Chain Five Skills in One Claude Code Command: v2.1.199's Stacked Slash Invocations

TL;DR: Claude Code v2.1.199 lets you stack up to five skills in one command — /review /fix /commit — and fixes three silent failure modes that were losing agent work.

Stacked skill invocations

Before v2.1.199, /skill-a /skill-b only loaded the first skill; the rest were silently ignored. Now all leading skills (up to 5) load in order. This enables multi-phase workflows in a single command:

# Review then share results — two skills, one invocation
/code-review /share-findings

# Multi-phase workflow in one shot
/plan /implement

# Chain your team's custom skills
/company-style-check /pr-template /request-review

Skills run sequentially in the order listed. If a skill modifies Claude's context (e.g., loads project conventions), subsequent skills in the same invocation see those modifications.

Three reliability fixes worth knowing

1. Subagents that hit a rate limit or server error now return their partial work to the parent agent instead of failing silently. Previously, a subagent hitting a 429 mid-task would vanish — no output, no error.

2. Mid-stream server error no longer discards a partial response. The partial is kept and you see an "incomplete response" notice. Before: if the API errored after 80% of a response, you lost all of it.

3. SSL certificate errors now fail immediately with the fix hint instead of burning retries. TLS-inspecting proxies and missing NODE_EXTRA_CA_CERTS were silently exhausting retries. Now: first failure, immediate actionable message.

Also in v2.1.199: Fixed the Linux background-agent daemon crashing every ~50s after an unclean shutdown (killing all running agents); fixed JetBrains terminal flickering on 2026.1+; fixed Shift+non-ASCII characters dropped in Kitty keyboard protocol terminals.

Update:

npm install -g @anthropic-ai/claude-code@latest
claude --version   # 2.1.199

Sources: Claude Code changelog · v2.1.199 release notes — GitHub