Claude Code 2.1.147 adds Workflow tool and `/code-review` effort levels

• ClaudeCodeLog's 2.1.147 release post says Claude Code 2.1.147 shipped 35 CLI changes, led by a new Workflow tool for deterministic multi-agent orchestration that stays off unless CLAUDE_CODE_WORKFLOWS=1 is set.
• In ClaudeCodeLog's 2.1.147 changelog post, /simplify becomes /code-review, the command now supports effort levels such as /code-review high, and --comment can post findings as inline GitHub PR comments.
• According to the 2.1.147 changelog post, Anthropic also hardened the REPL and Workflow sandboxes against prototype-pollution and thenable-based escapes.
• ClaudeCodeLog's 2.1.148 changelog post says 2.1.148 landed less than seven hours later to fix a regression from 2.1.147 where the Bash tool returned exit code 127 for every command for some users.
• ClaudeCodeLog's additional 2.1.147 updates and ClaudeCodeLog's additional 2.1.148 updates show prompt tokens continuing to climb across both releases, while the tool docs were updated to push grep harder for search tasks.

You can check Anthropic's official changelog entry for 2.1.147, the one-line 2.1.148 hotfix entry, and Marc Krenn's tracked 2.1.147 diff bundle. The weirdly fast follow-up is the story here: ClaudeCodeLog's metadata post clocked 2.1.147 at 1 day 14 hours after 2.1.146, then the 2.1.148 metadata post logged the next release just 6 hours 53 minutes later.

The new item in 2.1.147 is the Workflow tool. Anthropic describes it as deterministic multi-agent orchestration, and the feature is gated behind CLAUDE_CODE_WORKFLOWS=1 in both the release post and the official changelog entry.

That wording matters because Claude Code had already exposed the CLAUDE_CODE_WORKFLOWS environment variable in 2.1.146, where ClaudeCodeLog's 2.1.146 additional changes post listed it on the CLI surface before the actual Workflow tool appeared. This week turned that dormant flag into a named feature.

The /simplify rename started in 2.1.146, where ClaudeCodeLog's 2.1.146 release post said /code-review would add optional effort levels. In 2.1.147, ClaudeCodeLog's changelog post made the behavior more specific: /code-review now reports correctness bugs at a chosen effort level, --comment can write inline GitHub PR comments, and the old cleanup-and-fix behavior is gone.

That is a real product shift, not just a command rename. mattlam_ complained that /simplify was the more useful built-in skill for cleanup work, while the new command is aimed more squarely at review depth and PR workflows.

2.1.147 bundled a long fix list around automation and session management. The highest-signal items in the changelog post were the sandbox hardening for REPL and Workflow, restored enforcement for enterprise login restrictions on third-party-provider and API-key sessions, and a fix for auto mode suppressing AskUserQuestion when a user or skill explicitly relied on it.

Then 2.1.148 arrived as a hotfix. According to ClaudeCodeLog's 2.1.148 changelog post, 2.1.147 had introduced a regression where the Bash tool returned exit code 127 on every command for some users, which would break any automation that trusted exit status instead of output text.

The side-channel change across these releases is prompt and tool expansion. ClaudeCodeLog's 2.1.147 metadata post says prompt tokens rose by 4,954, up 6.6 percent, and tool tokens went from 65.9 percent to 69.4 percent of the mix. The 2.1.148 metadata post shows another 6,315 prompt tokens, up 7.9 percent, with tool tokens reaching 71.7 percent.

Those metadata posts also catch small but telling surface changes:

• The 2.1.147 metadata post lists a new --config option plus CLAUDE_BG_MEMORY_TOGGLED_OFF and CLAUDE_BRIDGE_REATTACH_OUTBOUND_ONLY.
• The 2.1.148 metadata post says the tool documentation now explicitly recommends always using grep for search tasks.
• Marc Krenn's tracked repo exposes the underlying prompt diffs, including the 2.1.147 CLI surface changes and the 2.1.148 grep wording change.