Skills issues on macOS — page 3
Issues the maintainers labelled both platform:macos and area:skills.
562 issues · 104 open · 458 resolved (81%) · first seen Feb 23, 2026
Is this getting better or worse?
This class of problem is converging. 140 new reports in the last 90 days vs 248 in the 90 before — -44%. The open backlog peaked at 117 in 2026-04 and sits at 104 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
390 of these reports name the Claude Code build they were running, spanning 150 releases. Heaviest reporters:
- Claude Code v2.1.119
- Claude Code v2.1.87
- Claude Code v2.1.112
- Claude Code v2.1.63
- Claude Code v2.1.150
- Claude Code v2.1.207
- Claude Code v2.1.92
- Claude Code v2.1.76
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 458 closures. Of the 458 closures with a recorded reason, 23% were closed as completed and 353 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
132 of these issues post a workaround someone says worked. The highest-engagement ones:
- [Bug] /loop slash command not recognized in v2.1.71
- …d on a recurring interval (e.g. /loop 5m check the deploy)" Workaround The underlying CronCreate tool works correctly and achieves the same functionality: - CronCreate with cron expression /5 and a prompt successfully sc…
Found in the description of #31759 · resolved - AskUserQuestion auto-submits without waiting for user input in skill/command context
- …t was never collected until they review the final artifact. Workaround implemented Added defensive instructions to skill workflow prompts: after each AskUserQuestion call, verify the result contains actual option labels.…
Found in the thread of #29674 · resolved - [BUG] Skill `model:` frontmatter triggers 429 "Extra usage required" on 2.1.76 when session uses opus[1m]
- …Sonnet promoted to . Rejected if account lacks entitlement. Workarounds - Stay on 2.1.75 — skill model switching works correctly without auto-promotion. Caveat: if a sonnet skill (200K) is invoked from an opus[1m] sessio…
Found in the description of #34296 · still open - [BUG] permissions.additionalDirectories does not load skills from added directories (--add-dir CLI flag does)
- …ilable-skills list. Only bundled skills ( , , etc.) appear. Workaround: Restored — all 22 skills load fine via the personal-scope path. Confirming this is identical symptom to the Windows repro in 43267 (auto-closed as d…
Found in the thread of #30064 · resolved - Regression: built-in `/code-review` rejects invocation from another skill (disable-model-invocation), breaking workflow composition
- …estore this without changing the safer default. Our current workaround is to pause the workflow and hand the developer an explicit manual step, then resume with the output folded in — but that reintroduces exactly the ma…
Found in the thread of #79560 · 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 201–300
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 562 matches; the statistics above cover all 562.
- [BUG] /ultrareview only checks uncommitted files, not full branch history
- [Bug] Permission bypass not working for .claude/skill directory creation
- [BUG] Getting Please run /login · API Error: 403 on skill load when using with LiteLLM
- [BUG] LLM is trying to invoke a non-exist tool: \"Skill\"
- Marketplace plugin source.path not used as root for skill discovery
- Third-party marketplace plugins cannot register skills/commands/agents
- [BUG] Unknown skill when CLAUDE_CONFIG_DIR is set and skill lives in $CLAUDE_CONFIG_DIR/skills/
- [BUG] User-level skills from ~/.claude/skills/ appear in Desktop Code tab autocomplete but don't invoke
- [BUG] Skill triggers ignored even with non-huge context
- Slash commands for plugin skills return 'Unknown skill' with custom CLAUDE_CONFIG_DIR
- /voice command triggers plugin skills instead of activating voice mode
- [Bug] Ralph-loop: Backticks in quoted prompts passed to Bash causing command substitution error
- [BUG] Claude Code no longer load skills invoked by slash commands
- Project skills silently dropped beyond ~28 limit — no warning or error
- Skill tool aborts with "Sibling tool call errored" when a parallel tool call fails
- Plugin --plugin-dir namespacing breaks skill allowed-tools resolution
- [BUG] Skill invocations persist across /clear and context compaction
- [BUG] Desktop: new session's slash menu omits anthropic-skills:* until the first message (temporary query not given the skills plugin)
- [BUG] /doctor prompt-audit: the documented "nothing changes until you ask" guarantee is enforced only by prompt text, and the method has no source-of-truth step
- [BUG] No disk-sourced skills load in 2.1.271 — user skills and enabled-plugin skills both absent; only bundled skills register
- Invoked skill not re-attached after manual /compact; only the summary's paraphrase survives
- [BUG] Clean boot sends skill catalog in both Skill tool schema and skill_listing message
- [BUG]Scheduled task stalls indefinitely on Write permission prompt despite matching allow rule (settings.json correct)
- /reload-plugins re-injects full skill roster without evicting the session-start copy (+~7.5k tokens per reload)
- Headless -p "/skill" with a failing !-preamble silently aborts pre-turn and reports subtype:success, num_turns:0
- Crashed ultrareview (cloud code review) still consumed a free ultrareview credit
- Bug: skill name + MCP server name collision causes silent server drop (no error, no warning)
- [BUG] Skill loader selects older plugin cache despite installed_plugins pointing to current version
- [BUG] Frontmatter-preloaded agent skills re-injected in full on every SendMessage resume (linear duplicate context growth)
- [Bug] Duplicate skill entries in /context output with narrow terminal and long skills list
- Skill `model` frontmatter field does not trigger automatic model switching
- [Bug] skill-creator run_loop.py burns ~50% of a usage window in minutes: full claude -p agent per query x run x iteration, no cost preview or cap
