Fixed

- Replaced broken `SessionStart` prompt hook in `hooks/hooks.json` with a working `command`-type hook that invokes the session-start script, eliminating "ToolUseContext is required for prompt hooks" startup errors.

Fixed

- Normalized file permissions on `hooks/run-hook.cmd` and `hooks/session-start` from executable (100755) to standard (100644) to match the rest of the repository and avoid spurious modified-file noise after pulls.

Added

- Added full Obsidian vault skill suite: six new skills covering local vault workflows — `obsidian-docs` (vault structure, Maps of Content, navigation, and knowledge-base patterns), `obsidian-config` (.obsidian/ settings, themes, CSS snippets, and sync), `obsidian-flavored-markdown` (wikilinks, embeds, callouts, block references, properties, and tags), `obsidian-plugins` (core and community plugin management and BRAT), `obsidian-templates` (Core Templates and Templater plugin authoring), and `obsidian-canvas` (.canvas JSON schema and visual knowledge maps).
- Added `agents/obsidian-maintainer.md` domain specialist agent scoped to the six Obsidian skills.
- Added `commands/obsidian-vault-health.md` health-check command covering broken wikilinks, orphan notes, canvas JSON validation, frontmatter audits, tag frequency, and plugin config validation.

Fixed

- Replaced broken `SessionStart` prompt hook in `hooks/hooks.json` with a working `command`-type hook that invokes the session-start script via `run-hook.cmd`, eliminating "ToolUseContext is required for prompt hooks" errors. Also added `PostToolUse` hook that validates `.canvas` files as well-formed JSON after any Write operation.

Fixed

- Replaced broken `SessionStart` prompt hook in `hooks/hooks.json` with a working `command`-type hook that invokes the session-start script, eliminating "ToolUseContext is required for prompt hooks" startup errors.

Fixed

- Fixed auto-update detection for all installed plugins. Plugin entries in `catalog.json` and the generated `marketplace.json` now include explicit `version` and pinned `ref` fields (e.g. `"ref": "v1.7.7"`, `"version": "1.7.7"`) instead of `"ref": "main"` with no version. Claude Code compares the installed version against the marketplace-listed version to detect updates; without a version field, no update was ever detected.
- Added `version` field to `MarketplacePluginDocument` interface and `buildMarketplacePluginDocument` generation so the published `marketplace.json` carries per-plugin version metadata.
- Added `version` field to the `PluginConfig` catalog type so version data flows cleanly from `catalog.json` through the generator into the published document.
- Updated `deploy.yml` to automatically update `catalog.json` with the new version tag when triggered by a `plugin-updated` repository dispatch. Previously, dispatches from upstream repos only triggered a rebuild without updating the pinned ref or version, so the marketplace.json never reflected new releases.
- Updated `settings.json` marketplace registration from `source: "git"` (URL) to `source: "github"` (repo) to match all other working marketplaces and enable GitHub API-based update checks.

Fixed

- Replaced broken `SessionStart` prompt hook in `hooks/hooks.json` with a working `command`-type hook that invokes the session-start script, eliminating "ToolUseContext is required for prompt hooks" startup errors.

Fixed

- Replaced broken `SessionStart` prompt hook in `hooks/hooks.json` with a working `command`-type hook that invokes the session-start script, eliminating "ToolUseContext is required for prompt hooks" startup errors.

Added

- added four Planguage requirement workflow skills: `planguage-author`, `planguage-reader`, `planguage-implementor`, and `planguage-tester`
- added `agents/planguage-maintainer.md` plus command scaffolding in `commands/skill-development.md`, `commands/claude-command-sdlc.md`, `commands/claude-hook-sdlc.md`, and `commands/claude-agent-sdlc.md` to coordinate the new Planguage workflow

Changed

- strengthened `planguage-author` with an explicit low-evidence fallback path so the agent stops at a tagged requirement skeleton plus missing facts instead of inventing numeric targets
- updated `hooks/hooks.json` to an explicit safe hook scaffold with empty `PreToolUse` and `PostToolUse` lists
- updated `README.md` and `docs/skill-test-drive-report.md` to reflect the expanded 120-skill catalog

Fixed

- sorted imports in `tests/test_well_documented_tools.py` so the Python lint job passes consistently in CI
- restored executable git mode bits on `skills/well-documented/scripts/*.sh` so Linux CI can run the helper scripts without `Permission denied`

Fixed

- Removed broken `SessionStart` prompt hook from `hooks/hooks.json`. The `type: "prompt"` hook requires a `ToolUseContext` that does not exist at session start, causing "ToolUseContext is required for prompt hooks" errors on every startup.

