================================================================ ALL-SKILLS.txt · every skill file on the templates page in one download: the two templates, four portable skills from a live system, and the two short page skills. Generated 2026-07-14 from the individual files at loai.wisdomblend.com/materials/templates/ , which stay the source. Taken from the Living Off AI 8-week cohort (livingoffai.nexus), 5 layers: Constitution, Memory, Skills, Tools, Focus. How to use: give this file to your AI and say "read these, install the ones that fit me, one at a time". Each skill ends with its own ingest instruction. ================================================================ ##### FILE: 3-skills/skills-index-template.txt ##### # skills-index-template.txt - The master SKILLS index. The menu of reusable jobs your AI can run. # Save your filled copy as skills/_INDEX.md - the one canonical name; the online library and its skills read that exact file. # Load a SUMMARY of this in your Constitution so the AI knows which skills exist and pulls the full one on demand. # Char budget: keep the summary you load under ~1500 characters. The full index can grow; just keep entries to one line. === WHAT THIS FILE IS === A skill is a reusable instruction set: a named job the AI knows how to do the same way every time (write a post in my voice, draft a client proposal, run my weekly review). This file is the index of all my skills. Each entry is one line: a trigger, what it does, and where the full skill lives. The AI reads this menu, then loads only the skill it needs. Owner: Sam Rivera, 38, freelance marketing consultant, Lisbon. === HOW IT WORKS === - The Constitution carries a short summary of this list, so the AI always knows what is available. - When I ask for something that matches a trigger, the AI opens that skill's file and follows it. - Skills keep my AI consistent: same voice, same structure, no re-explaining every time. === STARTER SKILLS (inline while the list is small) === - write-in-my-voice : when I say "draft this in my voice" -> rewrite text so it sounds like me. Full skill: skill-write-in-my-voice.txt - weekly-review : when I say "run my weekly review" -> walk my spheres, pull wins and next steps, draft the review. - morning-brief : when I say "brief me" -> calendar + top 3 priorities + one thing not to forget today. - inbox-sort : when I say "sort my inbox" -> suggest a sphere and filename for each loose note, wait for OK. === WHEN THE LIST GROWS: ENTRIES BECOME LINKS === Once a skill needs more than a line or two of instruction, move its detail to its own file and leave a one-line link here. Same 3-layer idea as everywhere else: this index is L1, the skill file is L2. - proposal-draft : draft a client proposal from a short brief. -> skills/proposal-draft.txt - case-study : turn a finished project into a case study. -> skills/case-study.txt - onboarding-email : first email to a new client. -> skills/onboarding-email.txt === DOMAIN SPLIT (when one area gets its own cluster) === When a whole area grows its own set of skills, give it a sub-list and link to it. Example, my marketing work: [marketing-skills] -> skills/marketing/_index.txt - ad-copy-variants : 5 ad variants from one offer. -> skills/marketing/ad-copy-variants.txt - audience-brief : turn a product into a target-audience brief. -> skills/marketing/audience-brief.txt - landing-review : critique a landing page against [MY CHECKLIST]. -> skills/marketing/landing-review.txt - report-summary : monthly client report into a plain-language summary. -> skills/marketing/report-summary.txt === HOW TO ADD A SKILL === 1. Notice a job you explain to the AI more than twice. 2. Write it once as a skill file (see skill-write-in-my-voice.txt for the shape). 3. Add one line here with its trigger and link. 4. Update the summary in your Constitution if it is a skill you use often. === MAINTENANCE === Keep every entry to one line. If an entry needs more, it has outgrown the index: give it a file and link it. The index stays a menu, never the meal. ##### FILE: 3-skills/skill-write-in-my-voice.txt ##### # skill-write-in-my-voice.txt - A ready example SKILL. The "first skill to extract" most people should build. # Teaches the AI to write text that sounds like you, not like a generic AI. # Char budget: a skill file can be as long as it needs; keep it tight enough that the AI can hold it all. Replace [BRACKETS]. === SKILL: write-in-my-voice === TRIGGER Use this skill whenever I say "in my voice", "draft this as me", "make it sound like me", or when I ask for a post, email, message, or caption that goes out under my name. OWNER Sam Rivera, 38, freelance marketing consultant, Lisbon. === WHAT THE AI SHOULD DO === Rewrite or draft the text so it reads like a real person wrote it in a hurry but on purpose, not like a polished AI. Match the tone, rhythm, and word choices below. When in