ghl-memberships
Build and operate GoHighLevel Memberships via the internal API — courses, chapters with drip/gating, lessons (text, video, audio, PDF, embed), quizzes with questions, assignments, offers, themes, credentials, enrollment, progress and submissions, plus community groups. Use when the user asks to build a GHL course or membership site, create an onboarding/client portal, add quizzes or assignments, enroll or revoke a member, issue a certificate, collect learner-submitted documents, or read course progress.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 12
- Forks
- 5
01 前言 Front Matter
- name
- ghl-memberships
- description
- Build and operate GoHighLevel Memberships via the internal API — courses, chapters with drip/gating, lessons (text, video, audio, PDF, embed), quizzes with questions, assignments, offers, themes, credentials, enrollment, progress and submissions, plus community groups. Use when the user asks to build a GHL course or membership site, create an onboarding/client portal, add quizzes or assignments, enroll or revoke a member, issue a certificate, collect learner-submitted documents, or read course progress.
02 资源文件 21
- 脚本engine/api.mjs13.2 KB
- 脚本engine/api.test.mjs7.0 KB
- 脚本engine/assessments.mjs9.3 KB
- 脚本engine/assessments.test.mjs8.4 KB
- 脚本engine/auth.mjs3.8 KB
- 脚本engine/communities.mjs6.0 KB
- 脚本engine/communities.test.mjs1.0 KB
- 脚本engine/course-builder.mjs25.2 KB
- 脚本engine/course-builder.test.mjs12.3 KB
- 脚本engine/credentials.mjs6.2 KB
- 脚本engine/members.mjs8.2 KB
- 脚本engine/members.test.mjs5.5 KB
- 参考文档references/course-spec.md5.3 KB
- 参考文档references/example-spec.json3.7 KB
- 脚本scripts/build-course.mjs3.7 KB
- 脚本scripts/build-course.test.mjs606 B
- 脚本scripts/cli-gateway.mjs2.6 KB
- 脚本scripts/cli-gateway.test.mjs1.7 KB
- 脚本scripts/conformance-flags.mjs637 B
- 脚本scripts/conformance-flags.test.mjs1.7 KB
- 脚本scripts/conformance.mjs25.0 KB
03 SKILL.md 正文
GHL Memberships Builder
MCP routing: If the
uxie-ghl-internal-mcpserver is registered in this session, prefer itsbuild_course/list_coursestools over running this skill's scripts directly — the tools wrap this same engine behind confirmation gates and post-build verification. Fall back to this skill's own scripts when the server is not registered.
Writes to a GHL account via the undocumented internal Memberships API.
Before any write
- Run BOTH gates in
${CLAUDE_PLUGIN_ROOT}/docs/write-rails.md. - Auth:
${CLAUDE_PLUGIN_ROOT}/docs/auth-jwt-capture.md— §8 is the Memberships section. The admin rail is §1 plus asourceidheader; the member rail is a different token class. Memberships auth is looser than the workflow builder — no iframe origin, noorigin/referer. Do not port the workflow CORS handling here.
Fastest path: compile a whole course from a spec
GHL_TOKEN='<LC JWT>' GHL_LOC='<locationId>' \
node ${CLAUDE_PLUGIN_ROOT}/skills/ghl-memberships/scripts/build-course.mjs <spec.json> [--dry-run]
--dry-run validates the spec without touching the account — always run it first.
Spec format + a worked example: references/example-spec.json and references/course-spec.md.
One command produces: course → chapters (order, drip, nesting, gating) → lessons (rich HTML body,
video/audio upload, PDF attachments, external embeds) → quizzes with questions → assignments →
free offer (draft; published only when the spec says publish: true) → branded theme → credential template attached to auto-issue on completion
→ optional enrollment. Then it verifies every item by read-back and prints a teardown command.
Remove: node scripts/build-course.mjs --delete <productId>
⚠️ Offers and credential templates do NOT cascade — delete those separately.