- [Bug] Skill listing truncated with descriptions dropped due to context budget constraints
- [BUG] claude-api skill trigger self-fires on /model banner (bare model names + ANSI [1m as trigger tokens)
- [Feature Request] Per-skill permission configuration in settings.json
- [BUG] personal skills not propagating to remote web sessions
- [Bug] Ultrareview consumed free credit on crash with zero findings
- [BUG/FEATURE] LLM agents fabricate non-existent `disabledSkills` settings field — silent no-op needs validation or the feature itself
- [BUG/UX] Custom user command silently shadows built-in alias with no warning at creation or invocation time
- [BUG] Desktop app (macOS): pasted slash command is silently swallowed — neither executed nor sent as text
- /ultrareview: billed cloud action triggered by default Return key with no cost confirmation
- [BUG] Claude code asks to read skill prompt files
- [Bug] Ultrareview crashes before producing findings, incorrectly consuming free attempt quota
- [BUG] Slash command invocation blocked by context-limit check on Opus 4.7 1M (false positive)
- [BUG] Dynamic /loop persists across session compaction without user re-invocation
- [BUG] Cowork serves stale cached content every session — CLAUDE.md never refreshes, skills revert to old version after install
- [BUG] Model ignores explicit project skills/rules even when memory references them
- ultrareview crashed before producing findings + rate limit error — 2nd occurrence same account, all 3/3 free quota now consumed
- [Bug] Anthropic API Error: Rate limited during ultrareview completion
- [Bug] /ultrareview command continues executing after cancellation and ignores spend limits
- Skill-loader returns "Launching skill" but body fails to load — 3 plugin skills with template-heavy SKILL.md + Agent dispatch references
- [BUG] /ultrareview completes cloud-side but CLI reports 'failed' and /tasks does not list the run
- /ultrareview crashed with empty findings — quota refund inquiry
- Skill picker should sort skills alphabetically
- Built-in /simplify skill fails: agent type 'code-simplifier' not found when code-simplifier plugin is installed
- [Bug] Compact/resume replays stale slash-command ARGUMENTS as current turn
- Crash: "H.toolUseContext.getAppState is not a function" in lN permission renderer when teammate invokes project-scoped skill
- [BUG] /commit-commands:commit doesn't work when there are no commits yet
- $0 / $ARGUMENTS template variables not substituted in skills with context: fork
- [Bug] Custom skills not resolved to worktree directory after EnterWorktree
- [BUG] Skills Missing after Claude for Mac update
- Allow skills to disable or customize @-mention autocomplete in arguments
- [BUG] /effort session command does not override effortLevel in settings.json
- /skills dialog uses toggle-like indicators with no toggle affordance
- Personal plugin upload: skills not registered in runtime
- [Feature Request] Restore /buddy command functionality
- Project skills in .claude/skills/ not discovered when running in a worktree
- [BUG] Skills silently fail in --print mode with no error or warning
- [BUG] Here's the issue ready to copy: Title: Skill model field not propagated to Agent sub-invocations when disable-model-invocation: true
- Plugin skills from skills/ directory invisible in slash command autocomplete, blocking adoption
- Slash command autocomplete missing skills from certain plugin namespaces
- Local personal skills don't suppress anthropic-skills: duplicates in skill list
- v2.1.89 regression: Project slash commands from .claude/commands/ no longer appear
- [Bug] Skill tool steps dropped during execution, skipping to manual gh commands
- [BUG] Issue: /init skill doesn't discover AGENTS.md
- [BUG] Specify model for insights command
- feat: Skip dynamic context injection inside markdown code blocks in SKILL.md
- [BUG] : /workflows:* commands fail with "Rate limit reached" at low usage
- Slash command autocomplete should search against group/namespace tags, not just skill display name
- Skill named "batch" silently blocked with misleading disable-model-invocation error
- fork: true ignores model override in command frontmatter (rate limit with Opus 1M default)
- Skill !command preload permissions never persist - rule format mismatch
- UX: allowed-tools message implies restriction but field only pre-authorizes listed tools
- [Bug] Skill frontmatter "model" field causes 404 API error
- Custom slash commands from ~/.claude/commands/ not recognized
- [Sonnet 4.6] AskUserQuestion returns without capturing user input during skill execution (intermittent)
- [Bug] User-defined skills not recognized in CLI and missing from /skills listing
- context: fork skill exits prematurely after nested Skill tool invocation
- Plan mode tool definitions silently override operator skill file guardrails
- [BUG] Slash command autocomplete truncates and randomly samples skills when total count exceeds display limit
- [BUG] Skill frontmatter model:/effort: ignored when Claude invokes the skill via the Skill tool (applied on /skill-name)
- Single code-review at medium effort using Fable 5.1 blows through entire session budget at $100/month subscription level without completing
- [Bug] /code-review --comment creates separate reviews instead of single review with summary
- [BUG] scheduled-task sub-agents self-invoke a skill and spawn duplicate agents, multiplying token spend
- [Bug] claude-api skill context usage balloons to hundreds of thousands of tokens on basic queries
- skill-creator: run_eval.py writes skills to .claude/commands/, so description optimization always measures recall=0%
- [Bug] Ultrareview results not delivered to local session and failed runs consume quota
- Skill tool substitutes invocation args into literal "$0" text inside SKILL.md
- [BUG] Dynamic /loop (ScheduleWakeup) swallows the assistant's replies to the user — before AND after the tool call; only stopping the loop restores replies
- [BUG] Claude Code Violates Settings