agentic-utilities

Analyze YouTube videos with content-type detection, multi-agent transcript chunking, optional GitHub repo cross-referencing for tutorials, and structured markdown output. Works in chat mode (inline) or document mode (saved to a configured output directory).

Unified 3D printing pipeline for Bambu Lab printers — design (OpenSCAD MCP), browse (MakerWorld), slice STL→3MF (OrcaSlicer-backed CLI), arrange multi-object plates, and printer control via Bambu MCP. Bundles a TypeScript slicing CLI with patched OrcaSlicer profile handling.

Opinionated, actionable analysis on any topic. Delivers ONE clear recommendation grounded in the user's actual stack — not a menu of options. Auto-detects topic category (operational, architecture, strategy, security, workflow), scales to quick/standard/exhaustive depth, and pulls live docs via find-docs when available.

OpenAI Codex CLI integration for Claude Code. Wraps `codex exec` and `codex resume` with model selection (gpt-5.4 / gpt-5.3-codex / gpt-5.4-nano), reasoning effort control (high/medium/low), sandbox modes (read-only / workspace-write / danger-full-access), session resumption, and stderr suppression for thinking tokens. Requires the `codex` binary on PATH.

Telemetry-driven report on Claude Code skill usage. Bundles a PreToolUse hook that logs every Skill invocation to a plugin-owned JSONL file, plus a report generator that surfaces top-used skills, recently-active skills, dormant skills (size + last-touched), and phantom invocations. Uses ${CLAUDE_PLUGIN_DATA} for telemetry storage so it survives plugin updates and never collides with other plugins.

Create GitHub READMEs that earn stars. Auto-classifies your project as Story-mode (real pain point + narrative arc), Utility-mode (single-purpose tool, visuals do the work), or Hybrid. Routes README visuals through the gpt-image-2 skill (ChatGPT Images 2.0 via the Codex CLI) for diagrams with ~99% text fidelity, with explicit fallback guidance when the user has a different image-gen backend.

Deep implementation review skill that sends a structured adversarial prompt to a heavyweight reviewer (codex gpt-5.5, Gemini 3 Pro, or Opus 4.7) to catch real bugs before they hit production. Returns a trinary verdict with file:line citations and a prioritized P0/P1/P2 fix list.

Babysit a PR end-to-end — push the branch, poll CI on a 3-minute cron, auto-fix red checks, address and validate PR comments, then alert or auto-merge when ready. Runs unattended until merged, closed, or cancelled.

Smart Git workflow engine — repo-aware Git Flow branching, CI monitoring & auto-merge, submodule handling, issue analysis/routing across coding-agent worktrees, deploy-workflow isolation, and changelog awareness. Bundles 8 workflow playbooks (Commit, Branch, Release, PullRequest, CIMerge, Submodule, IssueAnalysis, DeployWorkflow) plus a reusable issue-labeler GitHub Action template.

Generate images with GPT Image 2 inside Claude Code using your existing ChatGPT Plus or Pro subscription — no separate OpenAI API key, no per-image billing. Supports text-to-image, image-to-image editing, style transfer, and multi-reference composition via the local Codex CLI.

Audit a repository for autonomous-agent harness readiness across cold-start docs, rules, API documentation policy, lint, hooks, tests, PR automation, repo skills, and garbage-collection cadence. Adds a Symphony-style unattended ticket-execution overlay: workflow contract, disposable bootstrap, validation/evidence, ticket lifecycle, observability, safety, and smoke-ticket eval. Bundles 11 reference files covering 5 stacks (TypeScript, Swift, Python, Rust, Go) plus the full Symphony readiness and fix-pattern libraries.

Set up parallel git worktrees of the same repo, one per coding harness (Claude Code, Codex CLI, Cursor, etc.), so multiple AI tools can edit, typecheck, and test in parallel without trampling each other's node_modules, build artifacts, or branches. Audits shared-state hazards first, documents gotchas in CLAUDE.md/AGENTS.md, then creates durable work/<tool> branches and sibling worktrees with per-stack installs. Bundles an audit-hazards.sh script and a reset-worktree-to-main.sh script.

Linear issue management via linearis CLI — create, update, close, resolve, and plan issues with team inference, TDD-driven resolution, CI monitoring, and PR review handling. Bundles 12 workflow playbooks (CreateIssue, UpdateIssue, CloseIssue, ListIssues, ResolveIssue, PlanProject, CreateProject, ShipDaily, ManageInitiatives, ManageMilestones, MonitorCI, ReviewPR) plus a reusable GraphQL helper script.

Repository organization expert — structure, audit, and refactor via 8 framework archetypes. Bundles 4 workflow playbooks (NewProject, AuditRepo, RefactorPlan, ExecuteRefactor), a framework selection cheatsheet, health check suite, CI/CD guardrails templates, move map migration spec, and a working-tree indexer tool.

Enforce non-negotiable SaaS application standards — signup, mandatory onboarding, route guards, database schema, backend/ORM selection, and state-management architecture. Prevents the #1 failure mode: signup flows with only email/password and zero onboarding. Bundles 4 workflow playbooks (AuditSignupFlow, ScaffoldAuth, AuditStateManagement, SelectDataLayer) plus reference docs for signup/onboarding, route guards, healthcare SaaS, data layer, and state management.

Execute GitHub issues one at a time as vertical slices — sync, fetch, branch, TDD each acceptance criterion, functionally verify, self-review, open the PR, then babysit it to a terminal state before starting the next. Keeps `to-issues`-style slices from being re-horizontalized during execution: one issue, one branch, one PR, sequential. Bundles the full procedural playbook with completion-state and agent-delegation guidance.

Audit and improve CLAUDE.md / AGENTS.md instruction files against modern harness engineering principles: leanness, anti-staleness, discoverability, and failure-driven gotchas. Scores each file on a 100-point rubric, flags volatile data that will rot, detects fast-moving frameworks (Next.js, Tailwind, React 19) and recommends training-cutoff-honesty blocks, then applies targeted cut/point/add edits after approval.

Systematically explore and QA a web application in a real browser, find bugs and UX issues, and produce a structured report with full reproduction evidence: step-by-step screenshots, repro GIFs when recording is available, and detailed repro steps for every finding. Drives Claude in Chrome (mcp__claude-in-chrome__*) for interactive/authenticated flows and the Playwright CLI for deterministic headless screenshots, so findings can be handed straight to the responsible team.

Security-inspect an AI agent skill with the skillspector CLI, then render a plain-English verdict report in chat: SAFE / CAUTION / DO_NOT_INSTALL, a threat breakdown, the top findings to check first, and file hotspots. Deep mode adds LLM intent classification; the verdict is capability-clustered and intent-weighted, with confirmed source-to-sink exfiltration chains as the categorical do-not-install trigger. Bundles the render script and flags degraded (static-only) scans so an unreliable verdict is never mistaken for a trustworthy one.