v2.1.206

2.1.206

(☝️ Click to see full diff.)

+10,807 tokens

• NEW: Skill: Artifact dashboard; Skill: Artifact data table; Skill: Artifact explainer; and Skill: Artifact report — Add template-driven workflows for publishing operational dashboards, sortable/filterable data tables, visual concept walkthroughs, and long-form reports, including placeholder/data validation, light/dark-safe styling, and creation-versus-editing guidance.
• NEW: Skill: Code Review correctness finder angles; Skill: Code Review inline medium/high template; Skill: Code Review inline xhigh mode; and Skill: Code Review low effort expanded-findings mode — Add effort-scaled inline review workflows that prioritize recall, inspect changed functions and cross-file behavior, audit removed safeguards, include cleanup and convention checks at higher effort, deduplicate without verifier passes, and enforce bounded finding targets.
• NEW: Skill: PR explainer artifact-template mode — Makes PR explainers publish shareable HTML walkthroughs from the explainer template, organized around motivation, reviewer-facing before/after behavior, related code changes, non-obvious context, and review focus areas.
• NEW: Tool Description: EndConversation and System Reminder: End conversation background fork no-op — Add a conversation-ending tool restricted to sustained abuse after repeated redirection and an explicit warning, or a user-requested demonstration after confirmation; forbid its use for task failure, frustration, ordinary completion, harmful-content refusals, or self-harm/violence cases, and clarify that it has no effect in background forks.
• REMOVED: System Prompt: Proactive schedule offer after natural future follow-up and System Prompt: Strict proactive schedule offer gate — Remove the prompts that encouraged or gated proactive /schedule follow-up offers after completed work.
• Agent Prompt: /code-review part 9 fix application — Simplifies findings-tool follow-up after --fix, removing the explicit requirement to resubmit every finding with fixed, no_change_needed, or skipped outcomes while retaining explanations for skipped findings.
• Agent Prompt: Quick PR creation — Changes push guidance from always using origin to using the repository's configured remote, with origin treated as the usual default.
• Skill: Auto mode setup — Reworks setup around mechanically pre-gathered, untrusted local recon; uses the gathered repo, settings, sensitive-path, CLI-frequency, and denial data instead of repeating broad scans, adds recovery for failed settings gathering, respects CLAUDE_CONFIG_DIR, and preserves the existing approval and merge safeguards.
• Skill: /doctor slash command and Skill: /doctor slash command description — Add a checked-in CLAUDE.md trimming pass that removes codebase-derivable layouts, stack lists, standard commands, copied schemas, generic advice, and mechanically enforced rules while preserving gotchas, rationale, non-standard conventions, safety directives, and other non-derivable guidance; also refine version checks for npm/bun, native, and Homebrew installs with channel-aware, project-isolated lookups.
• Tool Description: ClaudeDesign — Prefer the live shared Claude Design canvas for co-editable presentations, decks, prototypes, demos, posters, and other visual artifacts unless the user requests local files or names another destination.