Keeping the record clean
The steps that read across what you approved: insights, skills, duplicates, same work and your GitHub projects.
Once you've approved facts, a few steps read across them: to say what's true of you across jobs, to list your skills once, to catch facts said twice, and to add your GitHub projects. Each proposes; nothing in your record changes until you decide in Record or Review.
The steps
Reading across what you approved.
Insights
After a story or project is fully reviewed, and when you choose Find new insights.
Reads
- Approved onlyYour roles, facts, context and directions
- Your wordsYour stories, for background only
- Rejected, to skipInsights you approved, still to review or rejected, so it doesn’t repeat them
- Approved onlyYour projects and their facts
May
Propose what’s true of you across roles and projects, each citing approved facts.
May not
Bring back an insight you rejected, or cite a fact you haven’t approved.
Gather skills
After a story or project is fully reviewed, when you approve a role after that, and when you choose Gather on Skills.
Reads
- Approved onlyThe skills and tools of your roles
- Approved onlyThe stack and languages of your projects
- Approved onlyYour facts
- Not yet reviewedSkills, tools and certifications you approved or that are waiting in Review
- Rejected, to skipSkills you rejected, with your reasons
May
Propose each skill, tool and certification once, with where your record shows it, and offer near-duplicates as merges.
May not
Bring back a skill you rejected, in any spelling, or change one you approved beyond adding where it shows.
Find duplicates
You choose Find duplicates on Record, or Find facts said twice on Review.
Reads
- Not yet reviewedFacts you approved or that are waiting in Review, by role, and the pairs you kept apart
May
Flag facts in the same role that say the same thing, for you to merge or keep both.
May not
Merge anything itself, or flag a pair you kept apart.
Find same work
You choose Find same work on Record, or you link a project to a role.
Reads
- Approved onlyThe project: its name, summary and stack, and the role it’s linked to
- Approved onlyThat role’s title, employer and dates
- Approved onlyThe facts of the project and of its role not yet connected, and the pairs you kept apart
May
Suggest pairs of a project’s facts and its role’s facts that describe the same work.
May not
Connect facts without you, or pair facts you kept separate.
Read a project
You choose GitHub repositories to read on Projects, or Read again on a project.
Reads
- Your wordsThe repository: its files, history and languages
- Not yet reviewedThis project’s facts, approved or waiting, so it doesn’t repeat them
- Rejected, to skipThis project’s facts you rejected, with your reasons
- Approved onlyYour directions, for what matters to you
May
Propose the project and facts about it, each citing the files it came from.
May not
Fetch anything that looks secret, like .env files, keys or certificates. It can open that one repository and nothing else.
Insights
Insights reads across your approved roles, facts, context, directions and projects, the way a career editor would, and proposes things that are true of you across jobs that you may never have put into words. Each insight cites the approved facts it rests on; an insight left with no approved fact behind it is dropped. A project's facts count as evidence just like a role's.
It runs on its own, quietly, once a story or project is fully reviewed: nothing left from it to approve or reject. It also runs when you choose Find new insights. It needs at least two approved facts. It never brings back an insight you rejected, and doesn't repeat one you already have or one waiting for review.
Skills, tools and certifications
Gather skills runs on its own once a story or project has nothing left to review (and again when you approve one of the story's roles after that), and when you choose Gather on Skills and tools in Record. It reads your approved roles' skills and tools, your approved projects' stacks and your approved facts, and proposes each skill, tool and certification once, named the way resumes and postings name it, with where your record shows it. Each item is one thing: "Radio advertising and media buying" is two skills, unless the joined name is itself one thing, like Mergers and Acquisitions. Skills are written in Title Case (Radio Advertising, Customer Onboarding, Go-to-Market Strategy), keeping acronyms and brands as they're written (SEO, B2B, iOS, HubSpot); tools and certifications keep their official names, and a name you edited stays exactly as you wrote it. A skill still waiting for review that combines two things is split into two. Vague ones are flagged as low value, and near-duplicates are offered as merges. It never merges anything you approved, and never pairs two items you kept apart. A rejected skill is never proposed again, in any spelling or casing.
Resumes list only the skills you approved here.
Duplicates
Find duplicates runs only when you choose Find duplicates in Record, or Find facts said twice in Review. It looks within each role for facts that say the same thing, often one from each of two stories, and flags them. Nothing changes until you choose: Merge keeps one wording with both facts' sources, history and notes; Keep both keeps them apart, and that pair is never flagged again. A new search replaces the last one's flags.
Same work
When a project is linked to a role, its facts and the role's facts often tell the same work twice: once from the repository, once from your story. Find same work runs when you link a project to a role, and when you choose Find same work. It reads the approved facts of the project and of that role and suggests pairs that describe the same work, the richer one as the lead.
Connect links the two facts without touching their words: resumes then write one line for them, from the lead, with the other as context. Keep separate means that pair is never suggested again. Unlinking or relinking the project drops the suggestions still waiting.
Reading a project
Read a project reads one of your GitHub repositories after you connect GitHub in Record and choose which repositories to read. Each repository is read whole, not just its code: the README, docs, planning and design files, changelogs, the stack from its manifests, a small sample of code and the commit history. It can open that one repository and nothing else.
Secret-looking files, like env files, keys and certificates, are never fetched, and anything that looks like a password, key or token inside what is read is redacted before it goes to the AI. A large repository is read in parts, then the facts from those parts are merged into at most six.
It proposes the project and facts about it, each citing the files it came from; a fact citing no file it read is flagged. It is told to write plainly: you built it, tools are tools, no inflated labels. The project and its facts wait in Review. Only approved ones reach insights, directions, role ranking and resumes. Read again on a project reads it against your record as it is now, and keeps what you approved or wrote.
Edit this page on GitHub