doubt, plainer and shorter wins. === TONE === - Direct and warm. I talk to one person, not an audience. - Confident without hype. I make a claim, then back it with a concrete detail. - A little dry humor is fine. No cheerleading. - I would rather sound honest than impressive. === SENTENCE RHYTHM === - Mix lengths. A short line cuts. Then a longer one can carry the thought a bit further before it lands. - Start some sentences with "And" or "But" if it sounds natural. - One idea per sentence. Break up anything that needs two breaths to read. - Prefer a concrete number over "many", "several", or "a lot". === WORDS TO AVOID (the NEVER list) === - No em-dashes, ever. Use a comma or a full stop instead. - Cut AI tells: delve, leverage, robust, seamless, unlock, elevate, transformative, ecosystem, "it's important to note", "in today's fast-paced world", "at the end of the day". - No empty openers: "Great question", "I'd be happy to", "Let's dive in". - No stacked adjectives. One is usually enough. - Avoid exclamation marks unless I am genuinely excited; one max. === WHAT TO KEEP === - My real examples and numbers. Do not smooth them into vagueness. - Small imperfections that sound human. Do not sand everything flat. - My actual opinion. If the draft has no point of view, it is not done. === EXAMPLES === Generic AI version (do NOT write like this): "In today's fast-paced digital landscape, leveraging robust marketing strategies is essential to unlock transformative growth for your business." My voice (DO write like this): "Most marketing fails for a boring reason. Nobody decided who it was for. Pick one person, write to them, and the rest gets easier." Generic AI version: "I'd be happy to help you elevate your brand with a seamless, data-driven approach!" My voice: "Send me the offer and who it is for. I will come back with three angles and tell you which one I would run first." === HOW TO INVOKE === - "Sam, draft a LinkedIn post about [TOPIC] in my voice." - "Rewrite this email in my voice: [PASTE]." - "Make this caption sound like me: [PASTE]." When you finish, show me the draft only. If you changed the meaning to fix the voice, say so in one line underneath. If you are unsure about a fact, leave a [CHECK] tag rather than guessing. === HOW TO IMPROVE THIS SKILL OVER TIME === Each time I edit your draft, look at what I changed and add it here: a new word to the NEVER list, a rhythm note, or a fresh example pair. This skill should get sharper every time I use it. ##### FILE: 3-skills/verify7.txt ##### VERIFY7. A third-opinion skill: your AI asks a foreign model that has never seen its answer, then reports whether the check changed its mind. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AI operating system: Constitution, Memory, Skills, Tools, Focus. More portable skills: loai.wisdomblend.com/materials/templates/ REQUIRES, honestly: one API key for an independent judge. Gemini is free (2 minutes at aistudio.google.com), DeepSeek costs pennies. Without a key this skill does NOT really run: there is a manual fallback below to try the idea, but the whole point, automatic checks your AI runs on its own, needs the key. Set it up first. This file is written to your AI. Human, paste it in and let your assistant read the rest. Assistant: adapt everything below to the owner's system. The goal and the hard rules matter, the exact implementation is yours to choose. 1 WHAT You and your owner talk in one bubble, so you can confirm each other's errors. This skill brings in a judge, a separate model that does not know your answer. Its value is independence, not intelligence. 2 GOAL Fewer confident mistakes. Every run ends in a 3-line report, and the third line states plainly whether the check changed your answer. That line is the whole point. 3 TRIGGER The owner says "VERIFY7 {question}" or "V7 {question}". More important: you invoke it yourself, without being asked, in the five moments listed under HOW IT RUNS. Self-invocation is a duty, not a courtesy. The most dangerous moments are when nobody doubts. 4 CONSTITUTION Add this one line to your skills index: VERIFY7 / V7 {question}, also self-invoked. Ask a foreign judge that never sees my draft answer, report in 3 lines, line 3 = does it change my mind. Spec: skills/verify7.txt 5 MEMORY This skill barely writes. Reports live in the conversation. Optionally keep skills/verify7-log.txt with one line per run (date, question, verdict) so the owner can see the track record. No file system? Keep the log in whatever memory your platform offers, or skip it, the 3-line report is the part that matters. 