The work came first.
Each substantial skill begins with a procedure that had to be understood, repeated, corrected, and made legible through actual use.
Brownsmith Dynamics publishes agent skills developed from procedures used in real projects. They preserve the sequence, judgement, checks, exceptions, and handoffs that make a workflow dependable, then make that knowledge easier to reuse.
From practice to portable procedure
Written steps are easy to copy. What is harder to preserve is the reasoning: what evidence is sufficient, which exception changes the route, when approval is required, and how to recover when ordinary work does not follow the ideal path.
Each substantial skill begins with a procedure that had to be understood, repeated, corrected, and made legible through actual use.
A dependable procedure explains why a decision is made, what evidence supports it, and when the context calls for a different route.
Good instructions name missing evidence, unsafe actions, approval gates, failure states, private information, and the conditions under which the agent should stop.
Several skills carry validators, schemas, test fixtures, provenance, version-aware references, revision history, or rollback guidance so the method can be inspected and improved.
Published skill library
Each card links to the original GitHub source. Review the instructions, supporting files, dependencies, permissions, and maintenance assumptions before adopting any agent skill.
Plans evidence-grounded multichannel campaigns from one controlled source of truth, then produces genuinely channel-specific assets instead of resized copies of one post.
What makes it operational
Includes campaign state, platform guidance, claim controls, revision history, schemas, and deterministic validation scripts.
Builds a persistent, plain-file editorial memory for an organisation, including positioning, audience, voice, content strategy, claims, and unresolved decisions.
What makes it operational
Separates confirmed, inferred, provisional, and unresolved guidance so an agent does not silently turn an assumption into brand policy.
Reduces unnecessary words while preserving meaning, evidence, technical distinctions, uncertainty, safety warnings, and the writer's intended tone.
What makes it operational
Treats brevity as an editorial decision rather than a mechanical shortening pass.
Diagnoses the audience, problem, awareness level, proof, objections, and central idea before drafting credible marketing and sales content.
What makes it operational
Routes work through context, strategy, drafting, editing, and quality control rather than beginning with generic conversion copy.
Designs, writes, personalises, compiles, and reviews reusable email systems for campaigns, newsletters, outreach, lifecycle sequences, and transactional messages.
What makes it operational
Combines MJML components and rendering tools with accessibility, deliverability, consent, compliance, testing, and measurement guidance.
Creates role-specific CV, cover-letter, recruiter-outreach, screening-answer, and application-review material from supplied evidence.
What makes it operational
Protects the boundary between source facts, approved claims, inferences, and gaps while improving ATS coverage and human readability.
Audits and improves Next.js sites across crawlability, metadata, structured data, performance, rendering, accessibility, analytics, and AI-search visibility.
What makes it operational
Detects the installed Next.js version and requires its bundled documentation, repository adaptation, focused changes, and production-like validation.
Studies the editing language of a supplied or researched short-form video, then translates it into an original and reusable Remotion system.
What makes it operational
Separates visual analysis, rights, template contracts, replaceable media, and rendering so the result is not a hard-coded clone.
Audits and improves existing skills when documentation, APIs, policies, observed failures, or better workflows make them stale or incomplete.
What makes it operational
Requires evidence, a scoped proposal, explicit approval, narrow edits, regression checks, provenance, and rollback before a skill changes.
Runs technical, copy, or combined website audits with representative-page selection, reproducible evidence, practical repair guidance, and clear confidence levels.
What makes it operational
Preserves raw Lighthouse and audit evidence, separates observed facts from likely causes, and validates structured findings before handoff.
Designs, audits, writes, and validates SEO across indexability, crawling, metadata, structured data, migrations, performance, measurement, and AI search.
What makes it operational
Starts with catastrophic search-risk and page-quality gates, then applies an explicit indexability matrix before pursuing keywords or marginal scores.
Discovery and evaluation
Agent-skill aggregators make a growing ecosystem searchable. Once something looks relevant, read the original instructions, supporting files, permissions, dependencies, and maintenance history before deciding whether it fits the work.
The open format reference for SKILL.md metadata, instructions, scripts, references, assets, and progressive disclosure.
A cross-agent directory and install ecosystem with searchable skills, rankings, packs, and security information.
A searchable marketplace for community-published skills with category, creator, language, and update filters.
A plain-language directory that links readers to skills from their original publishers.
Before installation
A skill can shape what an agent reads, writes, runs, changes, or recommends. Treat it with the same seriousness you would apply to a dependency or an operating procedure.
Start from the source