alexei-led/cc-thingz v1.9.1 on GitHub

[1.9.1] - 2026-05-03

Added

Architecture-tier linting in smart-lint.sh (dev-workflow): wires knip (unused exports / files / deps) and dependency-cruiser (boundary rules and import cycles) for TypeScript / JavaScript projects. Both opt-in by their own config-file presence (knip.json, .dependency-cruiser.cjs, etc.) — no env var to enable. Falls back through bunx and npx when the binary isn't on PATH; emits a stdout install hint (does not block) when no runner is available.
Layered hook config in smart-lint.sh: sources ~/.claude/.claude-hooks-config.sh (global defaults) then ./.claude-hooks-config.sh (project overrides) so per-project settings override per-user defaults. Per-tool linters keep reading their own project configs (.golangci.yml, pyproject.toml, .prettierrc, knip.json, .dependency-cruiser.cjs).
SKIP_ARCH=1 env var and .nolint-arch marker file to skip just the architecture tier without disabling fast-tier linters. Existing SKIP_LINT=1 and .nolint continue to skip everything.
plugins/dev-workflow/docs/lint-tools.md: new reference doc with install commands (brew / uv / bun) for every tool the hook touches, architecture-tier opt-in instructions, skip recipes, and a pointer to .golangci.yml for Go architecture enforcement (depguard, gomodguard, cyclop, revive).

Fixed

lint_shell() now skips .claude-hooks-config.sh files. They are sourced (not executed) and routinely written without shebangs, so shellcheck's SC2148 was incorrectly blocking edits when a project added per-project hook config.

[1.9.0] - 2026-05-03

[1.9.1] - 2026-05-03

[1.7.1] - 2026-04-19

Changed

reviewing-code skill: replaced unscoped Bash with scoped permissions (Bash(git *), Bash(gh pr *), Bash(gh api *), Bash(rg *), Bash(wc *)); added Read/Grep/Glob/LS/LSP for fallback inspection of user-provided file paths
24 review sub-agents (go-*, py-*, ts-*, web-* × qa/impl/tests/idioms/docs/simplify): scoped unscoped Bash to per-language read-only tooling. Top-level engineers (go-engineer, python-engineer, typescript-engineer, web-engineer) untouched
writing-python skill: expanded "Verify Generated Code" with explicit retry loop (ruff --fix → format → pyright → repeat until green)
testing-e2e skill: expanded Phase 3 with pass criteria, retry steps, and full-suite regression run
8 skills: TodoWrite → TaskCreate / TaskUpdate / TaskList in frontmatter and prose. Per CC spec, TodoWrite is non-interactive/SDK only; interactive sessions use Task*
linting-instructions skill: model opus → sonnet (rule-based regex linting doesn't need Opus reasoning)
looking-up-docs skill: removed dead WebSearch and mcp__perplexity-ask__perplexity_ask (description explicitly excludes general web search)

Fixed

mem-history skill: added context: fork — get_observations returns 500–2k tokens per result and was leaking into the main context

Notes

All 9 plugins bumped to 1.7.1 to align with marketplace tag
PR #6 (yogesh-tessl) closed without merge: the "frontmatter validation fix" was based on a third-party Tessl validator, not the Claude Code spec — which explicitly accepts both YAML lists and space-separated strings for allowed-tools. Useful prose changes (verify loops) cherry-picked manually

[1.9.1] - 2026-05-03

[1.7.0] - 2026-04-17

Added

Private mirror support: scripts/rewrite-mirror.py rewrites plugin manifests (repository URLs, homepage links, plugin names) for private GitHub mirrors — configurable per-mirror name maps in MIRROR_NAMES
Mirror sync workflow: .github/workflows/rewrite-mirror.yml auto-rewrites manifests on push to mirror repos; condition-gated on github.repository to skip the source repo; commits with [skip ci] to prevent trigger loops

Changed

Plugin manifests: All plugin.json and marketplace.json files enriched with full metadata — author.email, author.url, homepage URLs, expanded keywords arrays across all 9 plugins
make push: Simplified to plain dual-push (origin + mirror remotes); CI on mirror repos handles manifest rewrites automatically

