Community forks & adaptations #78
Replies: 40 comments 78 replies
|
🇻🇳 Vietnam Fork: ai-job-search-vn https://github.com/jamesng16/ai-job-search-vn Working:
Blocked: TopCV (Cloudflare 403), VietnamWorks (SPA) Built from HCMC 🇻🇳 |
|
Excellent - and genuinely glad to see this. Going from yesterday's closed PR straight to a working ITviec skill with passing tests in your own fork is exactly the arc this policy hopes for. Adding you to the list in the opening post. If ITviec's markup ever shifts and you find the /add-portal investigation notes lacking, feedback welcome. |
|
Thanks for setting up this discussion, related PR link: #67. I have published a China-mainland focused fork here: https://github.com/youngledo/ai-job-search-cn This version may be a better fit for Chinese users who want a localized workflow while still staying close to upstream. The main reason is that the country/market-specific behavior is kept isolated under a China market layer instead of rewriting the core upstream workflow. That gives it a few practical advantages:
The goal is not to push China-specific behavior upstream, but to keep a maintained Chinese version that benefits from upstream while respecting the market-specific boundary. |
|
Thanks for following through on this so quickly after #67 - a China-mainland adaptation with Chinese templates fills a real gap, and it's now listed in the opening post. Held to your own standard from the PR (gitignore discipline around personal job data), which deserved better than a plain decline. |
|
🇪🇸 Spanish market edition — https://github.com/erikpr1994/ai-job-search A fork adapted for the Spanish job market: adds portal search skills for InfoJobs, Tecnoempleo, and GetManfred (LinkedIn + freehire stay as country-agnostic defaults), with Profile files remain upstream placeholders — clone and run |
|
Checked the fork before adding it to the list: InfoJobs, Tecnoempleo, and GetManfred all follow the portal-skill contract (CLI with tests and a url-reference each), LinkedIn and freehire stay as the country-agnostic defaults, and the profile files remain upstream placeholders. This is exactly the shape the fork policy hopes for - additive portal skills on an unchanged core, so rebasing stays cheap. Adding you to the opening post. Also: your two Ideas (#92, #93) are the most carefully scoped proposals this repo has received - replies coming in those threads. |
|
🇧🇷 davicalvarenga/ai-job-search — Brazil adaptation
Thanks for the framework, @MadsLorentzen — hoping this helps Brazilian job seekers get started. 🙌 |
|
Welcome, and thanks for the thorough post. I looked through the fork: Gupy in particular is a smart pick - covering the ATS that hosts a large share of Brazilian postings rather than only job boards. Added to the opening post. 🇧🇷 |
⚙️ Codex adaptation: trikto/ai-job-searchThis fork adds Codex compatibility to the original What it adapts
Upstream trackingThe fork remains closely aligned with upstream. The adaptation is additive, with I plan to keep the fork updated with upstream improvements while maintaining the Codex compatibility layer separately. Thanks for providing a clear place to share community adaptations! |
|
Thanks - a Codex compatibility layer is the first alternative-harness port to land in this thread, and the design is the right one: thin command-named skills that each point at One heads-up: the layer currently lives on the |
|
@trikto, following up on this a couple of weeks late with something I should have spotted when I listed your fork. Two of the profile files on your The scope is narrow, which is the good news. I checked the rest and everything else is clean: Two separate decisions, and they're genuinely separate: Whether you want it public at all. If yes, no action needed, plenty of people put their CV online deliberately. If not, two things are worth knowing before you go looking for the obvious fix. First, deleting the content in a new commit is not enough, it stays in the history and stays reachable; GitHub's guide is removing sensitive data from a repository. Second, and this one catches people out: a fork of a public repo cannot be made private. GitHub's own docs put it plainly, "You cannot change the visibility of a fork by itself." So the settings toggle you'd expect isn't there, and the practical route is a history rewrite, or moving the work to a fresh private repo. What happens to your index row in the meantime. It stays up unless you'd rather it didn't. If you push a branch with the placeholders restored, tell me and I'll point the row at that instead, which is the tidiest outcome since the Codex work is what people are coming for anyway. Or if you'd prefer I pull the row while you sort it out, just say so and I'll take it down, no explanation needed. To be clear about where the fault sits: The Codex adaptation itself is unaffected by any of this and stays listed either way. |
|
May I present: This two make the skills harness agnostic. As for the folder structure, one skill can be added which initialises it. |
|