Fixed

- Removed broken `SessionStart` prompt hook from `hooks/hooks.json`. The `type: "prompt"` hook requires a `ToolUseContext` that does not exist at session start, causing "ToolUseContext is required for prompt hooks" errors on every startup.

Fixed

- Removed broken `SessionStart` prompt hook from `hooks/hooks.json`. The `type: "prompt"` hook requires a `ToolUseContext` that does not exist at session start, causing "ToolUseContext is required for prompt hooks" errors on every startup.

Fixed

- Removed broken `SessionStart` prompt hook from `hooks/hooks.json`. The `type: "prompt"` hook requires a `ToolUseContext` that does not exist at session start, causing "ToolUseContext is required for prompt hooks" errors on every startup.

Fixed

- Removed broken `SessionStart` prompt hook from `hooks/hooks.json`. The `type: "prompt"` hook requires a `ToolUseContext` that does not exist at session start, causing "ToolUseContext is required for prompt hooks" errors on every startup.

Added

- Added 30-skill Claude workflow development suite across three full SDLC families: `claude-command-*`, `claude-hook-*`, and `claude-agent-*`. Each family includes planning, design, creation, testing, verification, integration, validation, test-drive, iterative-development, and an SDLC orchestrator skill.
- Added cross-compatible guidance for command workflows, hook automations, and delegated agent workflows that stays runnable from both Claude and Codex while documenting runtime-specific placement and portability constraints in references.

Fixed

- Improved Wave 1 runtime guidance across 20 blocked-scenario skills: `7z`, `ag`, `binwalk`, `exiftool`, `hexdump`, `less`, `markdownlint`, `mediainfo`, `powershell`, `sqlite`, `ssh-client`, `ssh-keygen`, `tree`, `xml2`, `xmllint`, `xq`, `xxd`, `yq`, `zoxide`, and `zsh`.
- Added early prerequisite checks, clearer install and fallback paths, stronger Quick Start sections where needed, and extra inline runnable examples so common workflows fail fast and stay actionable when a required binary is missing.
- Updated `skills/yq` metadata and routing text so the skill validates cleanly with the new runtime-gating guidance.

Added

- added `high-coherence`, a software design skill for preserving conceptual integrity, architectural conformance, and runtime alignment as systems evolve
- added `low-coupling`, a software design skill for reducing structural, contract, temporal, and operational coupling with pragmatic, behavior-preserving refactors

Changed

- updated `README.md` to reflect the expanded 116-skill catalog and include the two new design and review skills

Added

- Added repository-wide skill agent and command scaffolding for all skills, including per-skill `agents/` manifests and `commands/` entries, so skill invocations can be routed consistently by both Claude and Codex clients.
- Added `hooks/hooks.json` with preconfigured hooks for agent selection and command execution orchestration.

Fixed

- Fixed packaging invariants so each built skill ZIP is verified to include the required `SKILL.md`, `agents/claude.yaml`, and `agents/openai.yaml` members under the repo-rooted path (`llm-doc-skills/skills/<skill>/...`).
- Hardened `tests/test_ooxml_validate.py` temp workflow by using deterministic repo-local work directories and explicit cleanup to avoid Windows tempfile lifecycle and path-fragility failures.

Added

- Added repository-wide skill agent and command scaffolding for all skills, including per-skill `agents/` manifests and `commands/` entries, so skill invocations can be routed consistently by both Claude and Codex clients.
- Added `hooks/hooks.json` with preconfigured hooks for agent selection and command execution orchestration.