[1.9.1] - 2026-05-03

[1.6.3] - 2026-04-16

Added

coding skill: Language-agnostic process discipline for all implementation tasks — surfaces assumptions before coding, defines verifiable success criteria first. Complements writing-go/python/typescript/web with process guardrails. Auto-activates on implement/write/create/build/add/develop intent; wired into go-engineer, python-engineer, typescript-engineer, web-engineer agents.
smart-lint.sh skip gate: Skip auto-linting via SKIP_LINT=1 <command> (transient) or .nolint file in project root (persistent, add to .gitignore). Useful when editing repos you don't own and want to avoid auto-formatting side-effects.

[1.9.1] - 2026-05-03

[1.6.2] - 2026-04-12

Fixed

test-runner.sh: Exit 0 (informational) instead of exit 2 (blocking) when no test framework found — prevents spurious "Claude must fix" errors for unknown project types, missing pytest, or missing cargo
smart-explore: Added context: fork — file-reading fallback now runs in isolated context
documenting-code: Phase 5 independently verifies changes via git diff --stat instead of trusting agent self-report
reviewing-cc-config: Slimmed from 447→237 lines (−47%); agents read RUBRIC.md directly instead of inline rules; added cross-check step; RUBRIC.md now required (no silent fallback)
exploring-repos: Removed unused Read/Grep/Glob from allowed-tools

Changed

Model routing (per Anthropic system cards): go-impl, go-qa, py-impl, py-qa, ts-impl, ts-qa downgraded opus→sonnet (checklist review, 3–5× cost reduction); go-docs, py-docs, ts-docs, web-docs upgraded haiku→sonnet (semantic doc quality)
Skill descriptions: Added trigger phrases and NOT-for exclusions to looking-up-docs, researching-web, writing-web, reviewing-code for cleaner routing
skill-enforcer.sh: Added negative patterns for 3 overlapping pairs (code/config review, docs/research, web/typescript) — all disambiguation tests pass

[1.9.1] - 2026-05-03

[1.6.0] - 2026-04-07

Added

reviewing-cc-config skill: Review Claude Code configuration (skills, agents, hooks, CLAUDE.md, commands) for context efficiency, signal density, and anti-patterns — derived from Anthropic's "Effective Context Engineering for AI Agents" and Claude Code best practices documentation
Co-located RUBRIC.md with 16 review rules across 4 categories: Context Budget (CB-*), Signal Density (SD-*), Architecture (AR-*), Anti-Patterns (AP-*)
Spawns up to 4 parallel review agents per component type with token-capped structured output
Skill-enforcer trigger patterns for "review config", "config review", "context review", "review skills/agents/hooks"
Skill count: 31 → 32 (dev-tools: 14 → 15)

[1.9.1] - 2026-05-03

[1.5.0] - 2026-04-03

New skill: explore public GitHub repositories via DeepWiki AI-generated documentation.

Added

exploring-repos skill: AI-powered exploration of 30,000+ public GitHub repositories via DeepWiki MCP — understand architecture, design patterns, component relationships, and cross-repo comparisons without cloning
Three DeepWiki MCP tools: read_wiki_structure (topic index), read_wiki_contents (full wiki), ask_question (semantic Q&A with multi-repo support)
GitHub CLI (gh) fallback strategy for repos not indexed by DeepWiki — gh repo view, gh api, gh search code work for any public repo
Tiered fallback chain: DeepWiki → GitHub CLI → Context7 → Perplexity → local clone
Clear DeepWiki vs Context7 decision table (architecture understanding vs API references)
Skill-enforcer trigger patterns for "explore repo", "deepwiki", "repo architecture", "how does owner/repo work"

[1.9.1] - 2026-05-03

[1.4.0] - 2026-04-02

AGENTS.md adoption and CC-first rebrand.