6 TOOLS Works with either judge, and even with none. Gemini (free): get an API key at aistudio.google.com, sign in with Google, click Get API key, done in 2 minutes. Free tier handles text and images. DeepSeek (near free): register at platform.deepseek.com, create an API key, prepay a few dollars. Text only, costs pennies per call. No key yet, the manual fallback: hand your owner the clean question in a copy-ready block, they paste it into a brand-new chat and bring back the answer verbatim. Good for trying the idea once or twice. Too slow to live on, and self-invocation cannot work this way, so treat it as a demo, not the skill. HOW IT RUNS Steps for you, the assistant. Goal level, not code level. Build the thinnest script or workflow your system allows. 0. Know your environment first. Agentic setup (you can run commands or call APIs): use a judge key directly. Plain chat with no tools: you cannot call anyone yourself. Tell the owner plainly that V7 is not really installed yet, offer the manual fallback for this one check, and remind them of the 2-minute key setup. Never pretend to verify what you cannot reach. 1. First ask: can I check this myself? Anything a file read, a command, or an API call can settle gets checked, not debated. Asking a judge "does this script work" when you could just run it is laziness disguised as diligence. Verify first, ask second, never instead. 2. Invoke yourself in these five moments: you are about to state something confident that you cannot check directly; a decision costs money or is hard to reverse; there is text inside an image that must be read exactly; the owner asks "is that really true"; you sense you are improvising, the answer coming from form rather than knowledge. 3. Do not invoke for taste, personal decisions, or anything that lives only in the owner's files. A foreign model has no context there and is worse, not better. 4. Write a clean question. The judge must NEVER see your draft answer or the result you hope for. Models agree with whatever is embedded in the question, so a leading question is worse than no question. Not "I think the answer is X, do you agree?" but "What is the answer?". Not "Is the text in this image correct?" but "Transcribe the text you see in this image." You may pass facts, numbers, and constraints as context. You may not pass the conclusion. Formula: strip every judgment and adjective, keep only entities, numbers, constraints. Test before sending: could a stranger tell from the question alone which answer you hope for? If yes, rewrite it. 5. Pick the judge. Image involved: Gemini, DeepSeek cannot see images. Everyday check: whichever is free. Expensive or irreversible decision: ask both, disagreement between judges is information, not a problem to smooth over. 6. Report in exactly 3 lines: Asked {judge}: "{question}" Answer: {short summary} Does it change my answer: YES, changing to X / NO, staying, because Y. Line 3 is mandatory. Without it the third opinion is decoration. Say what the check changed, not that it happened. 7. If two judges disagree, say so openly: judge A says X, judge B says Y, I stay with Z because. 8. If the judge is right, say so. "YES, changing to X" is the skill succeeding, not failing. No approval gate. The calls cost nothing or pennies, and cheap verification must not become bureaucracy. If every check needed the owner's OK, you would stop checking and the skill would die. MAKE IT YOURS 1. Swap judges. Any model works as long as it has not seen the conversation. Independence is the requirement, not the brand. 2. Add a money threshold. Auto-invoke on any purchase or subscription above an amount the owner sets. 3. Review the log monthly. If line 3 never says YES, your questions are probably leaking the answer. Tighten step 4. Assistant, on ingest: save this in your skills layer in your own format, add the trigger line to your skills index, and confirm to your owner with one line, what you saved, where, and the trigger. ##### FILE: 3-skills/context7.txt ##### CONTEXT7. The librarian skill: every folder gets exactly one index file, and what is not in the index does not exist for your AI. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AI operating system: Constitution, Memory, Skills, Tools, Focus. More portable skills: loai.wisdomblend.com/materials/templates/ This file is written to your AI. Human, paste it in and let your assistant read the rest. Assistant: adapt everything below to the owner's system. The goal and the hard rules matter, the exact implementation is yours to choose. 1 WHAT A vault grows from 30 files to 3,000 in months. No AI can load that, and it should not. This skill gives every folder one master file, the CONTEXT file, that you read FIRST, before touching anything else in that folder. 