Prove the API hasn't drifted
GHL_TOKEN='<LC JWT>' GHL_LOCATION='<locationId>' node ${CLAUDE_PLUGIN_ROOT}/skills/ghl-memberships/scripts/conformance.mjs
Runs the real admin lifecycle live and asserts on effects, not status codes (grant → member appears → revoke → member gone), re-checks the documented traps, and tears down. 21 passed / 0 failed / 4 skipped. This is an undocumented API that drifts — run this before trusting the reference on any account you haven't touched recently. The 4 skips are member-session writes that cannot run unattended (see Limits).
Which rail, per job
The public Courses API (2026-09-24) returned the same courses as the internal rail on the sandbox (2026-10-29).
<!-- rail-ledger:courses -->
generated from the rail ledger (mcp-internal/catalog/rail-ledger.json) — do not edit; the whole ledger is ghl-orientation/references/rail-ledger.md.
| Job | Decision | Routed today | Internal | Public |
|---|---|---|---|---|
| read | pool-internal | internal | list_courses |
courses-v3__list-courses, courses-v3__get-course, courses-v3__list-categories, courses-v3__list-lessons, courses-v3__get-lesson, courses-v3__list-offers |
| create | undecided | internal | build_course |
courses-v3__create-course, courses-v3__create-category, courses-v3__create-lesson, courses-v3__create-offer |
| edit | undecided | public | raw_request:PUT /membership/locations/{locationId}/products/{id}, raw_request:PUT /membership/locations/{locationId}/categories/{id}, raw_request:PUT /membership/locations/{locationId}/posts/{postId} |
courses-v3__update-course, courses-v3__update-category, courses-v3__update-lesson, courses-v3__sequence-categories, courses-v3__sequence-lessons, courses-v3__update-offer |
| delete | undecided | public | raw_request:DELETE /membership/locations/{locationId}/products/{id}, raw_request:DELETE /membership/locations/{locationId}/categories/{id}, raw_request:DELETE /membership/locations/{locationId}/posts/{postId} |
courses-v3__delete-course, courses-v3__delete-category, courses-v3__delete-lesson, courses-v3__delete-offer |
| enrol | undecided | internal | raw_request:POST /membership/smart-list/attach-offer-user |
courses-v3__create-enrollment, courses-v3__delete-enrollment |
| bulk | undecided | internal | raw_request:POST /membership/locations/{locationId}/assessments/quiz/questions |
courses-v3__get-quiz-questions, courses-v3__upsert-quiz-questions |
| logs | undecided | internal | raw_request:GET /membership/locations/{locationId}/products/user-progress/{productId} |
courses-v3__list-bulk-progress, courses-v3__list-enrollments, courses-v3__review-assessment |
- read: The memberships skill builds and edits on the internal rail; the public Courses API (2026-09-24) has not been proven on our sandbox yet.
- create: The memberships skill builds and edits on the internal rail; the public Courses API (2026-09-24) has not been proven on our sandbox yet.
- edit: D3 (Xander, 2026-11-06): course editing and deletion are public-only. Sequence and delete were ruled pool-public (orchestrator DECISION, w8-1); the public proofs are in [private capture] and delete.json. The tools below are the internal paths the ruling retires.
- delete: D3 (Xander, 2026-11-06): course editing and deletion are public-only. Sequence and delete were ruled pool-public (orchestrator DECISION, w8-1); the public proofs are in [private capture] and delete.json. The tools below are the internal paths the ruling retires.
- enrol: The memberships skill builds and edits on the internal rail; the public Courses API (2026-09-24) has not been proven on our sandbox yet.
- bulk: The public upsert replaces every question and drops option text (sandbox 2026-11-06); the internal batch add is the one proven on the build path.
- logs: Not run on either rail: needs an enrolled learner and a submission (G4). Grading is observed only in the engine.
Internal only: media, DRM, themes; clone; certificates; analytics. Public only: synchronous enrolment that says alreadyEnrolled; progress and review; import. <!-- /rail-ledger:courses -->
Object model
Product (course) → Category (chapter, self-nesting) → Post (lesson)
Offer = access gate Theme = per-product design tokens
Credential template + attachment = certificate/badge, auto-issued on completion
A post carries rich-HTML description (the body) + ONE primary media + N post_materials.
