kyzdes' personal Claude Code skill marketplace.
Cross-platform (macOS + Windows + Linux) secrets manager skill: stores API keys, SSH keys, server creds, and domain info in the OS-native credential store (macOS Keychain / Windows Credential Manager / Linux Secret Service, encrypted-file fallback on headless servers) via the `keys` CLI. Agents architecturally cannot reveal plaintext (Sealed wrapper + env-gated reveal). First-class Claude Code skill; `keys init` emits equivalent rule files for Cursor, Aider, Codex CLI, and Cline.
Turn loose product descriptions into structured English UX Specs ready for handoff to a visual-design agent.
Two-layer agent docs: a committed agent-docs/ navigation tree (MAP + per-domain deep docs) and a gitignored context-map-<slug>/ memory tree (decisions, known issues, gotchas, tasks). Generate, decompose, update, audit, reconcile, and conflict-check project docs for AI coding agents — navigation that survives a big codebase plus memory that survives a context reset.
Extract flight + hotel data from Aviasales / Ostrovok links and compile into a self-contained HTML itinerary.
Google Stitch AI UI generation — brainstorm, generate, edit, iterate on UI designs from text prompts. Bundles 4 sub-skills (design / theme / edit / upload).
Teach AI agents to work the Agentix issue tracker over MCP: orient with get_started + the wiki index, load one issue with get_context, and file well-specced issues (task-spec + checklist). Ships the full 43-tool reference (incl. delete/unlink), a REST-mirror map, a task-creation template, the index convention, and common workflows.
Instrument an app with SentryX over a remote MCP: error tracking + distributed tracing + product analytics/funnels, correlated by trace_id. Connect with an org Bearer token, then onboard a project (detect stack → 5 product questions → create_project → instrument_hint → define funnels → verify get_funnel) and fix bugs (search_issues → prepare_fix_bundle). Ships the full MCP tool reference, per-stack copy-paste snippets, and end-to-end recipes.
Build a polished, production Telegram bot around a model / HF Space / API — a 7-phase playbook (Discover→Design→Menus→Data→Backend→Build→Deploy), a 32-entry gotchas catalog, and real starter module templates. (Private repo — installs for the owner.)
Agent-native VPS deploy/ops over Dokploy — one command from a GitHub repo to a running app with SSL + auto-deploy, plus a local dashboard for servers, databases, domains, and logs. Deploys via a two-phase worker: a read-only pass analyzes the untrusted repo into a validated manifest, then infra is actuated from that manifest under a plan-then-confirm gate; destroy is guarded by a server-minted single-use nonce. Local, single-user (macOS/Linux).