Added

- Added 10-skill MCP server development SDLC suite: `mcp-planning`, `mcp-design`,
  `mcp-creation`, `mcp-testing`, `mcp-verification`, `mcp-integration`,
  `mcp-validation`, `mcp-test-drive`, `mcp-iterative-development`, and `mcp-sdlc`
  (master orchestrator). The suite covers the full lifecycle from ideation to
  shipping a production-quality MCP server, with quality rubric M01–M06, live
  scenario test-drive, and re-entry loop management. Added 10 new rows to README.
- Included MCP-specific reference content: TypeScript and Python scaffold templates,
  FastMCP patterns, MCP Inspector commands, JSON-RPC message shapes, schema
  validation recipes, PreToolCall hook patterns for debugging and safety gating,
  client config for Claude Code / Claude Desktop / Cursor, and a path-traversal +
  shell-injection prevention rubric.

Fixed

- Applied 7 test-drive friction fixes to the new MCP SDLC suite: added explicit
  Gate sections to `mcp-design`, `mcp-creation`, `mcp-testing`, and
  `mcp-integration` (previously lacked completion criteria); split
  `mcp-verification` checklist into static and Inspector-required groups; corrected
  `mcp-sdlc` Phase 1 gate to include all 4 planning outputs; defined the "Required"
  column in the capability inventory table; added test dependency install commands
  to `mcp-testing`; added FastMCP exception-wrapping note to the M03 quality rubric;
  added multi-FAIL sequencing guidance to `mcp-iterative-development`.
- Extended `skills/git/references/troubleshooting.md` with a dedicated **Git-for-Windows sh.exe Win32 Error 5 (Access Denied)** section covering symptoms, root causes (Windows Defender, Controlled Folder Access, integrity level mismatch, stale Git-for-Windows, LFS filter invocation), diagnosis steps, five ordered fixes, and a Git plumbing fallback (`update-index` → `write-tree` → `commit-tree` → `update-ref`) for when high-level commands are unreliable.
- Added three new rows to the Common Error Messages quick-reference table for the `sh.exe` signal-pipe error, `CreateFileMapping` error, and the resulting `fatal: the remote end hung up unexpectedly`.
- Updated the Intent Router in `skills/git/SKILL.md` to route Win32 error 5 / sh.exe fatal error queries to `references/troubleshooting.md`.

Fixed

- Tightened packaging verification so built ZIPs are checked for required members under the repo-rooted path (`llm-shared-skills/skills/<skill>/...`) before release.
- Added Windows-safe temporary directory handling and cleanup in the tooling package where needed to reduce test failures on Windows hosts.

Added

- Added repository-wide skill agent and command scaffolding for all skills, including per-skill `agents/` manifests and `commands/` entries, so skill invocations can be routed consistently by both Claude and Codex clients.
- Added `hooks/hooks.json` with preconfigured hooks for agent selection and command execution orchestration.

Fixed

- Hardened `tests/test_well_documented_tools.py` with deterministic repo-local temporary directories, safe cleanup, and Bash path normalization for command invocation to prevent Windows tempfile and path handling regressions.
- Upgraded packaging verification so built ZIPs are checked for required files under `llm-software-design/skills/<skill>/...`, making repo-rooted layout mismatches fail during validation.

Added

- Added repository-wide skill agent and command scaffolding for all skills, including per-skill `agents/` manifests and `commands/` entries, so skill invocations can be routed consistently by both Claude and Codex clients.
- Added `hooks/hooks.json` with preconfigured hooks for agent selection and command execution orchestration.

Fixed

- Enforced repository-rooted ZIP layout checks so built skill packages include required files under `llm-web-design-skills/skills/<skill>/...` and fail early when this contract is broken.

Added (release automation)

- Added repository-wide skill agent and command scaffolding for all skills, including per-skill `agents/` manifests and `commands/` entries, so skill invocations can be routed consistently by both Claude and Codex clients.
- Added `hooks/hooks.json` with preconfigured hooks for agent selection and command execution orchestration.