Engine API (prefer this over hand-rolling requests)
| Module | Use for |
|---|---|
engine/api.mjs |
products, categories, posts, video/audio upload, PDF materials, offers, theme, setEmbed, parseIframe |
engine/members.mjs |
invite, magicLink, grantOffer/revokeOffer, publishOffer, productProgress, allMembers, waitForEnrollment |
engine/assessments.mjs |
quiz + addQuestions, assignments, listSubmissions, grade |
engine/credentials.mjs |
template CRUD, issue, listIssued, attach (auto-issue on completion) |
engine/communities.mjs |
admin group CRUD + CommunitiesMember (portal rail: join, createPost) |
Traps that will cost you hours
contentTypeis a CLOSED MySQL ENUM:video | audio | quiz | assignment— or OMIT it. GHL has notextlesson type. A text lesson is avideo-typed post with the text in its description: sendcontentType: 'video'explicitly and put the text indescription. Do not read avideoread-back on a text lesson as a bug.text/embed/html/pdf/file/linkall return500 Data truncated for column 'content_type'. There is no embed or PDF lesson type.- Embeds:
POST /postsSILENTLY DROPSembedJson. It returns 200 and the embed is simply absent on read-back. Create the post, then PUT it (api.setEmbed()). The server mintsmetaData.embedMediaIditself — you never create that id, which is why no "embed media" endpoint exists. quizId !== postId.POST /assessments/quizcreates it; re-read viagetQuizByPost(postId)to get the quiz's own id, which is what the questions endpoint wants.- Auto-issue a certificate is a course-level ATTACHMENT, not a field on the lesson
(
credentials.attach()→certificate-attachments,eventType: product_complete). - The category tree is a PARTIAL projection —
GET /categories?posts=trueomitspost_materialsandasset_urls. Verify lessons withGET /posts/{id}or you get false negatives. - Theme needs TWO calls: PUT the theme, then PUT
products/apply-theme— saving alone does nothing. user-purchase/no-of-userslies about enrollment — it counts purchases, returns 0 for an admin-attached member. Verify withproducts/user-progress. 7b. The grant's own 200 lies too, and it is the worse of the two.POST /membership/smart-list/attach-offer-useranswers200 {ok:true, msg:"…successfully queued"}for an empty body, for fabricated ids, and for the wrong key names — proven on the test sub-account, with a nonexistent sibling path answering 404 as the control. Two operators have been told a grant was queued and granted nothing. The body is{contactId, offerId, source:'admin'}— singularcontactId, nolocationId(that rides on thesourceidheader). Then read back:members.waitForEnrollment(productId, {expectContactIds}), which looks for those contacts. A non-empty check is not a check — it passes on a member who was already there.- Deletes don't cascade to offers, credential templates, or assessment submissions. Submissions have no delete route at all.
- assets-drm wants a LEADING SLASH on the
paththatsigned-url/uploadreturns without. Client suppliesdurationInSeconds(ffprobe). - Community group list reads are eventually consistent (~1.5 s) — don't assert a write failed on one immediate re-read.
- A UI Save that "fires no request" is usually a lie. Twice on this surface it was a
product-tour overlay (
fd-tour-overlay, Pendo#pendo-base) swallowing the click, or a too-narrow network filter — GHL splits one feature's read and write acrossmembership/,courses/,services/andclientportal-middleware/. Re-check with a host-level filter before concluding a flow is client-side. createGroupsendsmarketplaceLocationId(the UI's Discovery switch ON) unlessdiscovery: false(2026-10 switch, default ON as in the UI; OFF omits the key — that the group then stays off the marketplace is the switch's stated purpose, not separately observed). Ask the operator which they want for a private client community.- Moderation keywords live on the ADMIN rail too — group rows carry
settings.moderationSettings.blockedKeywordsand the admin group PUT writes them (the portal read then shows them, proven 2026-10-30). Read the row and sendsettingsback whole with your change. Owner and admin posts are never held, so testing a keyword as the owner proves nothing; the held queue and member reports are portal-only (403). - Notification preferences are one document per member per group; the portal UI's save also turns the
member's
newsletterpush OFF (no push control on that row) — do not copy the UI's body.
