Rankup is a full lifecycle project management Skill for indie developers: research and demand validation → project initiation and positioning → build a website or native App (macOS, iOS, iPad; no Android App, Android stores are only used as demand reference) → pre-launch SEO / GEO → launch and integration → backlink acquisition → monetization and monitoring iteration.
SEO and GEO are the primary means of customer acquisition and validation,not the applicable boundary: project plans, handoff documents, feature iteration, launch operations and maintenance, and project record organization all belong here.
Each project's .rankup/ records long-term plans and current progress, and is responsible for cross-session handoff; this file only serves as the entry point and routing, with details in references/, reading only the part needed for one task at a time.
Data that cannot be obtained is written as "unknown / unavailable", not 0; distinguish "official requirements", "tool suggestions", and "project experience"; a phased report does not equal completion.
Mandatory process (read this table first, then do anything)
For project tasks, read the relevant entry point and handoff according to the table below, then choose the task branch, to avoid executing based on outdated impressions; for purely read-only and lightweight tasks, follow the corresponding exceptions.
| Trigger condition | Must read first | Must do | Completion marker |
|---|---|---|---|
| Start work: project task (including workspace root) | Handoff protocol, project handoff record | Restore, reconcile, and record status according to the handoff and the startup protocol below | Handoff can continue the work |
| Milestone: artifact, decision, or background task status changes | Same handoff protocol as above | Overwrite the current handoff and artifact pointers | Handoff reflects the latest status |
| Wrap-up: completed, abandoned, stopped, or awaiting user decision | Maintain "One" and "Two" | Close out according to the lightweight path or the five-step maintenance; scan using scripts/maintain/ref-scan.mjs |
Reply has a maintenance summary and evidence |
| "Continue", "pick up where we left off", "where did we get to last time", "let's start executing this project's plan" with project record context, or reading the plan / handoff / progress documents | Same as "How to pick up in a new session" above | Locate the project → read the handoff (if there is no .rankup/ use HANDOFF.md) → verify background artifacts and changes → directly execute the next step |
The next step already has execution evidence |
| Keyword selection, research, finding demand (Already has a primary keyword） | playbooks/entry.md → research.md |
Execute according to the entry point's four-step sequence and two-path interpretation; if the direction has not landed on a primary keyword, first go through selection.md, if there is no direction, first go through P1, and return to this row after the primary keyword appears |
The entry card is written into .rankup/research/<词根>-<日期>.md |
| ChatGPT query chain, natural recommendation validation | seo-geo.md steps 2–3 |
Queries, sampling, follow-ups, and the three lists only follow that single source; keyword-level orchestration and adjudication return to entry.md |
Entry card ② has the three lists and original evidence pointers |
| "Google quality," "HCU," "E-E-A-T," "will AI content be hit," "content review," "spam policies" | seo-helpful-content.md |
Distinguish official positions from industry inferences; check against the single-page or site-level self-check list and keep evidence | Pass the corresponding self-check list with no red lines |
| Build the site, build features, write UI | lifecycle/stage-3-build.md、cloudflare-stack.md |
Use the user-specified template; if not specified, use the default shadcn command; UI comes from the component library, do not hand-write basic controls; finish the Day-1 checklist the same day; checkRed lines during construction | checklists.md Section 3 passed |
| Write page content, write articles, write blog posts, model pages, model comparisons, product comparisons, add reference videos, cover images | blog（blog/SKILL.md） |
Execute according to its division of labor and process (X research to Grok, drafting to Gemini, review to Sonnet), do not make up content yourself; all English on-site page content must go through /blog |
sources, drafting, review, judgment (Ge Fei scoring, AITDK, ChatGPT anonymous Q&A), and read-back evidence are all complete |
