[MEDIUM] Auto-Update: Claude Code v2.1.118

[github-actions]

Auto-Update: Claude Code v2.1.118

Release: v2.1.118 (2026-04-23T00:42:21Z)
Relevance: MEDIUM - Review suggested
Release URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.118

Summary

Three releases spanning v2.1.116–v2.1.118 deliver a significant hooks expansion (MCP tool invocation, bug fixes for prompt re-firing and agent-type hook failures), task-directory retention changes, and performance/UX polish.

Why This PR Exists

All Claude Code releases get a PR for visibility. The relevance level helps you prioritize:

• HIGH: Contains changes that directly affect SDLC enforcement
• MEDIUM: Contains useful improvements worth reviewing
• LOW: Routine update, merge at your convenience

Relevance Analysis

{
  "version": "v2.1.118",
  "relevance": "MEDIUM",
  "summary": "Three releases spanning v2.1.116–v2.1.118 deliver a significant hooks expansion (MCP tool invocation, bug fixes for prompt re-firing and agent-type hook failures), task-directory retention changes, and performance/UX polish.",
  "wizard_impact": [
    {
      "area": "hooks",
      "description": "v2.1.118: Hooks can now invoke MCP tools directly via `type: \"mcp_tool\"` in settings.json — a new hook type alongside the existing `type: \"command\"`. Our three hooks (sdlc-prompt-check, tdd-pretool-check, instructions-loaded-check) all use `type: \"command\"` and continue to work, but this opens the door to richer hook logic without shell scripts.",
      "suggested_action": "No immediate change needed (KISS). Document the new hook type in ROADMAP or CLAUDE.md as a future upgrade path — e.g., a UserPromptSubmit hook could call an MCP validation tool instead of spawning a shell process. Revisit if shell hooks become unreliable or MCP tooling matures.",
      "files_affected": [".claude/settings.json"]
    },
    {
      "area": "hooks",
      "description": "v2.1.118: Fixed `prompt` hooks re-firing on tool calls made by an agent-hook verifier subagent. Our `sdlc-prompt-check.sh` fires on `UserPromptSubmit` — this bug could have caused it to double-fire whenever an agent-hook spawned a verifier subagent that itself made tool calls, injecting duplicate SDLC BASELINE text into context.",
      "suggested_action": "Reliability improvement only — no wizard file changes needed. Worth testing post-upgrade to confirm the hook fires exactly once per user prompt, especially in agentic/subagent sessions.",
      "files_affected": [".claude/hooks/sdlc-prompt-check.sh"]
    },
    {
      "area": "hooks",
      "description": "v2.1.118: Fixed agent-type hooks failing with 'Messages are required for agent hooks' when configured for events other than `Stop` or `SubagentStop`. v2.1.116: Agent frontmatter `hooks:` now fire when running as a main-thread agent via `--agent`. Both changes improve hook reliability across agent execution modes.",
      "suggested_action": "No code changes needed. Awareness item: if any future wizard hooks target agent-lifecycle events (Stop, SubagentStop), they will now behave correctly. Existing UserPromptSubmit / PreToolUse / InstructionsLoaded hooks are unaffected.",
      "files_affected": []
    },
    {
      "area": "tasks",
      "description": "v2.1.117: The `cleanupPeriodDays` retention sweep now also covers `~/.claude/tasks/`. The SDLC skill uses `TodoWrite` (stored under `~/.claude/tasks/`) for tracking SDLC checklist items — long-lived tasks could now be pruned if `cleanupPeriodDays` is set.",
      "suggested_action": "Verify whether the project or user settings define `cleanupPeriodDays`. If set to an aggressive value (e.g., 7 days), in-progress SDLC checklists from paused long-running features could disappear. Consider documenting a recommended minimum (e.g., 30 days) in SDLC.md or CLAUDE.md.",
      "files_affected": [".claude/skills/sdlc/SKILL.md"]
    }
  ],
  "plugin_check": {
    "new_official_plugins": [],
    "replaces_custom": []
  },
  "reasoning": "The new `type: \"mcp_tool\"` hook capability and the fix for prompt hooks re-firing directly touch the wizard's primary enforcement mechanism (UserPromptSubmit hooks), warranting MEDIUM rather than LOW. However, no custom wizard code needs to be replaced or rewritten — the hook bug fixes are platform-level improvements that make existing hooks more reliable, and the new MCP hook type is an optional future upgrade path. HIGH would require an official feature actively replacing custom code or a breaking change to existing hook behavior."
}

Review Checklist

• Aligns with KISS principle
• Actually needed for SDLC enforcement
• No over-engineering
• If replacing custom code, removal is clean

---

Generated by weekly-update workflow. You always decide what merges.