Proof levels — know what you're trusting
- EXECUTED (built and sent from code, effect asserted): everything the conformance suite covers, plus contacts, media rails, themes, embeds, credential attach.
- OBSERVED only (transcribed from the UI, never re-issued independently):
assessments.grade(), learner submission,user-post-completion. All member-session-bound; each was driven live through the real portal and verified by read-back, but a field the UI sends and we failed to record would only surface on first use.grade()is marked in-source.
Limits (state these rather than working around them)
- Paid offers are supported by the engine but need a connected payment integration (bl-877). On a sandbox with none, a paid create returns
400 Integrations not found. The enum isonetime(orsubscription);one_timeandrecurringreturn 500. Checkout is unproven. Onlyfreeis proven end to end. - Members cannot be fully provisioned headlessly — portal signup is OTP-gated. See auth §8.3 for the softer path (the user record exists before OTP verification).
- Whether the credential attachment FIRES on completion is unverified — it's created and reads
status: active, but no member has completed a course to watch it issue. - Module-level credential attachment (
altType: "category") is implied by the response shape, not verified —attach()only exposes product level. - Community channel create/update, and pending-invite list/revoke (403), are uncaptured.
Editing a course (public rail, ghl-multi)
Course editing is public-only (Xander, D3, 2026-11-06). Use execute_action on the ghl-multi server with these action ids. Every
write takes dry_run: true first; read the preview; then send the same call with confirm: true where the server asks.
After every write, read the course back with the internal get_course (one read: title, offers, normalised tree) and compare.
| Job | Action | Notes |
|---|---|---|
| change a lesson | courses-v3__update-lesson {productId, lessonId, …fields} |
partial body: only the fields you send change (measured 2026-11-06) |
| change a chapter | courses-v3__update-category {productId, categoryId, …fields} |
partial body (measured) |
| change the course | courses-v3__update-course {productId, …fields} |
partial body (measured) |
| reorder chapters | courses-v3__sequence-categories {productId, ids} |
ids lists every chapter in order (measured) |
| reorder lessons in a chapter | courses-v3__sequence-lessons {productId, categoryId, ids} |
ids lists every lesson of that chapter in order (measured) |
| move a lesson to another chapter | courses-v3__update-lesson with categoryId, then sequence-lessons on the new chapter |
two calls, see trap 3 |
| remove a lesson | courses-v3__delete-lesson {productId, lessonId} |
204; the lesson then 404s (measured) |
| remove a chapter | courses-v3__delete-category {productId, categoryId} |
cascades: its lessons go with it (trap 2) |
| remove a course | courses-v3__delete-course {productId} |
204 (measured on a probe product) |
| add a lesson or chapter | courses-v3__create-lesson / courses-v3__create-category |
create as visibility: "draft" unless the user has asked to publish |
Traps (the first three are measured on the sandbox; the rest are from the same runs):
- A reorder that names only some lessons is not measured. Always send every lesson of the chapter in
ids. What the server does with unnamed lessons is unknown, so do not rely on them keeping their place. - Deleting a chapter deletes its lessons. Before
delete-category, list the chapter's lessons (get_course) and say so to the user. The internal route also cascades (measured), so this is not a rail difference. - A cross-chapter move is two writes, and the lesson keeps its old number.
update-lessonwithcategoryIdmoves the lesson but does not renumber it (measured). Follow it withsequence-lessonson the new chapter, naming every lesson in order, or the moved lesson can collide with an existing number. - Do not use the internal whole-document PUT for these edits. The internal routes take the full object (the engine sends it whole), so an edit there is a read-modify-write of the whole object. The public routes merge a partial body (measured).