Added

AGENTS.md generation: scripts/generate-agents-md.py builds AGENTS.md from flat/skills-codex/ with categorized skill tables — compatible with 20+ AI coding tools via the Linux Foundation AGENTS.md standard
make agents-md and make validate-agents-md targets
AGENTS.md badge in README
Installation section for AGENTS.md-compatible tools (GitHub Copilot, Cursor, Windsurf, Devin)
Tests for AGENTS.md generator (16 tests)

Changed

Rebranded from "cross-platform collection" to "Claude Code plugin suite with portable skill export"
README tagline, badges, section headers, and Platform Support table updated to CC-first framing
"Cross-Platform Architecture" section renamed to "Skill Export Architecture"
Codex/Gemini badges changed from "compatible" to "skill_export"
.claude-plugin/marketplace.json description updated
gemini-extension.json description updated

Fixed

GEMINI.md skill drift: added 6 missing skills (evolving-config, learning-patterns, linting-instructions, mem-history, smart-explore, using-gemini) — now lists all 29 skills

[1.9.1] - 2026-05-03

[1.3.0] - 2026-04-02

Cross-platform plugin support for OpenAI Codex CLI and Google Gemini CLI.

Added

Codex CLI support: .codex-plugin/plugin.json manifests for all 9 plugins, .agents/plugins/marketplace.json Codex marketplace
Gemini CLI support: gemini-extension.json at repo root and per-plugin, GEMINI.md context file, skills/ symlink for Gemini skill discovery
Platform-aware skill overlay system (scripts/generate-overlays.py):
- Three-tier skill classification: GREEN (15 shared), YELLOW (8 auto-stripped), RED (6 hand-authored)
- CC-specific frontmatter stripping for all non-Claude platforms
-  body sentinel markers for section stripping
- Platform preamble injection with agentic anchors for o3/codex-1 and Gemini models
6 hand-crafted SKILL.codex.md overlays optimized for o3/codex-1 instruction following: reviewing-code, fixing-code, improving-tests, brainstorming-ideas, deploying-infra, testing-e2e
skills-codex/ build output directories in all 9 plugins (29 platform-optimized skills)
flat/skills-codex/ symlinks for cross-tool access
make overlays and make validate-overlays targets
Codex CLI and Gemini CLI badges in README
Installation guides for all 3 platforms in README
Multi-platform manifest templates in CONTRIBUTING

Fixed

smart-lint.sh: skip symlinks in markdown formatting (prettier errors on symlinked .md files)
CI: run overlay build before validation to handle cross-Python serialization
CC plugin versions synced from 1.2.0 to 1.2.2

Changed

All 9 .codex-plugin/plugin.json skills pointer: ./skills/ → ./skills-codex/
Codex and Gemini get platform-optimized skills instead of CC source
README structure diagram expanded to show dual-manifest layout
CONTRIBUTING directory structure shows all 3 platform manifests

[1.9.1] - 2026-05-03

[1.2.2] - 2026-04-01

Documentation accuracy fixes for README.

Fixed

Agent model table: engineers were listed as opus but are actually sonnet
Add 4 missing skills to user-invocable table (analyzing-usage, learning-patterns, linting-instructions, using-git-worktrees)
Move learning-patterns and using-git-worktrees from auto-activated to user-invocable
Add playwright-skill to auto-activated table, pdf-parser to agents table
Update dev-tools skill count: 13 → 14
Narrow linting-instructions enforcer triggers to skill/agent authoring context
Clarify linting-instructions description: references Anthropic model cards

[1.9.1] - 2026-05-03

[1.2.1] - 2026-04-01

System card-derived instruction hardening for all agents and skills.

Added

linting-instructions skill: model-based prompt review against system card rules
scripts/lint-instructions.py: advisory regex linter with uv inline deps
docs/instruction-lint-rules.md: 12 rules (6 universal, 3 opus, 3 sonnet) with citations
make lint-instructions target (advisory, doesn't fail CI)
Skill-enforcer trigger for linting-instructions

Fixed

Opus agents (6): add efficiency constraints, tool failure handling, grounding, read-only clauses
Sonnet agents (24): add anti-eagerness clauses, scope locks for review agents
infra-engineer: add destructive action safety for cloud CLI commands
docs-keeper: add write scope ceiling for Edit/Write/MultiEdit tools
web-engineer: add scope boundaries, failure handling, workflow structure
web-docs: add "Run Tooling First" section (was only doc agent missing it)
managing-infra: add mandatory dry-run before terraform/kubectl/helm apply
Writing skills (4): add verify-after-generate with build/lint commands
committing-code: add secrets detection guard for .env/pem/credentials
documenting-code, improving-tests, debating-ideas: add failure handling
mem-history: add scope description and output format template