| Make inner pages, "turn this keyword into a page" | lifecycle/stage-4-prelaunch.md"New inner page quick checklist" |
Page mapping according to Section 2 §2.3, register the target keyword in keywords.md, complete the page and health check according to Section 4; checkRed lines during construction |
checks.md Record ✅ and evidence |
| Pre-launch health check, "can it go live" | checklists.md Section 4,lifecycle/stage-4-prelaunch.md |
Full health check: AITDK, PageSpeed, icon-specific check, placeholder-specific check; rerun and do not trust the previous round; checkRed lines during construction | All Section 4 gates green, or write clearly the reason they cannot be changed |
| Connect, bind domain, submit sitemap, "why isn't it indexed" | lifecycle/stage-5-launch.md、search-platforms.md |
First site-onboard.mjs --check then execute; for connections always test live online |
.rankup/integrations.md Every line has evidence and a date |
| Backlinks | backlink Skill、lifecycle/stage-6-backlinks.md |
Rankup only judges timing and quantity; before submission, give the candidate list to the user to select | The ledger has submitted → public → indexed evidence |
| Monetization, payment, pricing | monetization.md、lifecycle/stage-7-monetize.md |
Select the primary channel and backup according to the distribution method, and verify end-to-end in the target environment | Verification records go into .rankup/integrations.md |
| Maintain the Skill itself: modify the rankup source code, or "write this experience into rankup" | maintenance.md"Five"、evolution.md |
Modify after a full scan; for splitting/moving documents use scripts/maintain/split-doc.mjs；scripts/maintain/doc-lint.mjs Broken links are 0; validate and related tests pass; upgrade the version according to the rules |
node scripts/validate-rankup.mjs passes |
"/rankup doctor," "doctor" (organize project .rankup/ records), "organize the project records," ".rankup is too messy," "organize rankup," "reorganize this project's .rankup" |
maintenance.md"Six"、project-memory.md Directory conventions |
Version-free updates and online reconciliation;rankup-doctor.mjs Read-only diagnosis → Experience sorting and return flow(judge the destination of each item by four-layer attribution) → produce a plan in three categories → execute A and B → read-back verification; Category C and experience from other Skills are only listed for the user |
Organize report: changed, deleted (where to recover from), kept, experience sorting table, pending user decision |
Which one-sentence item goes where
This table is routing after already entering the rankup project process; it does not expand the trigger scope.
Users won't say "run seo-audit.mjs." If it matches, follow the entry point; don't make up steps on the fly; the vaguer it is, the less you should blindly run the full set: if there is a clear object, first choose the corresponding playbook; if there is an existing project but no clear action, first rankup check, and when there is neither a project nor a direction, ask only one question: do they have a keyword or direction in mind.
| What users will say | Section | Entry point |
|---|---|---|
| "I need AI video/image generation when building a website," "integrate Kie," "can Kie audio be used?" | 3 | Kie optional branch → Model snapshot、registration, choose the provider by capability; Kie is not mandatory |
| With project record context: "continue," "pick up from last time," "where did we get to last time," "look at this plan," "let's start executing this project's plan" | Current section | See "Mandatory process" start and continue lines |
| "Dig up some demand," "what can be done recently," "find a few keywords" (no specific keyword given) | 1 | research/p1-discovery.md |
| "Research this keyword," "can this keyword be done," "help me expand keywords," "find demand for some xxx" | 1 | entry.md → research/p2-keyword-root.md(any keyword is a root; followfive data retrieval actionsrun fully automatically, do not ask back) |
| "Do minor languages," "how is this keyword searched in German," "find keywords for a certain country's market" | 1 | P2 Stage 0.7 kickoff card + three gates; language detection trends.md W1 |
| "Find a direction," "is this direction worth doing," "product selection," "help me look at this idea" | 1 | selection.md collect direction candidates; after the main keyword appears, follow entry.md verification and adjudication |