- Deleting an offer reads back as a ghost.
courses-v3__delete-offerreturns 204, but a laterget-offerreturns 200 with an empty body. Verify an offer removal withget_course(the offers list), not withget-offer(bl-801). - Do not use
courses-v3__upsert-quiz-questionsto edit questions. It replaces the whole question set (measured: two questions gone when one was sent) and the option text is not read back. To add questions, useedit_quiz_questions(below). - Gated, do not run without the owner's go: publishing a course, chapter, lesson, offer or channel (the spec's
visibility: "published"or any*-publish); any paid offer; any enrolment or grant that e-mails a member (G4). A dry run of these is fine.
Granting and revoking access
- Internal:
set_course_access {locationId, contactId, offerId, action: grant|revoke, confirm}. Preview first (no account call;sendssays whether a grant e-mails the member:unknownfor a grant, not yet measured;falsefor a revoke). Confirmed, it reads the course membership, sends the change, reads it again, and reportsgranted,already-had,revoked,already-absentorqueued-unconfirmed. Trust the outcome, never the 200. - Public:
courses-v3__create-enrollment(201 new, 200alreadyEnrolled),courses-v3__delete-enrollment(revoke),courses-v3__list-enrollments(read back). - A draft offer still grants access. A public enrol into a free draft offer answered 201 and read back (bl-833). Do not treat an offer's draft
status as a guard. Never put anything confidential in an offer, and treat its id as public (an unauthenticated GET of
/membership/offers/{id}returns the draft, with the location and feature flags, bl-321). - A grant or enrol may e-mail the member. Every send is a gated action (G4) for a real student. Check the preview
sendsbefore confirming.
Progress and completion
- Internal:
get_course_progress {locationId, courseId, email?}walks every page and returns each student's course percent (completedPercentage), the user's e-mail andlastLogin, plus a per-course summary.complete: false(withfailure) means the walk did not finish: do not read a partial list as the whole course. The internal route gives no lesson counts, socompletedLessonsis null there. lastActivityon that tool is the last login, not the last lesson. Do not read it as lesson activity.- Public (ghl-multi):
courses-v3__list-bulk-progress {contactId, productIds}returnstotalLessonsandcompletedLessonsfor a student, andcourses-v3__list-category-progressandcourses-v3__list-completionsgive per-chapter progress and completions. - Caveat (bl-834): on the sandbox the public bulk route returned
totalLessons: 0for a student on a course with 6 lessons. The likely cause is that the lessons were draft (a count of published lessons only). This is untested: it needs a published lesson, which is a gated publish. Until it is settled, do not trust a0/0lesson count without checking the lessons' visibility first.
Quiz questions and grading
- Add questions in a batch:
edit_quiz_questions {locationId, quizId, add: [{title, questionType, explanation?, options: [{statement, isCorrect}]}], confirm?}.quizIdis the quiz id (the oneGET …/assessments/quiz/{postId}returns), not the lesson id. Preview first (no write). A single-choice question needs exactly one correct option; duplicate question text in one batch is refused before anything is sent. - The batch write replaces the whole question set (measured on the sandbox, 2026-11-07: a one-question add to a quiz holding one question left two,
and the stored one was gone). So a confirmed
edit_quiz_questionsre-posts every stored question in full, then the additions, and it is refused before any POST when a stored question's option text cannot be read. It reads the quiz again on a separate request and reportsaddedonly when that read matches the expected set by content (question ids change on every save, so they are not compared, and a question id kept from before an edit is stale); otherwiseunverified, with the full before list. A failed batch POST is read back the same way (the route replaces, so a failed POST may still have changed the set): the result carriespostFailedwith the HTTP status, and the after read. Itsreplaceargument is refused. Do not reach forupsert-quiz-questions(trap 6). - Grade a submission (public):
courses-v3__list-assessments {productId, status?, pageSize?, pageNumber?},courses-v3__get-assessment {productId, assessmentId}, andcourses-v3__review-assessment(PUT) with body{status: passed | failed | processing, score?, feedback?}. A grade is visible to the learner. Only a submitted attempt can be graded, so the grading route needs a learner who has submitted in the portal. - Not exposed: the internal
Assessments.grade()is observed only (transcribed from the admin UI) and needs a member session. Do not present it as a tool. - Not yet proven on the sandbox: a learner's quiz or assignment submission and its grade. That leg needs a portal session and is tracked on the backlog (bl-842), not run in this build.
