Skills issues on Windows
Issues the maintainers labelled both platform:windows and area:skills.
212 issues · 52 open · 160 resolved (75%) · first seen Jan 5, 2026
Is this getting better or worse?
This class of problem is holding steady. 66 new reports in the last 90 days vs 84 in the 90 before — -21%. The open backlog peaked at 52 in 2026-10 and sits at 52 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
119 of these reports name the Claude Code build they were running, spanning 88 releases. Heaviest reporters:
- Claude Code v2.1.150
- Claude Code v2.1.126
- Claude Code v2.1.92
- Claude Code v2.1.119
- Claude Code v2.1.123
- Claude Code v2.1.108
- Claude Code v2.1.96
- Claude Code v2.1.266
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 33 days across 160 closures. Of the 160 closures with a recorded reason, 15% were closed as completed and 136 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
67 of these issues post a workaround someone says worked. The highest-engagement ones:
- [Feature Request] Support the standard .agents/skills/ directory for agent skills
- …ex, Cursor, ...), with no vendor-specific copies. - Current workarounds (sync tools, symlinks into ) duplicate content or rely on hacks, and create extra work: drift, conflicts, duplicate search results. Proposed behavio…
Found in the description of #16345 · still open - [BUG] Cowork /schedule skill fails — create_scheduled_task tool not injected into session context (Windows)
- Preflight Checklist - [x] I have searched [existing issues]( and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) -
Found in the description of #29022 · still open - [BUG] /claude-api bundled skill saturates context unconditionally — ~77% spike from neutral question
- …has the right file structure — it just needs a router. ▎ ▎ Workaround: A user command at ~/.claude/commands/claude-api.md overrides the bundled skill and implements on-demand loading. Effective but fragile — requires ma…
Found in the description of #63566 · resolved - Skill tool call returns 'Tool result missing due to internal error' and agent gets stuck
- …aming hangs indefinitely - 26224 — General hanging/freezing Workaround User must Ctrl+C and manually re-invoke the skill. The CLAUDE.md directive to "immediately retry on internal error" is not followed because the agent…
Found in the description of #43866 · resolved - [BUG] Invoking a skill (slash command) pins a large injected block and shrinks the assistant response into a tiny scroll area (VS Code extension)
- …. - So there's no userland (hooks/settings/skill-authoring) workaround — this needs a fix in the extension's message renderer itself: apply the same "Show more" collapse to skill-injected content that already applies to…
Found in the thread of #72590 · resolved
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).
- [Feature Request] Support the standard .agents/skills/ directory for agent skills
- [BUG] Cowork /schedule skill fails — create_scheduled_task tool not injected into session context (Windows)
- [BUG] /claude-api bundled skill saturates context unconditionally — ~77% spike from neutral question
- VS Code extension skill validator rejects valid frontmatter fields
- Skill tool call returns 'Tool result missing due to internal error' and agent gets stuck
- /buddy command removed — was told pet would persist
- [BUG] Invoking a skill (slash command) pins a large injected block and shrinks the assistant response into a tiny scroll area (VS Code extension)
- [Bug] Skill descriptions truncated due to context budget constraints
- [BUG] Slash command menu no longer triggers mid-input in the desktop app (regression; #44447 covered the terminal)
- [BUG] Skill subagent inherits parent's 1M-context tier but not extra-usage entitlement
- [BUG] Skills/commands with `disable-model-invocation: true` are hidden from agent, blocking user-initiated slash invocation
- [BUG] Nested <subdir>/.claude/skills and subdirectory CLAUDE.md never load in auto mode: trigger fires only on Read/Edit/Write, not Bash or Grep (nested-directory variant of #90004)
- Code-tab Customize -> Skills is missing skills that are present in ~/.claude/skills (non-deterministic; / menu and engine load them all)
- Skill invocation fails with '1M context extra usage required' even when already on 1M model
- [BUG] Custom skills from ~/.claude/skills/ not appearing in "/" autocomplete dropdown (Desktop app, Windows)
- [BUG] Cowork: skill upload truncates files >99 KB and leaves null-byte tail on re-upload
- [BUG] Windows Desktop app 1.37937 (2026-08-25 update): slash autocomplete blank, then first-position-only / single-command, chip styling gone
- [BUG] Skills are not loaded from `additionalDirectories`, but are from `--add-dir`
- [BUG] Skill invocation prompt is blocking the output.
