Skills Janitor vs Temporal AI

Temporal's product is a programming model: you write workflow logic in ordinary code, and the service persists state at every step so a crashed worker or abandoned session resumes where it stopped, with no checkpointing code. Activities retry automatically with backoff, four timeout classes, and heartbeating; Signals, Queries, and Updates mutate a running execution mid-flight; Workflow Streams add interactivity. AI-specific work is covered via OpenAI Agents SDK and Google ADK running LLM and tool calls as Activities, plus Serverless Workers for Cloud Run (public preview) and Bedrock AgentCore (pre-release). Saga patterns via compensating transactions handle rollback; durable timers sleep for months. Skills Janitor does none of this. It is a CLI that inventories skills, subagents, and MCP servers across user, project, codex, and plugin scopes, then lets you swipe-delete waste in a TUI sorted heaviest-and-least-used first. It cross-references MCP servers against real session usage, flags duplicate skills via Jaccard similarity above 30%, reports token cost of always-loaded descriptions, and scans for prompt-injection heuristics. It deletes only MCP config entries (leaving .bak backups) and flags plugin-bundled skills for review rather than deleting them. One is production infrastructure; the other is local developer hygiene.

Temporal is freemium: an open-source path plus paid Cloud tiers, with the enterprise-focused features listed as best-fit signals — SAML SSO, audit logging, SOC 2/HIPAA, and an SLA-backed support path. The static facts don't publish plan prices, so budget for a Cloud conversation if you need those controls; self-hosting trades license cost for operational effort. Skills Janitor is free and MIT-licensed, installs via plugin marketplace, npx skills add, or git clone, needs only Bash, Python 3, and curl, runs offline, and collects no telemetry. There is no paid tier and, per v1.8, no roadmap of new features — the project is in maintenance mode. Comparing the two on price is meaningless because the cost structures answer different questions: Temporal asks what a failure of long-running state costs you, while Skills Janitor asks whether token waste in your skill deck is worth five minutes of swiping.