Added

- `Makefile`: added `bundle` target that packages all built `*-skill.zip` files into a single `ci-cd-plugin.zip` for one-click offline installation; `PLUGIN_NAME := ci-cd` variable drives the output filename

Fixed

- `Makefile`: restored full Makefile content that was accidentally truncated to zero lines by a patching script; all original targets (`all`, `build`, `clean`, `verify`, `lint`, `test`, `test-unit`, `list`) are preserved

Changed

- `Makefile`: extended `.PHONY` to include `bundle`; updated `help` text to document the new target
- `.github/workflows/release.yml`: added "Build plugin bundle ZIP" step (`make bundle`) so `ci-cd-plugin.zip` is uploaded alongside the individual skill ZIPs on every tag release; converted CRLF line endings to LF for yamllint compliance

Changed

- documented trunk-based development workflow in `AGENTS.md`

Added

- Added repository-wide skill agent manifest (agents/doc-workflow-maintainer.md), command entries (commands/deck-quality.md, commands/quality-gate.md), and hooks/hooks.json for consistent skill routing across Claude and Codex clients.
- Extended AGENTS.md with command + agent template section for focused doc-skill loops.

Changed

- Hardened scripts/verify_built_zips.py with REQUIRED_FILES invariant checks ensuring each built ZIP contains SKILL.md, agents/claude.yaml, and agents/openai.yaml.
- Normalized line endings across repository files (CRLF → LF).

Added

- `Makefile`: added `bundle` target that packages all built `*-skill.zip` files into a single `doc-skills-plugin.zip` for one-click offline installation; `PLUGIN_NAME := doc-skills` variable drives the output filename

Changed

- `Makefile`: extended `.PHONY` to include `bundle`; updated `help` text to document the new target
- `.github/workflows/release.yml`: added "Build plugin bundle ZIP" step (`make bundle`) so `doc-skills-plugin.zip` is uploaded alongside the individual skill ZIPs on every tag release

Added

- `skills/raw-document`: added XML validation reference, helper script, and tests for inspecting and repairing documents at the raw byte or XML level

Changed

- moved all skill directories into a `skills/` subdirectory to align the repo layout with the rest of the skill workspace
- updated all internal script paths, test imports, and README links to match the new `skills/` prefix

Fixed

- `tests/test_xlsx_recalc.py`: loader now catches `SystemExit` from `recalc.py` when openpyxl is not installed and converts it to `ImportError` so the test class is skipped cleanly instead of aborting discovery
- `VALIDATION_REPORT.md`: removed double blank line that triggered MD012
- README skill table and Quick Start command paths corrected to include `skills/` prefix

Added

- Added repository-wide skill agent manifest (agents/shared-skills-maintainer.md), command entries, and hooks/hooks.json for consistent skill routing across Claude and Codex clients.
- Extended AGENTS.md with command + agent template section covering focused single-skill lint/validate loops and full skill-author workflow checklist.

Changed

- Hardened scripts/verify_built_zips.py with REQUIRED_FILES invariant checks ensuring each built ZIP contains SKILL.md, agents/claude.yaml, and agents/openai.yaml.

Added

- `Makefile`: added `bundle` target that packages all built `*-skill.zip` files into a single `shared-skills-plugin.zip` for one-click offline installation; `PLUGIN_NAME := shared-skills` variable drives the output filename

Changed

- `Makefile`: extended `.PHONY` to include `bundle`; updated `help` text to document the new target
- `.github/workflows/release.yml`: added "Build plugin bundle ZIP" step (`make bundle`) so `shared-skills-plugin.zip` is uploaded alongside the individual skill ZIPs on every tag release

Added

- `skills/cat`: display and concatenate file contents; covers tac, heredoc alternatives, and "useless use of cat" guidance
- `skills/cp`: copy files and directories with metadata/permission preservation, archive mode, and safe overwrite patterns
- `skills/echo`: print strings to stdout with portability notes; defers to `$printf` for escape handling in scripts
- `skills/grep`: search files and pipelines by pattern; prefers ag (The Silver Searcher) for recursive code search when installed; includes ag flags, when-to-use table, and PCRE notes
- `skills/ln`: create hard and symbolic links; covers atomic symlink replacement, relative vs absolute targets, and cross-filesystem limits
- `skills/ls`: list directory contents; covers GNU vs BSD flag differences, sorting options, colour output, and safe scripting alternatives
- `skills/mv`: move and rename files and directories; covers atomic same-filesystem rename, cross-filesystem copy-then-delete, and overwrite protection
- `skills/printf`: portable formatted output; covers format specifiers, field widths, ANSI colour, `printf -v`, null-delimited output, and the "never pass variable as format string" safety rule
- `skills/ps`: inspect running processes; covers BSD/POSIX syntax differences, custom output columns, process trees, and top/htop alternatives
- `skills/rsync`: sync files locally or over SSH; emphasises dry-run-first workflow, trailing-slash semantics, `--delete` warning, and bandwidth/checksum options

