shawn-sandy/agentics

Structured multi-dimensional code review across quality, bugs, security, best practices, complexity rating, breaking changes, and regressions

Analyze code and suggest specific, purpose-driven tests tied to actual behavior and intent — not arbitrary coverage

Automated git workflow — create branches, commit with conventional messages, create PRs, and open GitHub/GitLab issues from any context (selection, session, bug, feature) with host auto-detection and confirmation gate; ships background subagents and slash commands for fire-and-forget commit/PR/ship plus auto-linking plan issue references in PR descriptions

Audit and optimize CLAUDE.md project memory files against Claude Code best practices

Plan creation and review on demand or via ambient activation — /plan-agent:implementation-plan skill (command or model-invocable) produces self-contained interactive HTML plans with an opt-in Resources section (images, screenshots, and reference links used to create the plan, so readers can illustrate and verify the implementation), a rich copy-paste implement prompt (includes status, steps, acceptance criteria, and completion instructions), an always-present outcome-driven goal prompt (achieve the objective using the plan as reference but optimizing for the outcome), a one-click Save as PDF export button, and optional workflow prompt for complex plans; auto-activates on plan-document intent without colliding with built-in Plan Mode; /plan-agent:review-plan skill spawns a seven-reviewer Agent Team (architecture, completeness, testability, risk, conventions, UX, accessibility) to review implementation plans, synthesize findings, apply improvements in place, and emit shareable HTML review artifacts; /plan-agent:finalize-plan skill reviews a plan for completion evidence with per-criterion verification, checks acceptance-criteria boxes (all or only verified, per user choice), and sets status to completed or in-progress accordingly, with an --all sweep mode that discovers done-but-unmarked plans across the plans directory and batch-finalizes them behind one multi-select confirmation; accepts GitHub/GitLab issue URLs and #n references to auto-seed plans from backlog; accepts .md plan paths for conversion mode that turns an existing markdown plan into the HTML format with frontmatter carry-over and a keep-or-remove source prompt; built-in structured interview; mandatory acceptance criteria gate during implementation; Step 8 exit menu offers Review the plan option that runs the review-plan Agent Team on the current plan (foreground or background) before implementing; adaptive menu swap keeps all options within the 4-option limit; /plan-agent:refine-prompt skill interviews users and generates copy-pasteable AI prompts grounded in Anthropic best practices (role, XML structure, CoT, examples, output format); plans-open skill reopens the gallery without rebuilding; PostToolUse hook auto-regenerates the gallery index; filename hook enforces verb-target kebab-case; /plan-agent:build-proposal skill turns a vague idea into a decision-complete proposal under docs/proposals/ via an 8-step research-then-decide loop with a Tier 0/1/2 right-sizing gate, grounding claims in web + codebase research and separating facts from decisions, then hands the proposal to implementation-plan for execution planning; /plan-agent:setup-sites skill scaffolds the GitHub Pages deploy pipeline into any repo (deploy-pages.yml workflow, docs/.nojekyll, landing hub, serve-docs.sh preview), computes the live URL from the git remote, and guides the one-time Settings → Pages → GitHub Actions step so generated docs/ HTML publishes to a public URL; /plan-agent:prototype skill turns a completed HTML plan, a one-line idea, an image/screenshot/mockup, or a Figma design into a runnable, framework-free static-HTML prototype under docs/prototypes/ (inline JSON seed + per-prototype localStorage, escaped output, accessibility baked in), auto-indexed into a Prototypes gallery reachable from the docs hub

Stress-test implementation plans with structured multi-round interviews before coding begins — auto-routes product plans to the panel review skill, always emits an interview HTML artifact

Improve, optimize, and update product plans, PRDs, and feature proposals using a cross-functional Agent Team (PM, Lead Developer, UX, Frontend, Accessibility, Security) — produces a 15-section consolidated report, applies improvements to the source plan, and appends findings to any existing plan-interview HTML artifact

Back up and restore Claude Code user settings to a dedicated git repo — routine-compatible for automated backups

Review and optimize Claude Code skill files — score SKILL.md quality, plan and scaffold new skills, audit allowed-tools permissions, optimize frontmatter (descriptions + disable-model-invocation), and enforce the three-part 200-char description format

Discover teachable code, blog posts, videos, GitHub snippets, and selected/pasted code — scrub for secrets, draft instructional platform-aware copy with concrete takeaways, and generate styled dark-mode social cards for LinkedIn, Twitter/X, Bluesky, and Substack, plus a write-guide skill that writes long-form internal developer explainer guides to docs/

Review HTML/CSS and React/TypeScript code for WCAG 2.2 Level AA accessibility compliance