- [BUG] Skills in ~/.claude/skills/ not discovered by Desktop Remote (SSH) sessions, even after full restart + new session
- Skill auto-triggering: recall=0% in headless mode (claude -p) regardless of description content
- [BUG] Cowork slash commands return "Unknown skill" — affects custom org plugin skills AND built-in /feedback
- 3 free ultrareview runs consumed but crashed due to server-side rate limit — no findings returned
- [BUG] /ultrareview: three back-to-back runs produced no usable findings across different diff shapes (Dedupe crash → silent empty → silent empty)
- [BUG] Ultrareview crashes repeatedly on large PRs – no findings produced
- [SYS-032] Ultrareview crash kills Remote Control session — cascading failure, no isolation
- [Bug] Claude Code not following skill instructions
- [BUG]/insights generates empty report — no AI insights produced (facets directory never created)
- Slash-command menu returns zero matches when query contains `.` (regression 2.1.160 → 2.1.161)
- Plugin skills from custom/private marketplaces appear in slash menu but fail with "Unknown skill" on invocation
- [MODEL] Claude Code stopped delegating to custom subagents and loading skills — regression persists in Opus 4.6
- [BUG] `context: fork` and subagents appear to use a different (lighter) model despite model configuration
- [BUG] skill-creator: run_loop.py crashes on Windows with UnicodeEncodeError on HTML report
- [BUG] Skill and agent frontmatter hooks silently not firing (CLI, Windows)
- Feature request: background: true in custom command/skill frontmatter
- [BUG] Stacked skills: only the first is treated as user-invoked, so a disable-model-invocation skill in position 2+ is refused
- Skill with unquoted ": " in SKILL.md description is silently dropped from the skills list / slash menu (CLI, no diagnostic)
- Skills added mid-session (e.g. via `git switch` to a branch that adds `.claude/skills/<name>/`) are registered for the Skill tool but never appear in the slash-command menu; `/reload-skills` reports "no changes" and does not fix it
- Skill slash-command intermittently unresolved at session start, works in a fresh session
- [BUG] MCP servers intermittently fail to connect when invoking skills via claude -p (headless mode)
- [BUG] paths: frontmatter never triggers skill auto-loading when Claude reads/edits a matching file
- Multiple server-side flags incorrectly restricting Pro user features: auto mode, skills dashboard, permission friction
- C:/Program Files/Git/ultrareview crashed before producing findings — burned a usage
- [Bug] Agent fails to discover project-local skills when resolving slash commands
- [BUG] Skills with `context: fork` fail with "disable-model-invocation" when invoked via Skill tool (v2.1.113)
- Claude ignores installed MCP skills, writes wrong code 3x instead of verifying first (Shopify)
- [BUG] /simplify skill does not specify subagent_type, first agent launch always fails
- Slash commands should be case-insensitive (e.g. /TODO should work like /todo)
- Self-modification guard prompts for .claude/skills/ files despite documented exemption
- [BUG] Plugin skills not directly loaded when invoked via slash command — spawns unnecessary Explore agents
- [BUG] Flat ``.md-files`` in ``/.claude/skills/`` silently ignored — no error, confusing given "merged" messaging
- User message lost when slash command fails with 'Unknown skill'
- [BUG] Parent skill halts entirely after nested skill completes — remaining steps never execute
- [BUG] Cowork (Windows): "Save" greyed out on every skill proposal card, account-wide, including brand-new skills
- Custom slash commands with non-ASCII (Japanese) names cannot be sent from the composer
- [BUG] docx/pptx/xlsx skills: soffice.py raises "module 'socket' has no attribute 'AF_UNIX'" on Windows (re-filing #78841, closed as stale)
- Plugin skills with disable-model-invocation: true are hidden from the invocable skills list