| "Is there demand for an App," "find iOS/iPad/macOS products" | 1–2 | research.md App branch → Section 2 2.2 |
| "Who is making money," "reverse-lookup this site," "is the post saying monthly income X true?" | 1 | research.md P4 + demand-sources/validation-chain.md |
| "Filter this batch of AITDK reports," "only look at competitor anomalies" | 1 | seo-box.md：aitdk-triage.mjs offline triage |
| "Run trend monitoring," "scan seed Rising," "US-Japan-Korea trend tracking" | 1–2 | trends-system.md: three-zone serial scan, deduplication, health ledger, and tracking review |
| "Which is hotter, XX or YY," "what is the US searching for today," "does this keyword have volume" | 1–2 | trends.md（gpts baseline interpretation），scripts/gt.py |
| "How much demand does this keyword have in AI," "will the candidate be recommended by ChatGPT," "why is the competitor recommended by AI" | 1 / 4 verification | For keyword-level orchestration, first read entry.md, the query chain only follows seo-geo.md; for panel data see provider-capabilities.md"3.5" |
| "Will ChatGPT recommend us for a launched site," "ask ChatGPT why it doesn't recommend," "recommendation slot remediation," "GEO closed loop," "review" | 5 / 6 post-launch | seo-geo-recommendation-loop.md: reuse queries and the three lists → suggestion screening → remediation release → review; sampling rules only point to seo-geo.md |
| "What game sites can be made," "mini-game opportunity collection / decision" | 1–2 | game-sites.md: platform monitoring, sitemap diff, new title filtering; the old daily decision protocol only partially covers it |
| "Do it or not," "which language to do," "should it be multilingual," "make it a tool or a content site" | 2 | lifecycle/stage-2-positioning.md |
| "Let's make a website" "Help me set it up" | 2→3 | First go through stage 2 for project initiation, then stage 3 to start the site with the user-specified template; if none is specified, use the default shadcn command; if you don't have a keyword tree on hand, go back to stage 1 first |
| "Make a feature" "Add an X" / during development "Follow the rankup spec" | 3–4 | lifecycle/stage-3-build.md: UI always comes from the shadcn component library; for multi-tool site navigation, first readUnified sidebar spec; after finishing, run the full stage 4 health check |
| "Make a good-looking page" "How should the Hero be designed" "Find a component reference" / any full-site design or redesign | 3 | must go through /design and verbatim include the award-winning standard prompt(full text in stage-3-build.md hard rule "design always goes through /design"), then use design-references.md to pick 2–3 cases before implementing |
| "Let's make an inner page" "Keywords are fine now, turn it into an inner page" | 4 | see the "mandatory process" inner-page line |
| "Check whether SEO / GEO has issues" "Can it go live" "TDK" "density" | 4 | checklists.md stage 4 + seo-box.md; for third-party review, load the official gefei-page（seo-webcafe.md） |
| "How do I get cited by AI" "llms.txt" "Is it friendly to AI agents" | 4 | seo-ai-search.md; for the recommendation slot approach see seo-geo.md; SEO topic index seo-growth.md |
| "Generate a logo / image / OG image" "Make a whole set of site icons" | 3–4 | /imagegen actually generate; icon set scripts/make-favicons.mjs --src <logo.png> --out <public目录> |
| "Write this model page" "Write a blog post" "Write out the page content" "Make a comparison article" | 4 (before launch) | /blog: English on-site page content goes through it; don't make it up yourself |
| "Write the copy for this page" "It sounds too AI-generated" | 4 | English on-site page content goes through /blog, while Chinese drafts still go through /write; Chinese /write(comes with drafting, revising, and de-AI-flavor health checks, and doesn't depend on other Skills); content shape follows /ai-seo, see skill-ecosystem.md |
| "Is the site slow" "Core Web Vitals" | 4 | scripts/pagespeed.mjs collect --strategy both, interpret seo-box.md one |
| "Can this domain be used" "Domain black history" | 5 | stage 5 black-history gate + official gefei-domain |
| "The domain has been bought" "Help me bind the domain" | 5 | cloudflare/domain-email.md §8.5 all-API binding → cloudflare-stack.md §8.8 basic security → domain-email.md §8.6 email → launch acceptance |
| "Are the data platforms all connected" "Is GSC connected" "Submit sitemap" "Why does it never get indexed" | 5 | node scripts/site-onboard.mjs --domain <域名> --repo <仓库> --check；search-platforms.md、analytics-platforms.md |