Communities: paid channels and plans (know about)
- Announced 2026-11-02: paid channels and multiple pricing plans in Communities. Nothing here reads a paid flag, a plan or a price: the admin rail's group and location records carry no such key, and channel paid flags sit behind the PORTAL_USER scope.
- Every write is unproven. Creating a paid channel, creating or editing a pricing plan, and connecting a payment provider for Communities are all paid actions and gated. Do not run them to "test" the route.
- Know-about only: which surface shows what, and what is still open, is in the corpus page
70-research/2026-11-06-communities-paid-channels-and-know-about.md. - Course video analytics and the Discover Courses & Offers storefront have no route taught here; the courses analytics routes in the endpoint index are proven for dashboards, not for video watch data.
Routes with a proven raw_request write (settings, CSS and code, instructor fields, portal users)
These operations have no typed tool. Each one was proven on the sandbox (the write, then a read on a separate request, then a restore), so the
route and body shape below are evidence, not a guess. Send them with raw_request {locationId, method, path, body, confirm: true}. Preview first
where the tool offers it, and never send a body whose keys you have not read first: several of these writes replace the whole object.
- Course CSS, custom code and header/footer (product document):
PUT /membership/locations/{locationId}/products/{productId}with the FULL product object (read it withGETon the same path first). The custom fields arecustomCss,customJs,customHeader,customFooter. Proven: set to a TEST value, read back on a separate GET, restored tonull. Clearing a field back tonullworks; an empty string is not the same and is not used. - Instructor name, title, bio and heading (customization):
PUT /membership/locations/{locationId}/products/customizations/{customizationId}, with the customization record read first (GET /membership/locations/{locationId}/products/{productId}, thecustomizationsblock). Proven: name, title and bio set and read back, the heading read back unchanged, then restored. - Portal settings (portal banner text, app permissions, and the other portal-settings keys):
PUT /clientclub/portal-settings/{locationId}with the full document. The keys the GET carries must equal the read value, except the one test field. Keys the GET does not carry are sent only as the form's own defaults. Proven forportalBannerTextand for app permissions: set, read back on a separate GET, restored. Once a portal name is set the UI cannot clear it (an empty name is refused before any request), so a name is a one-way write on a sandbox. - Invite a user to the client portal:
POST /clientclub/{locationId}/users/invitewith{locationId, email: [address]}; returns201with{success: true}. Proven: the portal settings read (invitedUsers) went from absent to a count of 1 on two separate reads after the invite. The invite e-mails the address: send it only with an approval that names the address, and read the user back afterwards. - Course delete (irreversible; deletes the object):
DELETE /membership/locations/{locationId}/products/{id}. It deletes the course. Deleting a product does not delete its offers, so clean up the offers separately. Proven: the course is gone from the list on a separate read after the delete. Delete only a TEST-CONF course you created for the test. - Offer delete (accepted, not applied):
DELETE /membership/locations/{locationId}/offers/{id}returns 200{ok:1}, but the offer stays on the product record on a separate read (the course Offers list shows nothing). Do not rely on it to delete an offer: the delete is accepted but not applied, the same finding as the public delete-offer route. - Course comment (edit and delete):
PUT /membership/locations/{locationId}/comments/{commentId}edits a comment;DELETE /membership/locations/{locationId}/comments/{commentId}removes it. Read the comment first withGETon the same path and send the full object on the PUT. Proven: set and read back on a separate GET, then restored. - Lesson video (upload, then attach): the assets-drm upload first (
POST /assets-drm/assets/signed-url/uploadwith{source:"courses", entityId:"{locationId}", type:"videos", mimeType:"video/mp4"}, thenPUTthe raw bytes to the signed URL, thenPOST /assets-drm/assetsfor theassetsLicenseId). ThenPOST /membership/locations/{locationId}/videoswith{url, title, postId:"{lessonId}", assetsLicenseId, id}. Note: thepathreturned by the upload needs a leading slash before the assets call. - Course theme (save, then apply):