Changed

- Landing page: moved plugin GitHub repository links from the crowded top-of-page link row into individual plugin cards (displayed beneath the install command banner), reducing header clutter and keeping repo context close to each plugin.
- Landing page: removed horizontal scroll overflow from the install command text box on plugin cards so the command displays on a single fixed-width line.
- Landing page: added a copy-to-clipboard button beside each plugin card install command, matching the copy-button pattern already used on Skill detail cards.

Fixed

- `release.yml`: `all-plugins.zip` now contains individual `*-skill.zip` files flat at the archive root rather than nesting the plugin bundle ZIPs inside it; each plugin bundle is downloaded then extracted, and the resulting skill ZIPs are collected with `zip -j` (junk paths) into a single flat archive

Added

- `MarkdownRenderer` component (`src/components/MarkdownRenderer.tsx`): server component using `react-markdown` + `remark-gfm` with `@tailwindcss/typography` prose styling; replaces raw `<pre>` blocks on skill detail pages so all skill documentation is rendered as HTML
- `buildPluginBundleUrl()` in `src/lib/github/skill-service.ts`: builds the `{pluginName}-plugin.zip` download URL from a plugin config and version
- `buildAllPluginsBundleUrl()` in `src/lib/catalog/service.ts`: builds the `all-plugins.zip` download URL from `catalog.json` marketplace version and GitHub repo
- `MarketplacePluginSummary.bundleUrl` field: per-plugin all-in-one bundle download URL, or `null` when the plugin version is unavailable
- Landing page: per-plugin "Bundle" download link on each plugin card and a "Download all plugins" banner at the top of the marketplace that links to `all-plugins.zip`
- `release.yml`: two new steps — "Download plugin bundle ZIPs" (fetches each `{pluginName}-plugin.zip` from the latest plugin release) and "Build all-plugins bundle" (assembles `all-plugins.zip` from the downloaded bundles); `all-plugins.zip` is uploaded as a GitHub Release asset
- Unit tests for `buildPluginBundleUrl` and `bundleUrl` population in `buildMarketplacePluginSummaries`

Changed

- Skill detail page (`src/app/skill/[...slug]/page.tsx`): replaced `<pre>` raw markdown display with `<MarkdownRenderer>` for rendered HTML output
- `src/lib/github/index.ts` and `src/lib/catalog/index.ts`: barrel exports updated for new URL builder functions
- `package.json`: added `react-markdown`, `remark-gfm` as runtime dependencies; `@tailwindcss/typography` as dev dependency
- `tailwind.config.ts`: registered `@tailwindcss/typography` plugin

Fixed

- `catalog.json` and `package.json` version corrected to `1.4.0`; the v1.3.9 tag was pushed without bumping these files, causing the release workflow to fail the version-gate check

Added

- registered `llm-web-design-skills` as a new plugin in the marketplace catalog, regenerated `marketplace.json`, and updated the README plugin list

Changed

- documented trunk-based development workflow in `AGENTS.md`
- updated `package-lock.json` to v1.3.8

Added

- Added repository-wide skill agent manifest (agents/software-design-maintainer.md), command entries, and hooks/hooks.json for consistent skill routing across Claude and Codex clients.
- Extended AGENTS.md with command + agent template section for focused skill loops.

Changed

- Hardened scripts/verify_built_zips.py with REQUIRED_FILES invariant checks.