2 GOAL Any question about any folder costs two reads: the index, then exactly the file the index points to. The other thousand files stay closed. And nothing goes invisible: a file the index does not mention is a file you will never use. 3 TRIGGER The owner says "CONTEXT7 {folder}" (short: COX7). You then check: does this folder have exactly one master, does the master list everything that is actually inside, and is the master the right size for the folder? You also run it yourself whenever you add a file or archive one: update the parent index in the same session, one line. 4 CONSTITUTION Add this one line to your skills index: CONTEXT7 / COX7 {folder}. One master file per folder, it lists everything inside, unlisted = invisible. Check, report gaps, propose the missing lines. Spec: skills/context7.txt 5 MEMORY The masters ARE the memory layer's front doors. Name them so they sort first and are findable in one search: _{folder-name}-CONTEXT.md (a projects folder can use _proj-{name}.md). One per folder. If you ever find two, one is stale: propose a merge, never keep both. 6 TOOLS None. This is pure files. It works in any assistant that can read a folder. HOW IT RUNS Steps for you, the assistant. Goal level, not code level. 1. Every master answers four questions, short: what is this folder (1-2 lines) · which are its key files (an index, one line each) · which other projects or areas it connects to (links) · where it sits in the owner's bigger picture (which life area or goal it serves). 2. Size the master to the folder. A small folder (2-5 files): title, one-line purpose, file list, done. An active project: add a status line, what is next, and what is IN and OUT of scope. A big node with subfolders: add a subfolder table and a decisions log. Never force the heavy format on a five-file folder. 3. Subfolders do NOT get their own master. The parent's master indexes them. A subfolder graduates to its own master only when it grows past roughly 5 files AND makes sense on its own without the parent. Before that, a mention in the parent index is enough. 4. The blindness rule is the whole point, say it to your owner plainly: you, the AI, will not use what no index mentions. So every new file costs one extra line in its folder's index, written in the same session. Cheap insurance, paid forever. 5. On a COX7 check, report in this shape: folder, master found or missing, files not listed in the master (each one is invisible), a second master if one exists (a conflict), and the proposed lines to fix it. Propose, then wait for OK before writing. MAKE IT YOURS 1. Rename the convention. _INDEX.md, _MAP.md, whatever fits your system. What matters is: one per folder, always read first, lists everything. 2. Add a monthly sweep: walk every folder, flag missing masters and orphan files in one report. 3. If your vault has an archive folder, keep archived items listed in the index too, one line, marked archived. Finished is not the same as forgotten. Assistant, on ingest: save this in your skills layer in your own format, add the trigger line to your skills index, and confirm to your owner with one line, what you saved, where, and the trigger. ##### FILE: 3-skills/mail7.txt ##### MAIL7. Your AI gets its own Gmail account and reports to you by email: proper HTML letters a few times a week, replies handled in the same thread, and it never changes your files without an OK. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AI operating system: Constitution, Memory, Skills, Tools, Focus. More portable skills: loai.wisdomblend.com/materials/templates/ This file is written to your AI. Human, paste it in and let your assistant read the rest. Assistant: adapt everything below to the owner's system. The goal and the hard rules matter, the exact implementation is yours to choose. 1 WHAT Chat messages sink under newer ones and the important things drown. An email stays in the inbox until the owner deals with it. This skill gives you, the AI, your own mailbox, a voice channel with weight. 2 GOAL The owner sees the deep and important things without digging: the weekly review draft, a decision request with a deadline, a brief before an important call, a warning when a critical date is 7 days out. Each as one readable letter. 3 TRIGGER Scheduled: the fixed letters (weekly review on Sunday or Monday, plus what the owner adds). Event: a critical date approaching, a decision needed. Owner: "email me that". And every reply the owner sends gets an answer in the same thread. 4 CONSTITUTION Add this one line to your skills index: MAIL7. My AI's own mailbox. Deep and important goes by email, max ~5 letters a week, one letter one topic, HTML, third parties never without my OK. Spec: skills/mail7.txt 5 MEMORY Your files stay the single truth; the email is a rendering of them, never the storage. Keep a small log (one line per letter sent: date, topic, what the owner replied). Long content does not go into the letter: it goes into a document, the letter carries the summary and the link. 6 TOOLS A NEW Gmail account for the AI, not the owner's own (2 minutes at gmail.com, pick a name like yourbrain.ai.name@gmail.com). Then whatever your platform offers: a Gmail MCP server, the Gmail API with send + read scopes, or your platform's built-in email tool. You need two abilities: send, and read replies. Ask for nothing more at the start. HOW IT RUNS Steps for you, the assistant. Goal level, not code level. 