PUT /courses/locations/{locationId}/product-themes/{productId}/theme/{themeId}saves the fullthemeDatatree. Without the next step the course keeps its old theme:PUT /membership/locations/{locationId}/products/apply-theme/{productId}?template_id=NeoClassic(empty body) applies it. Proven: the UI save and the apply both returned 200; a replay of the theme save (weight bold, read back on a separate GET, then restored) was read back; the product customization was unchanged after the apply. - Community group (create):
POST /communities/{locationId}/groupswith{name, slug, description, locationId, marketplaceLocationId, branding:{}}returns 201. Creating a group does not list it on the Gokollab marketplace: the listing is a separate outward step and is not done here. - Portal notification preference (toggle, saves on change):
POST /clientclub/notification-preferences/{locationId}with{category, notifications:{EXPIRE_CERTIFICATE:{config:{email, sms}, email_template}}, locationId}; returns201. The page has no save button: unchecking sends the POST. Read the current value first on the page (the GET behind the page is not in the catalogue, so this row does not name it). Proven: the page read (200), the change (201), and the restore (201, same route, email set back to 1). - Member grade (assignment review):
PUT /courses/locations/{locationId}/assessments/{submissionId}/reviewwith{score, status:"passed"|"failed", feedback}; returns200with{error:false, updated:true}. Note the/coursespath, not/membership. Proven: the grade save returned 200, and an admin read of the same submission on a separate request showed it. - Course site details (title and the site-info save):
PUT /membership/locations/{locationId}/settings/site-info/{siteInfoId}carries the title. Proven: the title was set to a TEST value and read back on a separate GET, then restored to empty and read back. The same restore save also sentPUT /membership/locations/{locationId}/settings/domain/{id}andPUT .../settings/builder-settings/{id}with their values unchanged (200 each); those two routes are not proven as writes. - Course duplicate (clone; leaves a new course):
POST /membership/locations/{locationId}/products/clone/{productId}with{locations: [locationId]}returns the newproductIds. Proven: the clone was read back on a separate GET (its title carried the "- Clone" suffix). The clone is a new course, so it stays as residue on a sandbox; delete it only by its own id. Not covered here: the course analytics dashboards (their data comes from read-only summary queries that the catalogue does not carry yet), the course analytics progress view (no proven write), and learner-side rendering. Those stay as the corpus describes them.
Credential templates (certificates): list, create, designer save, delete
Proven on the sandbox with a read on a separate request after each write. The template is the credential's design; a credential is issued from it.
- List templates:
GET /certificates/locations/{locationId}/templates→{templates: [{_id, title, ...}]}. - Create a template:
POST /certificates/locations/{locationId}/templateswith{title, ...},confirm: true. Returns201. Read it back on a separate list GET (_idandtitle). Create only with an explicit approval naming the type. A created template stays as residue unless it is deleted with the route below. - Designer save (title and canvas fields):
PATCH /certificates/locations/{locationId}/templates/{templateId}with{locationId, title, ...}; send the body keys you change. Proven: the title change read back on a separate list GET. Rules measured on the sandbox: the endpoint refuses an emptytitle(send the unchanged title with every PATCH); the merge keeps keys you omit; the designer'stemplateJsoncannot be cleared back to an empty string through this route (a restore to""is stored as the two-character string"", not the empty value). Read the template first and send the exact before value for any key you restore. - Delete a template:
DELETE /certificates/locations/{locationId}/templates/{templateId}, with confirm. It returns200withtype: TEMPLATE_DELETED. Read back on a separate list GET: the id is gone. Delete only a template you created for the test, by its own id.
