apply-in-browser
Skill1 starsChanged 46 days agoMITView on GitHub
Fill a job application in a web portal using the harness browser tools, the stored form-answer memory, human-paced actions, and candidate confirmations.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- find-jobsSkill
- outreach-playbookSkill
- review-loopSkill
- session-startSkill
- tailor-applicationSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 29
Never guess identity, legal, authorization, or compensation fields.
- Distinctive: line 37
If a CAPTCHA, human-verification challenge, or block page appears, stop and hand it to the candidate.
- Very specific: line 41
Keep a running log in workspace/applications/<appId>/notes.md: pages completed, anything odd, where you stopped.
- Very specific: line 48
If a confirmation number is shown, capture it too (--kind confirmation) and note it in chat.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Apply in the browser
The browser is the default way to apply. Use the harness's browser/computer-use tools when they exist. Without them, prepare answers and ask the candidate to drive while you coach.
- Open the application portal for the tracked
<appId>(see itsnotes.mdartifact for the URL and any portal quirks). - Load the form memory first:
node bin/applymate.ts answers list. For each field,node bin/applymate.ts answers get "<question>". Fill only what is known. Free-text answers you compose (essays, "why this role", summaries) are drafted from dossier facts and passed through thehumanizerskill before they go into the field — no AI-sounding prose under the candidate's name. - Portal form state rarely survives between sessions — on a return visit,
expect to refill. If the portal parses resumes (SmartRecruiters, Workday,
Greenhouse do), upload the resume PDF first and let the parser autofill
experience and education, then patch the gaps by hand. In-app browsers
cannot drive the OS file picker; the working path is the page's own upload
widget: locate the
<input type="file">(often inside open shadow DOM) and setfilesvia aDataTransfer, or dispatch a syntheticdropevent carrying aDataTransferwith aFileonto the drop zone — either lands the file and triggers the site's parse. Verify every parsed field against the approved resume before moving on (parsers drop months, merge words). - Never guess identity, legal, authorization, or compensation fields. For
unknown ones, ask in chat (
chat reply), wait for the candidate, apply the answer, then persist it:node bin/applymate.ts answers set "<question>" "<answer>". - Pace yourself like the person you are acting for. One session, one tab per site; let pages load and scan them before acting; move through multi-page flows (Next / Save & Continue) in order with natural pauses; type field by field with small pauses and the occasional correction instead of instant field-stuffing; no parallel sessions or batch-blasting.
- If a CAPTCHA, human-verification challenge, or block page appears, stop and hand it to the candidate. Never bypass, solve, or disguise your way around one — no stealth plugins, fingerprint spoofing, or proxy tricks, ever. If a site rejects automation, that site belongs to the candidate.
- Keep a running log in
workspace/applications/<appId>/notes.md: pages completed, anything odd, where you stopped. - At the final Submit: stop and ask the candidate for an explicit yes. Only the candidate submits.
- The moment it goes through, preserve the proof: screenshot the
confirmation/submit page and store it:
node bin/applymate.ts app evidence <appId> submit.png --kind submit-screenshot. If a confirmation number is shown, capture it too (--kind confirmation) and note it in chat. - Record the facts:
node bin/applymate.ts app submitted <appId> --portal "<portal>" --confirmation "<number if any>"— this stamps the exact submission time and moves the application tosubmitted.
Portal mechanics — the recipe library
Hard-won portal mechanics (resume-upload workarounds, reload behavior,
click fallbacks, LinkedIn messaging/invite recipes) live in the committed
library docs/portal-recipes/ — one file per site family, indexed in its
README. Read the file for this portal before opening the portal, and
after any difficulty you solve in here, write the recipe back into that
library and commit it (docs(recipes): <domain> — <what was learned>).
The AGENTS.md "Browser recipes" section is the standing rule: never
re-derive what the library already knows, never fork a second copy.
$ARGUMENTS
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

