Reusable Workflow Library

Agent workflows treated not as one-off prompts but as durable, shareable artifacts — discovered, named, cataloged, versioned, and published so any person or agent can reuse a workflow that worked once without rebuilding it from scratch. A published workflow (a "loop") carries a fixed, self-describing anatomy so it is portable: Use when (the problem it solves), Prompt (the copy-ready instruction), Verify (the evidence that proves success), Steps (the feedback cycle in readable form), Notes (limits, risks, setup), and Related (nearby workflows). A companion skill supplies the lifecycle around the catalog — discover recurring work, find an existing fit, audit and repair weak checks or unsafe actions, adapt to local tools and limits, craft new loops by interview, run in bounded passes, debrief receipts, and prepare drafts for publication. The catalog is deliberately machine- consumable (JSON / plain text / an llms.txt agent guide) so an agent can browse and use it directly, without the companion skill installed.

Claims

Agent workflows are reusable artifacts, not one-off prompts — worth naming, cataloging, and sharing so they can be reused without rebuilding. principle — durable: the value of a good workflow is realized only when it can be reapplied; treating it as an artifact is what unlocks reuse across people and agents.
A shareable workflow needs a self-describing anatomy: use-when, prompt, verify, steps, notes, related. best practice — context: making a workflow portable across people/agents; the anatomy is what makes it discoverable and safely reusable, and it is contingent on wanting shareability (a private one-off does not need it).
Expose the catalog in machine-readable form (JSON / plain text / llms.txt) so agents can consume it directly. best practice — context: catalogs meant for agent consumption, not only human browsing; contingent on treating agents as first-class readers of the library.
Before publishing, check the live catalog for overlap and require explicit human approval of the exact preview; owner actions default to a draft. best practice — context: contributing to a shared catalog; the overlap-check plus approval-gate guard catalog quality and prevent unauthorized or duplicate publication.
Auditing a workflow ("loop doctor") repairs only material problems — weak checks, unsafe actions, unclear stopping. best practice — context: maintaining catalog quality; minimal-change repair avoids churn and preserves what already works.
A local task runner (e.g. a justfile) can serve as the reusability front-door — one place cataloging every command and its parameter permutations so you, your team, and your agents can discover and invoke them. best practice — context: a personal/team codebase of slash commands and scripts (vs. a published cross-org catalog); a second independent source (IndyDevDan's Bowser) uses just as the top layer over skills/sub-agents/commands. Contingent on the team already standardizing on such a runner.
Distribute the cataloged agentics by reference through a single manifest, not by copying — one source of truth every device, teammate, and agent syncs and can push edits back to. best practice — context: sharing agentics across 10+ codebases/devices/agents; a second independent source (IndyDevDan's "the library" meta-skill) implements this as a package.json-style YAML of pointers with add/use/push/sync commands. Contingent on a trusted single source that always wants latest (no versioning). See Agentic Distribution.

Agent Loop — the artifact the library catalogs, adapts, and distributes.
Agentic Workflow Patterns — the underlying workflow shapes the cataloged loops instantiate.
Agent-Computer Interface (ACI) — a machine-readable catalog (llms.txt / JSON) is itself an ACI for the library: agents consume the catalog as a tool surface.
Spec-Driven Development — the same move one altitude up: share the definition (spec), not the rebuilt implementation ("if agents are good enough, do we need to share code?").
Meta-Prompt — a reusable prompt/plan generator (a slash command like plan_w_team.md) is a concrete workflow artifact deployed over and over.
Agentic Distribution — the distribution/sync layer for a library of agentics: a reference-only manifest + meta skill that keeps skills/agents/prompts in sync across devices, teams, and agents.
Layered Agentic Architecture — the top two layers (commands cataloged by a just task runner) are a personal reusable-workflow library; "solve a class of problems once" is the same thesis at the codebase scale.
Distillate: Forward-Future/loopy — a catalog and skill for bounded, reusable agent loops.
Distillate: My 4-Layer Claude Code + Playwright CLI Skill (Agentic Browser Automation & UI Testing) — just as the reuse front-door; templating your engineering to solve a whole class of problems.
Distillate: Learnings from a No-Code Library: Keeping the Spec-Driven Development Triangle in Sync — "a library without code" is a workflow-library argument extended to application code.
Distillate: Claude Code Task System: Orchestrating a Team of Agents Through a Task Graph.
Distillate: The Library Meta-Skill: How I Distribute Private Skills, Agents and Prompts — the library as a reference manifest of pointers; sync/push keeps one source of truth across surfaces.
Distillate: Anthropic's Claude Cookbooks — the canonical recipe index — the vendor-official recipe catalog (registry + CI-validated notebooks); the first-party analogue of a personal loop library.

Reusable Workflow Library

Claims

Related

Linked from