github alexei-led/cc-thingz v1.6.2

latest releases: v4.9.5, v4.9.4, v4.9.3...
one month ago

v1.6.2 — Context Engineering Review Fixes

Addresses all findings from the reviewing-cc-config audit. Sharpens skill routing, right-sizes agent models per Anthropic model cards, and eliminates false-positive blocking errors.

Fixes

  • test-runner.sh: Exit 0 (informational) instead of exit 2 (blocking) when no test framework is found. Prevents spurious "Claude must fix" errors for unknown project types, missing pytest, or missing cargo.
  • smart-explore: Added context: fork — file-reading fallback (Read/Grep/Glob) now runs in isolated context instead of polluting the main window.
  • documenting-code: Phase 5 now independently verifies changes via git diff --stat rather than trusting the docs-keeper agent's self-report.
  • reviewing-cc-config: Slimmed from 447→237 lines (−47%). Agent prompts reference RUBRIC.md rule IDs instead of inlining full descriptions. Added cross-check step to verify agent findings against actual files. RUBRIC.md is now required (no silent fallback).
  • exploring-repos: Removed unused Read/Grep/Glob from allowed-tools — the skill only uses DeepWiki, Context7, Perplexity, and GitHub CLI.

Model Routing (per Anthropic system cards)

  • opus→sonnet: go-impl, go-qa, py-impl, py-qa, ts-impl, ts-qa — checklist-based review agents don't need frontier reasoning. Estimated 3–5× cost reduction per code review.
  • haiku→sonnet: go-docs, py-docs, ts-docs, web-docs — doc quality review requires semantic code understanding that haiku under-delivers on.

Skill Routing & Disambiguation

  • looking-up-docs: Added trigger phrases ("look up docs", "how to use", "API for", "syntax for", "examples of"). Explicitly excludes comparisons → researching-web.
  • researching-web: Explicitly excludes API/library doc lookups → looking-up-docs.
  • writing-web: Added file-type triggers (.html, .css, .htmx). Explicitly excludes React/Vue/Angular → writing-typescript.
  • reviewing-code: Explicitly excludes config/setup/skills/agents/hooks prompts → reviewing-cc-config.
  • skill-enforcer.sh: Added negative patterns for 3 overlapping pairs to prevent mis-routing. All disambiguation tests pass.

Stats

  • 27 files changed across 9 plugins
  • 208 lines added, 584 removed (net −376 lines)
  • All codex variants updated in sync

Don't miss a new cc-thingz release

NewReleases is sending notifications on new releases.