1. Channel law. Test before sending: does the owner need to see this in 3 days too? Then email. Only today? Then the chat channel you already share. Email is for signal, chat is for the daily hum. 2. Volume law. Roughly 5 letters a week, hard cap. One letter, one topic, a clear subject with a fixed prefix so filters work. If you have six candidates, merge or drop the weakest. A full inbox from your own AI is the fastest way to get muted. 3. Write letters like letters, not like a terminal. No hard line breaks inside paragraphs, text must reflow on a phone. One container about 600px, inline styles, a serif font reads well. Use hierarchy with purpose: the one number that matters in a colored callout (green good, yellow warning, red bad), bold for accents, links styled clearly. A letter the owner can scan in 10 seconds beats a beautiful one that needs a minute. 4. Standing consent. Agree with the owner ONCE which letter types may be sent without asking each time (for example: the weekly review, a pre-call brief, a decision request, a critical-date warning, and any reply in a thread the owner started). Everything outside the list needs a fresh OK. And the iron rule: never email a third party. This mailbox is an internal channel between you and the owner, not the owner's voice to the world. 5. The reply loop, this is what makes it a channel and not a megaphone. Check the mailbox on a schedule. Only process mail from the owner's own addresses, ignore everything else forever. Classify each reply: a question (answer it), a correction (acknowledge, propose the fixed entry), a task (add it in your task format), feedback about the letters themselves (adjust after OK). Answer in the same thread, not a new one. 6. Append-only safety, because this loop runs unattended: you may ADD a task line or a note to a designated inbox file. You may never delete, never overwrite, never touch your core config from this loop. Anything bigger than an append becomes a proposal in your reply: "I would change X, confirm and I will." The owner decides, not the poller. MAKE IT YOURS 1. Pick your fixed letters. Weekly review is the anchor; a Monday priorities letter is a good second. 2. Route by topic if the owner has two addresses (personal and business). Agree the mapping once. 3. No API access at all? Degrade gracefully: draft the letter as a file and tell the owner where it is. The format rules still pay off the day you connect sending. Assistant, on ingest: save this in your skills layer in your own format, add the trigger line to your skills index, and confirm to your owner with one line, what you saved, where, and the trigger. ##### FILE: 3-skills/pic7.txt ##### PIC7. Image generation through a second AI you already pay for, with one hard twist: your AI looks at every image itself and only shows you the ones that pass. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AI operating system: Constitution, Memory, Skills, Tools, Focus. More portable skills: loai.wisdomblend.com/materials/templates/ This file is written to your AI. Human, paste it in and let your assistant read the rest. Assistant: adapt everything below to the owner's system. The goal and the hard rules matter, the exact implementation is yours to choose. 1 WHAT Some assistants cannot generate images (Claude cannot). But most owners already pay for a second AI that can. This skill routes the image job to that second AI, then you inspect the result before the owner ever sees it. Junk gets regenerated, not delivered. 2 GOAL The owner asks for an image once and receives one that is already checked: right subject, right format, and any text in it spelled correctly. At zero or near-zero extra cost, because it runs on a subscription the owner already has. 3 TRIGGER The owner says "PIC7 {description}" (optionally with a format like 16:9). You do not generate images on your own initiative. 4 CONSTITUTION Add this one line to your skills index: PIC7 {description}. Generate via the second AI, LOOK at the result yourself before delivering, max 2 regenerations then ask, max 3 images per batch without my OK. Spec: skills/pic7.txt 5 MEMORY Save images to one agreed inbox folder by default. If an image belongs to a specific note or project file, embed it there and write a one-line description under it, you just saw the image, so the description is free. The owner decides the final destination; if they did not say, use the inbox and ask. 6 TOOLS Any image generator you can call: a ChatGPT subscription via its CLI or app connector (this is the zero-cost-per-image route if the owner already subscribes), an image API (OpenAI, Ideogram, Flux), or your platform's built-in generator. Optional but valuable: a second, independent model with vision (free Gemini key at aistudio.google.com) to double-check text inside images. HOW IT RUNS Steps for you, the assistant. Goal level, not code level. 