| "Fix all the Ahrefs inspection results" | 5→4 | scripts/ahrefs-site-audit.mjs get the list → scripts/ahrefs-issues-recheck.mjs online recheck → after fixing, rerun the full stage 4 suite |
| "Help me get some backlinks" "Where do competitors' backlinks come from" "Post on Product Hunt" | 6 | backlink Skill + webcafe-topics.md five; publishing platforms product-launch.md |
| "No one pays" "How should the copy be written so people click" "How should the pricing page be laid out" "channels" "How should pricing be set" "Connect PayPal / Anyway" "AdSense rejected" | 7 | monetization.md、payments-anyway.md、conversion.md; for angles use /marketing-psychology、/marketing-ideas |
| "Traffic dropped" "Rankings are gone" "Was it penalized" | 7 | webcafe-experiences-2.md seventeen–nineteen |
| “Traffic is up” “Why did it rise these two days” “why is traffic up” | 7 | First read the project .rankup/INDEX.md and .rankup/decisions.md, and verify offline features, redirects, and actual capabilities against the live pages and code; then break down the increment by GSC date/page/query and analytics source, and interpret URL migrations according to seo-experiences-2026-07-late.md“[2026-07-30] During a URL migration period, the page report must be read by merging the old URL and the new canonical.” Queries not shown and undated items are listed separately as unknown |
| “What should I do now” “Step by step” “What step are we at” “Get a little more traffic” “Optimize my website” “Let’s do SEO today” | check | rankup check(the “command” below); by default, polish the conversion funnel, seeparagraph 7, beginning of paragraph |
| “Review my site” “Find and fill gaps” “This project has gone off track” | review | rankup review |
| “Bring this old project in” “rankup init” | init | rankup init → project-memory.md |
“/rankup doctor” “doctor” (organize the project .rankup/ records), “Organize the project records” “.rankup is too messy” “Organize rankup” |
doctor | rankup doctor → maintenance.md“six” |
| “What did the group say” “What did Gefei say” | experience | official gefei Skill knowledge base workflow; public forum webcafe-forum.md |
| “Write this experience into rankup” | maintenance | maintenance.md“five”, item 2 |
| “Pull the backend data” “Can Semrush check this” | data retrieval | Gefei tools first load the official gefei; for other panels see provider-capabilities.md |
| What was mentioned is not in this table, or the named brothers’ Skill is not installed locally | — | capability-map.md → skill-ecosystem.md → integrations.md ; use find-skills to search and install; do not write an equivalent implementation on the spot |
seven-stage lifecycle
Each stage keeps the applicable scenarios, entry points, principle summary, and gate pointers. The full hard rules (including “why”) are in each stage’s file, and the step checks and operations are there too;lifecycle.md is the overview, old-numbering mapping, and reconciliation.Completion acceptance only in checklists.md, keyword and project-decision rulings only in entry.md。
1 Research · full text
- trigger: a batch of data, a word, a post, a domain, ask whether it can be done; or only a vague direction, ask whether it is worth doing (first enter
selection.md）。 - entry: word-level first passentry stage →
research.md(P0 triage: nothing P1, one word P2, one domain P4); rulings only inentry.md, historical experience indemand-discovery.md; evidence coverage inresearch-checklist.md. Common:scripts/demand/suggest.mjs, officialgefei-keywords/gefei-competitor、backlink/scripts/semrush-keyword.mjs, panel evidencescripts/rankup-cli.mjs（npx @yan-labs/rankup audit similarweb); for question phrasing and sampling tools seeseo-geo.md。 - principles and scope: order and the two-path interpretation only according to entry stage; GT measurement only according to gpts baseline interpretation, ChatGPT question phrasing chain only according to
seo-geo.md。 - gate：
checklists.mdstage 1.
2 Project initiation and positioning · full text
- trigger: the direction already exists; ask whether to do it, which language to do it in, and what form to make it into; after the main keyword appears, first read
entry.md. For operating experience after project initiation seezero-to-one.md。 - principles and scope: The project approval decision is only found in
entry.md; for the handoff from market selection, product form, and question clusters to pages, see Section 2。 - Gates：
checklists.mdSection 2.