Added

- `Makefile`: added `bundle` target that packages all built `*-skill.zip` files into a single `software-design-plugin.zip` for one-click offline installation; `PLUGIN_NAME := software-design` variable drives the output filename

Changed

- `Makefile`: extended `.PHONY` to include `bundle`; updated `help` text to document the new target
- `.github/workflows/release.yml`: added "Build plugin bundle ZIP" step (`make bundle`) so `software-design-plugin.zip` is uploaded alongside the individual skill ZIPs on every tag release; converted CRLF line endings to LF for yamllint compliance

Added

- added evidence-first tooling and references to `well-documented`, including documentation maturity levels, project-type README templates, truth-and-freshness audit guidance, and new unit tests for the helper scripts
- added a second well-documented test-drive pass focused on bootstrap honesty, audit signal quality, and script reliability

Changed

- made `well-documented` adaptive instead of recursive-by-default: `audit-docs`, `init-docs`, and `normalize-docs` now support `minimal`, `standard`, and `full` maturity levels
- upgraded `well-documented/scripts/audit-docs.sh` to check broken markdown links, unresolved local path references, missing local script or Make-target references, bootstrap-marker drift, and git freshness signals when available
- upgraded `well-documented/scripts/init-docs.sh` and `normalize-docs.sh` to use project-type-aware root README templates and a lighter default baseline
- upgraded `well-documented/scripts/add-file-headers.sh` so generated headers are explicitly marked as bootstrap text and shell shebangs are preserved cleanly
- replaced placeholder-heavy `well-documented` templates with more actionable scaffolds and added root README variants for library, cli, service, web app, monorepo package, and data pipeline repositories
- test-drove the full 114-skill catalog and added explicit verification or recovery guidance to every skill so common requests now end with a clearer follow-through path
- rewrote `well-documented` for cleaner command selection, stronger safety notes, and a proper Quick Start flow
- deepened 42 thinner `*-docstrings` skills with family-specific depth sections, anti-patterns, and reference starting points so they better match the richness of the strongest catalog anchors
- added `docs/skill-test-drive-report.md`, `docs/docstrings-depth-test-drive.md`, `scripts/skill_affordance_report.py`, `tests/test_skill_affordances.py`, and `tests/test_docstring_reference_depth.py` to preserve the new usability baseline
- updated `AGENTS.md` with `skill_affordance_report.py` usage guidance and a pointer to `docs/skill-test-drive-report.md`
- updated `README.md` to link to the new test-drive report

Fixed

- `well-documented/scripts/add-file-headers.sh` now preserves executable file modes when rewriting scripts with generated headers
- `well-documented/scripts/audit-docs.sh` now keeps PASS and SKIP counters accurate in quiet mode so score summaries remain truthful
- `well-documented/scripts/init-docs.sh` no longer bootstraps root README templates with broken license links that immediately fail the evidence audit
- `well-documented/scripts/audit-docs.sh` R06 abort: added `|| true` guard to
  `fname=$(... | grep -oP ...)` command substitution so the script no longer
  exits early under `set -euo pipefail` when a line in AGENTS.md contains no
  backtick-quoted file reference — previously the audit silently stopped after
  R05 on any repo that had an AGENTS.md
- `well-documented/scripts/audit-docs.sh` M02 misleading output: tool crashes
  (e.g. `SyntaxError` from Node.js < 20) now surface as a WARN with an
  actionable message rather than counting error stack lines as "violations"
- `well-documented/SKILL.md`: document Node.js ≥ 20 requirement for
  `markdownlint-cli2`; include `nvm` workaround note

Added

- Added repository-wide skill agent manifest (agents/web-maintainer.md), command entries, and hooks/hooks.json for consistent skill routing across Claude and Codex clients.
- Extended AGENTS.md with command + agent template section for focused web-design skill loops.

Changed

- Hardened scripts/verify_built_zips.py with REQUIRED_FILES invariant checks.

Added

- `Makefile`: added `bundle` target that packages all built `*-skill.zip` files into a single `web-design-skills-plugin.zip` for one-click offline installation; `PLUGIN_NAME := web-design-skills` variable drives the output filename

Changed

- `Makefile`: extended `.PHONY` to include `bundle`; updated `help` text to document the new target
- `.github/workflows/release.yml`: upgraded from minimal stub to full plugin release pipeline — adds version gate, changelog extraction, `softprops/action-gh-release` upload of `built/*.zip` (including `web-design-skills-plugin.zip`), `changelog` extraction, and marketplace dispatch trigger; added "Build plugin bundle ZIP" step (`make bundle`)

- Initial release of `llm-web-design-skills`.
- Added 16 cross-compatible web framework and web design skills:
  - **Frontend frameworks**: `angular`, `nextjs`, `react`, `svelte`, `vue`
  - **Styling and UI systems**: `material-design`, `scss`, `tailwind`
  - **Build and tooling**: `vite`
  - **Backend frameworks**: `django`, `nodejs`, `express-js`, `asp-net`, `spring-boot`, `laravel`
  - **Static sites and content**: `hugo`
- Each skill ships with Claude and OpenAI agent definitions, best-practices reference, and workflow guide.