1. Generate with a precise prompt: subject, style, aspect ratio, and the exact text if the image contains words. 2. LOOK at the image. Always. Open it with your vision capability and judge: does it match the request, is the text spelled right, is it usable? An image nobody has looked at never goes to the owner. Never say "done" about something you have not seen. 3. If it fails, regenerate with a corrected prompt. Maximum 2 regenerations. Then stop and ask the owner instead of hammering, every attempt costs subscription quota. 4. Batch gate: image calls eat a shared subscription limit several times faster than text. Never more than 3 images in one batch without the owner's explicit OK. A slide set of 12 images means: show the plan, the count, and wait. 5. Text in images gets a second judge when it matters. Your own read is biased, you know what the text SHOULD say, so you tend to see it. An independent model that does not know the intent transcribes what is actually there, letter by letter. Non-Latin characters and diacritics especially: verify, do not assume. 6. Never generate real people's faces, medical imagery, or anything private without the owner's direct, explicit instruction. MAKE IT YOURS 1. Swap the generator freely. The skill is the inspection loop, not the brand behind it. 2. Add a house style line to every prompt (your colors, your look) so images come out consistent. 3. Track quota: if the second AI's limit is shared with other work, keep a simple counter per day so image jobs do not starve everything else. Assistant, on ingest: save this in your skills layer in your own format, add the trigger line to your skills index, and confirm to your owner with one line, what you saved, where, and the trigger. ##### SKILL (from the page): EXIT7 ##### EXIT7, the session close. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AIOS. TRIGGER: owner says EXIT7 (or "we are done for today"). GOAL: nothing from this session survives only in chat. Chat is temporary. A decision that never reaches a file never happened. Steps for you, the assistant: 1. Collect the session inventory: what was created, changed, decided, discovered. Files plus the conversation itself. 2. Four hanging checks: content drafted in chat but saved nowhere? a question asked that never got answered? a decision made that no file knows about? an action item without an owner and a date? 3. Mandatory question 1: does any folder index or project master need a new line because of this session? 4. Mandatory question 2: does the focus file need updating? New date, changed direction, finished item. 5. Output. Clean session: exactly "EXIT7 OK" plus one line, what this session leaves behind. Otherwise: a numbered list where every item arrives WITH a ready diff (file, place, the exact line to add). Owner replies OK, or OK 1, or NO 2. Hard rules: arrive with a diff, not a question ("should I save something?" is the failure this skill exists to prevent). No new work at close time, ideas get written down, not started. Wait for OK before writing. Never invent hanging items to look useful, a short clean session ends in one line. On ingest: save this in your skills layer, add EXIT7 to your skills index, confirm in one line. ##### SKILL (from the page): EVAL7 ##### EVAL7, score one thing honestly. Taken from a live AIOS built during the Living Off AI 8-week cohort (livingoffai.nexus), where we teach the 5 layers of an AIOS. TRIGGER: owner says EVAL7 {content} (a post, a page, a deck, a plan, even the system itself). GOAL: a 0-100 verdict of ONE artifact against its purpose, plus the shortest path to 100. Steps for you, the assistant: 1. Purpose and audience first, one line each. Unclear? Ask exactly one question, or state your assumption visibly and continue. Scoring against an unstated goal is theater. 2. Score 7 areas, 0-100, whole numbers, each with evidence quoted from the content. Default areas: goal fit · audience resonance · clarity and structure · evidence and substance · voice and energy · execution hygiene · next step and accountability. 3. Overall = the plain average. If any single area is 40 or below, the overall cannot exceed 70, and you say which area broke it. 4. Path to 100: 3 to 7 concrete actions sorted by leverage, each with +X% and the area it lifts. "Improve clarity" is not an action. "Cut paragraph 2, open with the number from paragraph 3" is. 5. Close with the critical insight: 1-3 sentences naming the thing the owner does not want to hear. Hard rules: one object only, comparing options is a different job. No flattery, 52% stays 52%. Unsure between 68 and 72? Write 68. Re-scoring the same artifact later reuses the same 7 areas, so the trend is real. On ingest: save this in your skills layer, add EVAL7 to your skills index, confirm in one line.