Full reference
references/course-spec.md — spec format.
Semantics and capture evidence: the corpus surface knowledge/corpus/memberships-courses/, the
source of truth for anything this file summarises. The endpoint index is search_endpoints
on the internal MCP: every /membership/*, /courses/*, /certificates/* and /clientclub/*
route, mined from the memberships front-end's own source — most never exercised here.
describe_endpoint gives the exact raw_request, and raw_request sends the sourceid this
surface's front-end pins. The engine modules above remain the way to WRITE: a catalogued route is
a path, not a proven write.
04 同仓库更多 Skills
ghl-workflow-specialist
Senior GoHighLevel workflow architect — designs automations (nurture, speed-to-lead, no-show recovery, internal ops, data hygiene, routing) grounded in the full trigger/action catalog and known anti-patterns, and reviews or fixes live workflows. Use when the user asks what automation to set up for X, wants a workflow designed, or asks why a live workflow is not getting through, is failing, or whether an A/B test is done. Recons + reads the client brief before proposing anything. Once approved, the build itself goes to create-ghl-workflow.
ghl-workflow-fast-forward
Fast-forward contacts parked at a GoHighLevel / HighLevel workflow WAIT step to the next step — the builder's "Action statistics → move to next step" control, driven via the internal API. Use to accelerate end-to-end workflow testing (drive a multi-day wait ladder to completion in minutes instead of waiting real time), to unstick contacts stranded at a wait, or when the user says "fast-forward this wait", "skip the wait", "move parked/stuck contacts to the next step", "requeue stuck contacts", "test this long delay without waiting days", or gives an app.gohighlevel.com workflow URL and asks to push contacts past a wait. This MUTATES live enrollments (it fires whatever steps come next, i.e. real SMS/email to real people) — it is a WRITE skill and follows the plugin's write-rails. It is the mutating counterpart to the read-only get-ghl-workflow-logs.
ghl-voice-ai
Build and configure GoHighLevel Voice AI phone agents — the prompt, the voice, the actions (appointment booking, call/agent transfer, contact-field extraction, SMS, workflows, custom API), knowledge bases, MCP servers, working hours, and outbound consent. Use when the user says 'set up a voice agent', 'build a phone agent', 'my AI should answer calls', 'book appointments over the phone', 'outbound calling', 'the agent isn't transferring calls', 'change the voice', or names a Voice AI agent. Internal API — the public rail exposes only how the agent sounds, not what it does.
ghl-system-conventions
House conventions for building GoHighLevel systems — workflow naming and structure, tag/field/custom-value discipline, pipeline and opportunity decisions, copy rules, and the pre-build approval document the operator signs off before anything gets created. Account-agnostic, for ANY GHL sub-account. Use whenever designing, building, editing, reviewing, or discussing GHL workflows, automations, pipelines, stages, opportunities, tags, custom fields, custom values, calendars, or AI agents — even if the user never says "conventions" or "standards". Load it BEFORE proposing a design or creating any GHL object, so the proposal already looks the way the operator wants it.
ghl-snapshots
GoHighLevel snapshots — capturing a sub-account's assets into a reusable snapshot, refreshing one, curating what it carries, and checking load conflicts. Use when a task involves creating or updating a snapshot, deciding what an account template should include, asking why a snapshot came out empty or is stuck processing, or planning a rollout onto other sub-accounts. Read it BEFORE calling any /snapshots path by raw_request — two of the write endpoints silently do the wrong thing.
ghl-reverse-engineering
Map a GoHighLevel internal surface exhaustively — enumerate every screen, panel, control and save path a feature has, capture the real request behind each one, prove what each field does by differential, and write the result into the corpus. Use when the user wants to understand how a GHL feature works under the hood, extend an engine to an object the public API can't reach, re-sweep a surface that already has partial coverage, or asks to reverse-engineer / sniff / capture / trace anything internal. GHL permits inspecting your own account's traffic. This is a MAPPING skill, not a lookup: it finishes when the surface is exhausted, not when the first endpoint is found.