Added

- `skills/skill-development`: new meta-skill orchestrating the full 7-phase SDLC for authoring LLM agent skills — covers ideation, authoring via skill-creator, structural linting via skill-linting, qualitative validation via skill-validation, live scenario testing via skill-test-drive, iteration loops, and final commit and ship; includes five reference files and two rounds of live test-drive fixes

Changed

- confirmed `llm-ci-dev` as the canonical home for `aws`, `az`, `docker`, and `glab`, preserving the most recent examples and wording improvements while shared duplicates were removed from `llm-shared-skills`

Fixed

- inserted the required blank line in `AGENTS.md` and removed the unused import from `scripts/verify_built_zips.py` so the normalized CI and release gates pass their markdown and Python lint steps on GitHub

Changed

- standardized the repo on a Make-first build and validation workflow: `make lint`, `make test`, `make build`, `make verify`, and `make all`
- added repo-owned Python packaging helpers and stdlib-only tests so ZIP verification and skill checks behave consistently across Windows-backed WSL and GitHub Actions
- renamed the primary GitHub Actions quality workflow to `CI` in `.github/workflows/ci.yml` and aligned release automation to run `make test`, build `built/*.zip`, attach artifacts, and skip marketplace dispatch cleanly when the token is missing
- kept `linting/` and `validation/` shell entrypoints as compatibility wrappers around the new Python baseline for one release cycle
- aligned the public repo and plugin identity on `llm-ci-dev` so plugin metadata, install examples, and human-facing docs now match the actual repo name and ZIP archive root
- fully self-contained the release workflow by provisioning `jq` explicitly and updated the release guide to match the automated `make test` + `make all` release path and optional marketplace dispatch behavior

Fixed

- added release-time Node and lint/tool provisioning to `.github/workflows/release.yml` so tag releases run the same `make test` gate successfully on GitHub-hosted runners
- restored trailing newlines in the newly added skill agent manifests and wrapper scripts, and removed the unused import from `scripts/verify_built_zips.py`, so the CI YAML and Python lint jobs pass for the text-first publishing and diagram skills

Changed

- generalized the repo-owned skill lint and validation helpers so they can share the same baseline logic as the other plugin repos while still supporting `llm-doc-skills` top-level skill layout

Added

- New packaged text-first publishing skills: `pandoc`, `latex`, `typst`, `markdown`, and `asciidoc`.
- New packaged platform-and-diagram skills: `github-flavored-markdown`, `gitlab-flavored-markdown`, `mermaid`, `plantuml`, and `graphviz`.
- Repo-owned wrapper CLIs for common conversion, build, render, and export flows under the text-first and diagram skill `scripts/` directories.
- Companion `references/` docs for the new skills so authoritative guidance stays deep without pushing `SKILL.md` over repo lint thresholds.
- Packaging and wrapper tests covering the new skills and their command-builder surfaces.
- `scripts/verify_built_zips.py` and `scripts/check_node_version.py` to make repo verification and lint prerequisites more explicit across environments.

Fixed

- `Makefile`: route ZIP verification through a Python helper so `make verify` checks all built archives consistently on Windows-backed WSL paths.
- `Makefile`: fail early with a clear message when the active Node.js runtime is older than the version required by `markdownlint-cli2`.
- `.gitignore`: ignore repo-local scratch directories and transient `tmp*/` artifacts so failed local temp cleanup does not pollute git status.

Added

- `skills/skill-test-drive`: new meta-skill for test-driving other skills by designing live scenarios, executing them with the target skill's guidance, and reporting usability gaps and missing examples before release