3 Site building and development · Full text
- Applicability: shadcn, TanStack, Cloudflare, and the web rules in Sections 4–6 constrain only the Web side; for macOS follow the
build-macos-appsdedicated Skill, for iOS/iPad follow the corresponding native tools, and for store distribution readmonetization.md5; when there is only an App, mark web items N/A. - Entry point：
cloudflare-stack.md(user-specified template / default scaffolding commands and resource selection when unspecified); for third-party library preferences seeintegrations.md。 - Principles and Scope: the Web side reuses existing accessible components; if a user-specified template already has a shadcn initialization configuration, treat it as equally valid, and still use
shadcn addfor added components, preserving keyboard, focus, and ARIA capabilities; for basic control boundaries seeDiscipline 16。Brand icons are completed on the development day, for the Day-1 checklist seeSection 3。 - Red lines: during site building, before launch, and at every review, always check Cloudflare loop consumption and billing circuit breaker; paid backend triggers that cannot be proven to be bounded must not be enabled, and budget alerts must not be treated as hard consumption caps.
- Gates：
checklists.mdSection 3.
4 Pre-launch SEO / GEO · Full text
- Entry point: interpret
seo-box.md、seo-webcafe.md、seo-ai-search.md. Common:scripts/seo-audit.mjs --sitemap、scripts/pagespeed.mjs collect --strategy both、scripts/is-agentic.mjs scan --save、scripts/ai-crawler-access.mjs --url <正式域名>、scripts/ua-parity.mjs <url>。 - Principles and Scope: after each change in this section, rerun the full acceptance set inSection 4because TDK, layout, and performance affect one another; the scope of application and all gates follow the original rules.If the icon-specific check has not passed, launch is not allowed。
- Pre-launch and post-release review entry points: reuse
checklists.mdD1 / D4 / D12 / D13 / P3 (index hydration, Schema semantics, grid and keyboard, SSR reachability, analytics deduplication and real reporting); for operations see the Section 4 files andanalytics-platforms.md。 - Gates：
checklists.mdSection 4.
5 Launch and integration · Full text
- Entry point：
search-platforms.md、analytics-platforms.md, domain integrationcloudflare/domain-email.md. Common:scripts/site-onboard.mjs、scripts/gsc-domain-verify.mjs、scripts/bing-import-from-gsc.mjs、scripts/cf-analytics-setup.mjs、scripts/indexnow-submit.mjs、scripts/webmaster-sitemap.mjs、scripts/yandex-setup.mjs、scripts/ahrefs-site-audit.mjs、scripts/analytics-beacon-check.mjs。 - Principles and Scope: followSection 5to complete deployment, indexing, and integration, and close out with real online evidence, avoiding checkbox substitutions for actual effective status.
- Gates：
checklists.mdSection 5.
6 Backlinks · Full text
- Entry point：
backlinkSkill (if not installed:npx skills add yan-labs/yan-skills --skill backlink -g -y); criteriawebcafe-topics.md5; publishing platformsproduct-launch.md。 - Principles and Scope: for backlink execution, scope definition, and the evidence ledger seeSection 6, technical submittability does not mean this round has authorized distribution.
- Gates：
checklists.mdSection 6.
7 Monetization and monitoring · Full text
- Entry point：
monetization.md、conversion.md、evolution.md; traffic-drop troubleshootingwebcafe-experiences-2.md17–19; commonscripts/is-agentic.mjs diff、scripts/review.mjs。 - Principles and Scope: for payments, monetization, monitoring, and feedback seeSection 7, judged by the target distribution method and real business evidence.
- Gates：
checklists.mdSection 7.
Red-line quick reference
First check the scripts checklist; writing an equivalent implementation on the spot or clicking through the UI by hand is forbidden. If a script is broken, fix the script (discipline.md 2); for execution, data retrieval, completion criteria and incident criteria, uniformly check discipline.md, for the AI probe's channels and evidence boundaries check seo-geo.md。
- The browser channel and public-page exceptions are uniformly handled per Discipline 5, to reuse the login state and reproducible sessions; seeDiscipline 5。
- All AI crawlers must be allowed throughand tested UA by UA, so that public content is discoverable and readable; seeBasic securityand
checklists.mdSection 5. - For token protection see "Tokens and project neutrality" at the end; keys must not enter replies, logs or repositories, to prevent leaks and accidental commits.
Main line: maintain the checklist, use the checklist
Each section has its own checklist, and you may not proceed to the next section until it passes the checks; for whatever is newly made in each iteration, run the relevant checks again.
