Conciseness

The body is lean and command-first — nearly every section is executable code or a config with one-line rationale (e.g., "if step 3 of 5 fails, only step 3 retries"), assuming Claude's competence throughout. It misses score 5 because of minor trimmable material: the service-naming gotcha is explained twice (Tier 3 block and Gotchas #3), and the time-sensitive "As of Feb 2026" date stamp sits in the main flow rather than a deprecation/notes section; it is well above score 3's "some unnecessary explanation".

Actionability

Almost everything is copy-paste executable: docker run commands, a full k8s StatefulSet+Service manifest, a six-step typed worker, a curl test event, and a launchd plist. It falls short of score 5 because the Step 2 client example uses `new EventSchemas().fromRecord<Events>()` without importing `EventSchemas`, so that file fails as written; the curl example's `/e/key` endpoint also assumes an event-key setup not wired to the earlier `.env.inngest` values.

Workflow Clarity

The process is clearly sequenced (intent-alignment questions → tiered setup → worker steps 1–6 → patterns → persistence → gotchas) with an explicit Verification checklist including a failure-recovery test ("Kill the worker mid-function — restart worker, function resumes from last step") and reboot checks — matching the score-5 anchor's validation steps, feedback loops, and checklists. Error messages for failure modes ("Error: signing-key is required", "Cannot find module '@inngest/ai'") give concrete recovery guidance.

Progressive Disclosure

Section headers are clear, but the skill is a ~430-line monolith with no bundle reference files (references/, scripts/, and adrs/ do not exist), while the body links to nonexistent paths like [ADR-0010](/adrs/0010-system-loop-gateway) and defers to an external joelclaw.com setup script — content that belongs in separate files (k8s manifest, launchd plist, patterns, ADR context) is inlined and navigation is partly dead. This matches the score-3 anchor ("references present but not clearly signaled; content that should be separate is inline") and is above score 2 only because the section structure itself is well-organized.

Specificity

The description names concrete capabilities — "Set up self-hosted Inngest on macOS as a durable background task manager for AI agents" and "from Docker one-liner to full k8s deployment with persistent state" — with "Interactive Q&A to match intent" describing the delivery mechanism. It stops short of score 5 because it omits several capabilities the body actually covers (building a typed worker, event chaining, cron functions, reboot persistence), and is above score 3 because multiple specific actions and deployment tiers are explicitly stated rather than generic.

Completeness

It explicitly answers both questions: what ("Set up self-hosted Inngest on macOS as a durable background task manager for AI agents... from Docker one-liner to full k8s deployment with persistent state") and when ("Use when: 'set up inngest', 'background tasks', ..."), with concrete trigger phrases — matching the score-5 anchor exactly. Voice is third-person imperative, so no voice penalty applies.

Trigger Term Quality

Six quoted natural phrases — 'set up inngest', 'background tasks', 'durable workflows', 'self-host inngest', 'event-driven functions', 'cron jobs' — plus the catch-all "any request for a local workflow engine" give good coverage of what a user would actually say. It falls short of score 5 because common synonyms like "job queue", "task queue", or "scheduled tasks" are missing, but clearly exceeds score 3's "missing common variations" bar.

Distinctiveness Conflict Risk

The named tool (Inngest) and "self-host inngest" trigger give it a clear niche with minimal conflict risk, but generic triggers like 'background tasks' and 'cron jobs' could overlap with other job-queue or scheduling skills — matching the score-4 anchor ("mostly distinct; minor overlap risk with closely related skills") rather than score 5's minimal conflict.