// CUSTOM SKILLS · CALLABLE WORK
A custom skill packages one job—its inputs, method, evidence, limits, and output—so an agent can perform it again without rebuilding the thinking from scratch.
01 / MODEL
The command is only the handle. The useful part is the behavior behind it.
A repeatable function starts when it is called, follows the same evidence contract, and returns a named result. It can be used at any point when its inputs are available.
Example: inspect the live surface and return the exact source that owns it.A bounded loop runs the function again after a schedule, change, or failed check. It stops on a clear success, risk, time, cost, or approval condition.
Example: audit, repair, verify, and recheck until every owned route passes.02 / CATALOG
These examples expose the reusable shell while keeping the underlying project context private.
/seo-cluster-auditAUDIT LOOPCompare queries, existing URLs, search intent, overlap, content gaps, internal links, and available evidence before assigning one owner to each opportunity.
Returns keyword-owner map · cannibalization risks · ranked action queue/live-truth-checkFUNCTIONTrace the public result through redirects, headers, deploy state, revisions, source files, generated output, and cache identity before any edit begins.
Returns live surface · owning source · safe edit point · proof window/content-refreshMAINTENANCE LOOPReview claims, sources, freshness, metadata, structured data, internal links, and the page’s current search job; change only what new evidence supports.
Returns keep/change ledger · source-backed edits · next review trigger/mobile-site-auditQUALITY LOOPInventory every route and check real viewports, overflow, tap targets, navigation, modals, forms, keyboard behavior, accessibility, images, and performance.
Returns route matrix · severity-ranked findings · verified fixes/release-verificationBOUNDED LOOPBuild, inspect the blast radius, test a hosted preview, release to the intended target, and independently verify the live user path. Stop on any red gate.
Returns release verdict · deployed identity · live evidence · rollback point/handoff-packetFUNCTIONCompress current state, decisions, evidence, scope boundaries, unresolved risks, and the smallest safe next action into a continuation another agent can trust.
Returns state · decisions · proof · stop conditions · next move03 / CONTRACT
A memorable slash command is useful. A visible evidence and exit contract is what makes it dependable.
The request, schedule, change, or failure that starts the skill.
The exact sources, state, and authority the skill is allowed to use.
The ordered checks and actions that make the work reproducible.
The observable proof required before the skill may claim success.
The success, approval, risk, time, cost, or rollback condition that ends it.
Templates make the inputs and outputs concrete. The Library explains the methods behind the callable work.
Search pages, field notes, deep dives, skills, and templates. Use arrow keys, then Enter.