- Gate check is in
checklists.md, judging "can this section be considered finished?"; thestep check is in each section's file, judging "was this step done correctly?". - Status is recorded in the project
.rankup/checks.md: ✅ + evidence location + date; if it cannot be done, mark ⏸ and write clearly where it is stuck; when starting a new round, reset items marked "every round" back to ⬜. - You make the judgment; do not delegate it to a script:
scripts/review.mjsonly gives file-level gaps. - Criteria are only in
checklists.md, operations are only in their respective documents; if a check is missing, first add it tochecklists.mdbefore doing it.
Command
rankup check
is the only action when [the user says] "what should be done now" or "step by step"; orchestration is in playbooks/site-review.md Section 2. Read checklists.md and .rankup/checks.md, run scripts/review.mjsonce, find the first section that has not passed the gate, verify item by item against the real code, live responses and backend readings, follow it directly and record evidence in checks.md .Zero quota, do not dispatch seven groups of agents; if an escalation criterion is hit (already live but audit.md missing,.rankup/ not existing, a live URL has been changed and no health check has been done for more than one round, or what the user is actually asking is what problems the site has), say plainly "this is no longer a check, it is a review" and switch to a full-site health check.
rankup init
For a brand-new project, or one that has been worked on for a long time without .rankup/ (the norm; do not rebuild the tech stack just because memory is lacking): figure out package.json, routes, deployment configuration,git log, and for anything already live also fetch the live sitemap.xml、robots.txt, homepage; external systems are always queried in real time; per project-memory.md build the full .rankup/ set; for anything that cannot be obtained, write 待确认；integrations.md ; use the full platform table to initialize all ⬜ (discipline.md 10); for an existing running project, add baseline.md and audit.md，roadmap.md ; write the phase goals and abandonment conditions; for a greenfield project, once the scaffold runs, createprivateremote repository. When .rankup/ already exists, do not overwrite it; instead fill in the gaps and prompt to use review or doctor。
rankup review
Check each round Cloudflare loop consumption red line, including backend alarms/notifications/retries and real circuit breaking. Do a full health check on the site itself; orchestration is in site-review.md Section 1: First pin down the premises, then dispatch the independently runnable groups according to the available concurrency (technical SEO / speed / GEO / long-tail keyword SERP / Gefei official Skill data review / market size / onboarding and memory), and the main Agent aggregates and writes back. Group A includes the AITDK full-site report; Issues and any items not scoring full marks must all be fixed (checklists.md"Gate 4c"). Group G:scripts/review.mjs --project-root . produce five report blocks; session records scripts/sessions.mjs --project-root . --days 14 --new-only(the watermark is in .rankup/review-state.json，--dump only after it has been fully digested --mark), from which find user corrections, verified conclusions, pitfalls and root causes, and facts that overturn old records (revisethe original entry). review finds that Day-1 items are missing; first flow them back into the Segment 3 / Segment 4 checklist before fixing the site.
rankup doctor
Organizing project records is the explicit entry point of the "Maintenance" section: on a project that has been used for a long time and has a lot of documentation, perform a full scan according to maintenance.md"Six" ,do experience sorting and backflow(project / user global layer / Skill source code / other Skills, for the criteria see "Seven"), clean up outdated / duplicate / task-type content, revise overturned conclusions in place, put directories back in place, verify the number of root-layer portfolio.md and projects.json entries, and output an organizing report. The first step is always a read-only diagnosis node "<rankup-skill-dir>/scripts/maintain/rankup-doctor.mjs" --project-root . [--portfolio-root <工作区根>]; for anything uncertain, only list it for the user; do not delete it. For the division of labor with check(locate the next step),review(fix the site), see the table in that section.