Changed

Skill count: 29 → 30 (new linting-instructions in dev-tools)
All instruction fixes derived from Claude Opus 4.6 and Sonnet 4.6 system cards

[1.9.1] - 2026-05-03

[1.2.0] - 2026-03-31

Optional claude-mem integration for AST-based code navigation and cross-session memory.

Added

smart-explore skill: AST-based code navigation via smart_outline/smart_search/smart_unfold (10-20x token savings)
mem-history skill: cross-session memory search with 3-layer workflow and graceful fallback
Claude-mem MCP tools added to all 30 agents (25 review + 5 engineer) as optional tools
Historical context check in reviewing-code (Step 0) and fixing-code (Pre-Phase 2)
Project history option in brainstorming-ideas Phase 3 checkpoint
Smart Explore column in searching-code decision table
smart-explore and mem-history detection patterns in skill-enforcer hook
Claude-Mem Integration section in README with installation guide and resilience docs

Changed

Skill count: 27 → 29 (2 new skills in dev-tools plugin)
All review agent frontmatter converted to multi-line tools format
Engineer agents gain ### Memory (claude-mem) body section

[1.9.1] - 2026-05-03

[1.1.1] - 2026-03-31

Full repository review and cleanup.

Fixed

Broken specctl test path after plugin restructuring (10 test failures)
Ruff config referencing deleted scripts/ce (linter was scanning nothing)
Plugin version mismatch (all plugin.json now match marketplace 1.1.1)
Skill-enforcer missing 4 user-invocable skills
learning-patterns skill name mismatch (was learn, now matches directory)
CI gate not detecting cancelled job state
README skill counts (26 → 27)

Added

Per-plugin README.md for all 9 plugins
Makefile with lint, test, validate, fmt, flat, ci, setup, release targets
CONTRIBUTING.md with plugin authoring guide and PR checklist
Pre-commit hook running full CI pipeline
Release tag script with automatic version bumping
CI badge, version badge, and project narrative in README
Dependabot pip ecosystem tracking
using-gemini skill documented in README

Changed

Replaced monolithic GUIDE.md with per-plugin READMEs + expanded project README
CI and release workflows now use Makefile targets
All lint/test jobs run unconditionally on push to master
Dependabot frequency: monthly → weekly

Removed

GUIDE.md (split into per-plugin READMEs)
Orphaned root files: claude-powerline.json, MCP_Sequential.md, .claude-hooks-config.sh, .claude-hooks-ignore
install-tools.sh (user-specific, not marketplace-related)

[1.9.1] - 2026-05-03

[1.1.0] - 2026-03-30

Restructured as a 9-plugin marketplace for community sharing.

Added

MIT LICENSE file
Marketplace metadata (description, version, categories, tags)
Plugin-level plugin.json manifests for all 9 plugins

Changed

Restructured from flat config into 9 installable plugins
26 skills, 34 agents, 9 hooks, 9 commands across all plugins
Updated README with correct installation syntax per official plugin docs
Updated GUIDE with plugin-relative paths and companion tool notes

[1.9.1] - 2026-05-03

[1.0.0] - 2026-02-28

Initial versioned release of the Claude Code configuration.

Added

23 skills: brainstorming, committing, debating, deploying-infra, documenting,
evolving-config, fixing, improving-tests, learning-patterns, looking-up-docs,
managing-infra, refactoring, researching-web, reviewing, searching,
testing-e2e, using-cloud-cli, using-git-worktrees, using-modern-cli,
writing-go, writing-python, writing-typescript, writing-web
14 agents for Go, Python, TypeScript, web, infrastructure, and planning
Spec-driven development system (specctl + spec skills)
CI workflow with config validation, linting, and tests
Global CLAUDE.md with universal development practices
Project CLAUDE.md with config-repo-specific guidance