Workflows Resources

Multi-step processes for agents. Check each listing's source, license and requirements before use.

This directory includes skills from OpenAgentSkills and AI-generated templates supplied by SimplyUtils, alongside community submissions. Check each listing's source, license, compatibility and permissions before use. Listed compatibility describes intended use; it does not establish that a resource has been tested in every named app. Missing source or license details still require verification. The AI-generated templates are examples; package availability, successful operation and redistribution terms have not been verified.

Choosing and checking a resource

A workflow template describes a sequence of work and review decisions; it is not a running automation or evidence that a repository has passed its gates. This directory groups six SimplyUtils AI-generated examples covering evaluation, API contracts, release notes, incident investigation, feature work and migration. Choose one by the task you actually need to complete, then open its full instructions and inspect the worked example. A familiar title can help discovery without proving that the listed commands, packages or assumptions fit your project. Keep the intended outcome and the missing evidence separate before asking an agent to apply a workflow.

For a concrete directory exercise, search for OpenAPI while viewing workflow resources. The API contract and code generation listing should remain available because its title identifies that subject. Open it and compare the static type-generation discussion with the separate runtime payload example. A search result is a metadata match, not an installation check or proof that a package can validate incoming JSON. Clear the query to restore the complete workflow list. If no result appears, inspect the selected resource type and category before treating an empty search as evidence that a capability is supported or unsupported.

Select the evaluation workflow when you need to define cases, scoring criteria and release observations for a model-dependent change. Its local exercise combines supplied scores and critical-failure flags; it does not grade a response or call an AI provider. The useful review question is whether the evidence covers the intended failure, not whether a generic aggregate threshold looks impressive. Keep deterministic assertions, human judgment and model judging distinct, and record unavailable connected evidence explicitly. A passing arithmetic gate cannot establish zero hallucinations, complete prompt-injection resistance or consistent behavior across every input a production model might encounter.

Select the API workflow when a request or response contract needs to agree across documentation, implementation and callers. Generated TypeScript types can improve static checking while leaving runtime JSON unvalidated. Inspect the selected tool output before importing a schema that may not exist. The local title-payload exercise shows one explicit grammar and invalid-input policy, rather than creating an endpoint or changing a database. Confirm success and error shapes against the existing application. A directory compatibility label describes intended use; it does not show that a generated client, middleware and real server have been tested together.

Select the release workflow to prepare a version proposal and accurate changelog from reviewed included work. The local helper increments a stable version tuple after a person chooses the change category. It neither infers compatibility from commit labels nor publishes a package. Check reverted changes, feature flags, migration notes and the actual artifact before describing a feature as available. A successful Git push, a completed build and a verified production behavior are different observations. Keep publication as an authorized action following a concrete review, and preserve the distinction between proposed and released version information in the final handoff.

Select the bug-triage workflow to organize a reproduced symptom, causal evidence, repair check and affected validation. It does not determine severity or infer the true cause from a stack frame alone. Select the feature-delivery workflow when a bounded user task needs acceptance conditions, relevant implementation and visible-state checks. Its checklist model allows a justified exclusion only where the local contract permits one; a failing observation cannot be dismissed by adding a reason. In both cases, inspect actual source and evidence before accepting state labels. The templates cannot give an agent permissions or turn a generated plan into owner approval.

Select the migration workflow when old and new representations need a defined compatibility comparison. Its owned catalog example preserves false values, compares identities and reports missing, added and changed records. Row order is intentionally outside that small contract, which may be inappropriate for an ordered API. A production dual run can duplicate writes, emails or paid calls, so do not assume that executing both paths is harmless. Identify a controlled comparison boundary and review side effects separately. A set of matching sample outputs is supporting evidence for those cases, not proof that all traffic or every important client is covered.

Read each listing before downloading or placing it in an assistant rule directory. These workflow templates were generated by an AI agent using an LLM and are supplied by SimplyUtils; they are distinct from the separately attributed upstream skills. Their redistribution terms remain unconfirmed, while named third-party tools and documentation retain their own terms. The directory and resource pages remain ad-free under the existing placement rules. A download is text to inspect and adapt, not a complete project with every dependency, script or integration it mentions. Keep an unmodified copy when comparing a revised workflow with the original assumptions.

Opening this directory does not run its workflows, read a repository or send project data to a model. If you use another provider or connected tool, review what files, prompts and logs it can receive and which actions it can perform. Start with owned public examples, remove unnecessary secrets and personal records, and test the smallest relevant path in a disposable environment. If instructions fail, report the listing URL, actual application version, missing requirement and a sanitized reproduction through SimplyUtils support. Preserve blocked observations and uncertainty instead of inventing a successful command, verified credential or professional endorsement to complete the template.