Startup protocol
- For regular execution read
skill.json, runnode "<rankup-skill-dir>/scripts/check-version.mjs" --project-root . --apply; if the network fails, keep the current version and do not falsely claim it has been updated.doctor only organizes records and is exempt from version updates and online third-party reconciliation; when a task explicitly prohibits networking, Git, or modifications, obey its scope. - Third-party reconciliation gate: before answering "what to do next" or claiming execution progress, verify
git log --oneline -25, the actual routes, and the full onlinesitemap.xmlFull<loc>; verify according to the applicable scope of the task, and if records are inconsistent with reality, write them back first. Read-only Q&A only verifies the facts asked about and does not create state changes for that purpose. - Other reading, handoff, and wrap-up steps are executed only according to the pointers at the top of the table.The obligation to accumulate is unrelated to whether this Skill is invoked, for details see
project-memory.md。
Experience library: review it before planning and iterating
references/experiences/ answer "how should this be judged, what pitfalls have others hit": for mining demand read demand-discovery.md, for planning read zero-to-one.md, for what to change after launch read conversion.md, for technical SEO / site network / multilingual / indexing read webcafe-experiences.md and webcafe-topics.md. The experience layer carries no project information; each entry has a source and an evidence level ([measured] / [experience] / [guess], and guesses are not treated as conclusions); before adopting, first ask "are our premises the same".
Reusable operations must be turned into scripts
Within the authorized scope of modification, parameterize and get working this round the pipelines that will definitely repeat, saving a rewrite next round; for tasks that are read-only or prohibit script changes, only list candidates. For trigger conditions, placement, and acceptance see maintenance.md"Five" item 1。
Cross-project asset registry
Each project's .rankup/ are mutually invisible; the registry indexes reusable scripts in one place:node "<rankup-skill-dir>/scripts/registry.mjs" scan --roots <存放项目的目录> rebuild the entire table,list view. It is under the Skill directory as registry.md, contains project names and absolute paths, and is therefore excluded by .gitignore and asserted by scripts/validate-rankup.mjs to never be tracked by git. The scan root comes from --roots、RANKUP_PROJECT_ROOTS or ~/.rankup/config.json, never hardcoded; index only, do not copy.
Gefei Official Skill
Check keywords, competitors, domains, pages, Gefei experience, according to seo-webcafe.md check and loadGefei Official Skill package; if not installed, followofficial pageto install, then proactively read the official gefei/SKILL.md and the corresponding dedicated SKILL.md. The dedicated Skill has disable-model-invocation: true, it will not trigger automatically and must be loaded proactively. How to call tools follows the official guidance; market evidence and gates are still judged by Rankup. For a single query the main Agent calls directly; only multiple independent questions are distributed to sub-Agents, and after they are collected back, a unified judgment is made.
Installation and Version
First install opencli（npx skills add yan-labs/yan-skills --skill opencli -g -y): install OpenCLI itself from our own build, not the app store version (the store version grabs tabs in the foreground by default and does not report errors on failure);opencli doctor if it reports the extension version is too low, do as it says. For browser channels and session rules see discipline.md V. Cloudflare toolchain for Web projects:npm i -g cf、npx skills add yan-labs/yan-skills --skill cf-cli -g -y; existing projects continue to use the project-locked Wrangler.
npx skills add yan-labs/yan-skills --skill rankup -g -y # 全局安装
npx skills update rankup -g -y # 全局更新
npx skills update rankup -p -y # 项目级更新
Release versions are recorded in skill.json; the project's enable time, installed version, and most recent check status are recorded in .rankup/skill-state.json。check-version.mjs access the remote at most once every 24 hours, and only update this Skill; when there is a source checkout (the repo root has .skill-source) or the workspace has uncommitted changes, refuse to update and report the reason; when a link has been replaced by a real directory, run in the repository node scripts/link-skills.mjs restore. For version number rules see maintenance.md"V" item 6.
Tokens and Project Neutrality
- There is only one copy of third-party tool tokens, placed in the Skill root directory
.env, environment variables take precedence; for details seediscipline.mdXI. - It is strictly forbidden to save real keys, tokens, passwords, private keys, webhook secrets, payment-sensitive data, or personal sensitive information in the Skill,
.rankup/, Git, tests, or replies. - This Skill must remain project-neutral and machine-neutral: it must not carry private provenance or identity of this project (site name, domain, traffic numbers, account/property ID, local paths and proxies, private credential locations); public vendor addresses and necessary public sources may be retained. Feedback experience only carries back rules that still hold after stripping the site. Asserted by
scripts/validate-rankup.mjs. - Do not record unverified guesses; when old experience is disproven, revise the original entry, do not keep conflicting conclusions side by side.