- `deep-research` skill permanently disappears from autocomplete after a session hits usage limit mid-run (all other built-in skills unaffected)
- Slash command/skill with `context: fork` returns result as `<local-command-stdout>` instead of a visible assistant message (output never renders)
- Desktop app omits Windows junction/symlink skills from the / command menu (agent loader resolves them fine)
- [BUG] Cowork: skill scripts silently truncated on disk at a token boundary, passing py_compile and becoming canonical
- [BUG] /init silently misses source files in repos with many files due to Glob truncation
- [BUG] effort: frontmatter silently ignored in commands/*.md slash commands (works in skills/SKILL.md)
- [BUG] Skills upload returns "Failed to fetch" on Claude Desktop
- skill-creator run_eval.py: Windows incompatibility — select.select() fails on pipe handles (WinError 10038)
- [BUG] Review crashed before producing findings
- Skills from marketplace plugins not activated after installation (missing copy to ~/.claude/skills/)
- [BUG] Skill tool's args parameter drops the first token before $1/$2 substitution
- Used up 1 Ultrareview credit, but claude crashed and threw error API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
- Plugin slash-command autocomplete drops namespace prefix on accept
- [BUG] Slash command autocomplete shows only ~5 built-in commands, hides all 78 custom skills (regression from #22020)
- Custom slash commands from .claude/commands/ not populating / autocomplete menu (v2.1.108)
- [BUG] /insights report makes misleading productivity judgments from incomplete local-only data
- [Bug] Permission allow rules not applied for Bash(mkdir) in skill contexts and default mode
- [BUG] Skills and hooks not available in worktrees created by `claude -w`
- .claude/commands/ files deleted within seconds of session start (v2.1.81)
- [BUG] Why is the `context: fork` skill ineffective?
- skill-creator: run_loop.py fails on Windows with socket error in claude -p subprocess
- [BUG] /clear built-in command being intercepted as a skill lookup
- [BUG] skill-creator crashes on Windows due to non-UTF-8 terminal encoding (GBK, Shift-JIS, etc.)
- Skill tool fails with EEXIST on Windows when username contains non-ASCII characters
- [Desktop 2.2553.1.0] Slash picker opens only at position 0, even after an accepted /skill
- Skill catalog descriptions intermittently missing from system prompt (frontmatter on disk is correct)
- ScheduleWakeup / self-paced loop never fires - the session only wakes when the user types
- [BUG] disable-model-invocation: true still hides skills from model entirely (regression from #43660)
- [BUG] Claude for Windows (desktop): built-in /create-pr (GitHub) runs instead of project custom command; terminal & VS Code work correctly
- [BUG] the connector-query API the skill uses is returning empty when connectors are present.
- System prompt advertises uninstallable Cowork skills as "available"
- Ultrareview crashes before producing findings (2 consecutive failures)
- [BUG] v2.1.140 VS Code extension on Windows: project skills missing from slash-command autocomplete (#9710 / #32331 cluster unfixed)
- [BUG] unable to use ultrareview
- [BUG] /ultrareview crashes at Dedupe step after successful Verify — findings unrecoverable
- [Bug] /ultrareview crashes mid-stream on a normal-sized PR (no findings)
- [BUG] `/ultrareview <PR#>` fails when origin remote uses an SSH host alias for multi-account setups
- [Bug] Anthropic API Error: Dedup stage crash on /ultrareview with no findings and misleading billing
- [BUG] Open Claude Desktop and Bit Defender tells me malware has entered my system download-skill_[0126S7qbJMFG8MgX2vmyeRHX-1777218280219.zip
- [Bug] `/buddy` command disappears or becomes unavailable
- Auto Mode auto-disables after Plan Mode cycle (triggered by plan-first skills); documented workaround keys appear ineffective
- My project skills don't auto-load as commands in Claude Code
- Bug: anthropic-skills:docx hangs silently on Windows 11 (Write permission denied)