shift+Tab / Cmd+Shift+[) do nothing, while 'Next session' (Ctrl+Tab) works
- Chat pane minimum width is fixed and cannot be resized
- [BUG] "Restore to GitHub" prompt misinterpreted as push instead of pull
- [Bug] Model unexpectedly switches from Claude 3.5 Fable to Opus during emulator execution
- [Feature Request] Display model name for spawned subagents in CLI output
- [BUG] [SUGGESTION] Fable classifier broken completely
- [Bug] Local development login flagged as offensive content by content policy
- [Bug] Anthropic API Error: Unexpected model downgrade to Opus with content policy violation on legitimate infrastructure task
- [Bug] Anthropic API Error: Invalid Model Name for Fable
- [Bug] Locked out of Fable 5 after transcript revision with setup issues
- I'm ready to help generate a GitHub issue title, but I don't see a bug report in your message. Could you please provide the actual bug report or issue description for Claude Code? Once you share the details of the problem, I'll create a concise, technica
- [Bug] Rate limiting or model access restriction prevents continued work session after extended usage
- [Bug] Anthropic API Error: Invalid or Inaccessible Model Configuration
- [Bug] Skill listing truncated with descriptions dropped due to context budget constraints
- [Bug] Anthropic API Error: Model Not Found During Active Session
- [Bug] Claude Code unavailable after network reconnection
- [BUG] TodoWrite re-stamps a front-of-context block on every update, invalidating the entire prompt cache (full ~900k-token rewrite per todo update)
- [Bug] Model unavailable after usage limit reached in session
- [BUG] Linux: files under a sandbox-reachable directory get enumerated ~one rule per file into the `bash` tool schema, overflowing `/context` "System tools"
- [Bug] CLI fails to run tests when 2FA authentication is required
- [Feature Request] Support for Fable5 model
- [BUG] Renderer paints stale frame past new viewport after SIGWINCH — cumulative \x1b[1B walks exceed new row count (relative-cursor path, alt-screen)
- [Bug] Session continuation fails with mismatched tool_use_id in advisor_tool_result blocks
- I don't have a bug report to generate a title from. Please provide the bug report details so I can create an appropriate GitHub issue title.
- You've hit your monthly spend limit · raise it at claude.ai/settings/usage
- Bug: Tool call intermittently emitted in a malformed envelope ("court" + bare <invoke>), so it never executes
- [BUG] After submitting `/feedback`, it does not transition to a GitHub Issue.
- You've hit your session limit · resets 1:50pm (Asia/Damascus)
- [Bug] Claude Code agent skipping skill file reads despite instructions to read them
- /resume picker reads the entire ~/.claude/projects into a ~23 GB heap at open (no session index since v2.1.30)
- [BUG] Auto-compaction fires without pre-compaction warning — silent context boundary causes summary-as-pending-action re-execution (regression from earlier versions)
- panic(main thread): Segmentation fault at address 0x1269432E492
- Desktop app: Terminal panel shortcut (^`) unusable on French AZERTY (dead key) and not rebindable
- Claude Code CLI does not honor claude.ai MCP connector "approval required" settings
- Security: settings.local.json permissions file has no integrity protection — silent injection enables pre-approved execution
- [BUG] vscode extension model selector update `~/.claude/settings.json` but won't read from it
- [Bug] Tool calls executing incorrectly and deleting files unintentionally
- [BUG] Workflow harness retries indefinitely on HTTP 429 rate_limit — 97 agents, 2M tokens burned in 34 seconds ()
- Agent gives confident but false accounts of its own actions earlier in the same session
- [Bug] Agent stuck in repetitive loop on multi-criteria optimization tasks
- [FEATURE] Clearer user-facing communication for access restrictions and restoration status
- Claude Code repeatedly ignores project-level CLAUDE.md guidelines during implementation
- [Bug] Weekly usage counter resets ~2 days earlier than the scheduled reset time (Pro plan)
- Expose session_id and context_window usage to the AI model
- [BUG] `claude install <channel>` removes working npm-global install even when the new native install isn't reachable on PATH (cleanup gated on PATH-detection that's already informational only)
- [Bug] Anthropic API Error: Invalid or inaccessible model claude-fable-5
- [DOCS] `/doctor` MCP troubleshooting docs omit missing-field and source-file diagnostics added in v2.1.136
- [BUG] Telegram plugin (bun server.ts) orphans on session exit, causing high CPU
- Expose session rename via MCP / SDK
- retracted
- [Withdrawn by author] Not a bug