🇵🇱 Poland Adaptation (Pracuj.pl & JustJoin.it) Hi everyone! Disclaimer: I did this primarily for my personal use to quickly start applying, so I modified the core template files directly. This means it might be difficult to pull future upstream updates into this fork without conflicts. However, if you are from Poland and just need a working, out-of-the-box solution right now – it works great! Anyone is free to use it, or take my code and refactor it into proper isolated skills if they have the time. Happy job hunting! Link to the fork: https://github.com/pandawidkur/ai-job-search-pl |
|
Thanks for sharing, and especially for the honest disclaimer - "I modified core files directly, so upstream pulls will conflict" is exactly the information a would-be user needs, and most posts wouldn't volunteer it. Pracuj.pl and JustJoin.it are the right two boards for Poland (the generalist giant plus the IT-specialist one), so as an out-of-the-box snapshot for Polish job seekers this fills a real gap. Adding it to the opening post with a note that it's a working snapshot rather than a tracking fork. If you (or anyone reading) ever want to refactor the two portals into isolated contract-shaped skills so the fork can rebase cleanly, |
|
🇺🇸 chenyuan99/ai-job-search by @chenyuan99 — US market adaptation
|
|
Welcome - and a US adaptation fills a real gap in the list. I checked the fork before adding it: One heads-up before people come looking: the skill currently lives only on the |
|
@MadsLorentzen Thanks Mads, I just have it merged |
Google Antigravity (AGY) Adaptation (Gemini-powered local run)I have created an adaptation for running this framework using Google Antigravity (AGY), providing a cost-effective, Gemini-powered alternative to Claude Code for scraping, triage, and tailoring.
|
|
I think creating separate forks for each LLM/agent tool might lead to a combinatorial explosion. If we have one fork per LLM/agent framework (e.g., Claude Code, Google Antigravity, etc.) and then country-specific forks for each of those (e.g., Spain, Germany, etc.), we will end up with a recursion of forks that is impossible to maintain and track. Since |
|
You've named the real problem with the fork model, and more precisely than anyone so far: tools × markets is a matrix, and forks can only ever pick a cell. The Poland fork and the AGY adaptation cannot compose today, and that's a structural gap, not a fork-quality issue. Two parts of the proposed fix don't survive contact with this repo, though. Symlinks break on Windows - git checks them out as plain text files unless the user has enabled developer mode and But there's a one-file version of your idea that I think is right: a root If that shape sounds right to you, PR welcome - you know the contract by now. |
|
Fair on the Windows symlink breakage and the drift risk of parallel config trees — I'll drop the symlink angle. One correction on the mechanics, though, because I think it actually reinforces the case for fixing the split rather than just pointing at it. There isn't really a namespace collision to design around. The sharper mechanic is almost the reverse of a collision: Concretely — here's Antigravity CLI surfacing one of the portal skills straight from 
Net effect, the workflow is split across two roots that no single runtime reads together:
So So I don't think a pointer file alone closes this — an The honest catch is that no single folder is read by all four: put everything in |
|
One more shape, in case the answer to "who owns the portals" is that they shouldn't be standalone skills at all — i.e. the goal is that no agent surfaces In that case, rather than picking a canonical skills root, you'd fold the portals into This tells the truth about how they already work — The trade is the mirror image of the "canonical So really two clean ends of the same decision: (A) portals are shared skills → make Having thought it through, I'd lean B together with your
|
|
First, the correction you're owed: you're right about That correction is also what settles the direction, though - and it lands on neither A nor B. Working through #161 (which did exactly what an RFC should: made the trade concrete enough to evaluate on evidence), the deciding fact is one the diff can't see from inside this repo: the pinned index above lists 15 portal skills across 8 community forks, every one at What your evidence actually convinced me of is that the split should stop being an accident and become the design:
The residual cost of the split after the pointer lands is "other runtimes reach I'm closing #161 with thanks - it was the cheapest possible way to buy certainty on a direction question, which is precisely what a draft RFC is for. The analysis in this thread is what forced the architecture to be decided explicitly instead of by drift, and that credit lands in the merge when the AGENTS.md goes in. |
⚙️ OpenCode adaptation: Affan-p/ai-job-searchCurrently on the Codex, Antigravity, and Gemini CLI all have adaptations in this thread, but OpenCode doesn't yet. OpenCode reads What it adapts Adds an opencode.json mapping all 11 workflow commands to OpenCode:
Keeps the existing Claude commands and workflows unchanged. Upstream tracking The fork remains closely aligned with upstream. The adaptation is additive, with |
|
Welcome, and verified + listed. I opened Worth saying out loud: yours is the first adaptation to arrive after the root On your question: master preferred, yes. The index carries "work branch" notes when needed (trikto's Codex layer has one), but forks get cloned directly, and master is what people land on. I've listed you now with the branch note - say the word when it's merged and the note comes off. One observation, not a blocker since it's your fork's call: your Thanks for a clean, honest announcement - claims that survive verification make listing easy. |
|
https://github.com/emanuelet/ai-job-search in case anyone can benefit from it, I have been keeping my fork in sync as well, with a thin layer to have compatibility with opencode and pi. |
|
Merged to master. The permissions block is now a curated allowlist with a Also saw @emanuelet's fork with OpenCode + Pi compatibility, had a look at the commits behind it. Mine stays focused purely on the |
|
rasstamann/ai-job-search-germany by @rasstamann - German/DACH market adaptation
Main functionalities have been tested. So far, only Xing allows data scraping, but at reason. They don't outright prohibit data scrapers, but it is stated that it should be used sparringly. |
|
Verified and listed - and I want to point at the part of this fork other portal builders should copy, which is not the part that worked. Your Same for the One connection you'll want: @StarOnDavid built two more German portals last week (#162 - closed only because it targeted upstream, where market skills don't live): arbeitsagentur-search on the official Bundesagentur für Arbeit API (government job board, static public key, no scraping questions at all - and Germany's largest portal) plus arbeitnow-search. Between your Xing skill and that official-API work, the DACH market would have real coverage with clean consciences all around. If the two of you combine forces - either fork adopting the other's portals, per the index's read-the-code rule - I'll gladly update the rows. Given what you wrote in #140 about your own search, having the Bundesagentur's full listing flowing into your Happy hunting - in both senses. |
|
🇳🇱 Dutch market edition: https://github.com/Fabian-Schreuder/ai-job-search-nl A Netherlands-focused fork built directly on top of upstream MadsLorentzen/ai-job-search. What it adapts
Language / market
Tests / CI
Personal use only
Tracks upstream: closely; one Dutch-market delta commit on top of a recent upstream head. Feedback welcome. |
|
Welcome, and verified - this is one of the cleanest fork announcements this thread has had. I checked the tree against every claim: the skill is real (zero deps, single public-API endpoint, mocked-fetch tests, and the personal-use/rate-limit note is exactly the standard this thread has converged on), the Danish demos are disabled via the One question before I add the index row, purely provenance: this fork belongs to @Fabian-Schreuder while the announcement comes from @3dhq, a brand-new account - same person? Every row in the index is an implicit recommendation to the people who clone from it, so I confirm who's behind a fork before listing. (No concern about the content itself - it screened clean.) Confirm and the 🇳🇱 row goes in. |
|
Yes, same person — @3dhq is a newer account I used to post the announcement, but the fork is under @Fabian-Schreuder. Happy to post future replies from the @Fabian-Schreuder account if that makes the index cleaner. Thanks for the careful verification and the kind words. |
|
Listed - both the adaptations entry and the 🇳🇱 portal-index row are in, credited to @Fabian-Schreuder. Posting from either account is fine going forward; the index credits the fork owner either way. First fork to ship with the |
|
Index refresh - the forks index is growing 🌍 We're up to 11 adaptations across ~9 markets (China, Vietnam, Spain, Brazil, Poland, US, Germany/DACH, Netherlands) plus Codex and OpenCode harness ports - all from people who found this thread and shared. If you maintain a market or language fork that isn't listed yet, this is the nudge: reply with your fork link + what it adapts (format in the opening post), and it'll be added to the index and the portal-skill table. The whole point is that a job-seeker in your market can find your work instead of starting from scratch. One reminder now baked into the opening post: share a clean branch, not your filled-in profile - |
|
Two adaptations worth surfacing 🇳🇴 🇮🇳 - spotted while going through forks, inviting them here rather than listing without the maintainer's say-so. @ThomasHjartarson - I came across your fork ai-job-search-no and noticed you built @ashutoshgh - I had a look at your fork ai-job-search-india and you've done substantial work: skills for Apna, Hirist, IIMJobs, Instahyre, Indeed India, Randstad India, and more. That's easily the most complete Indian-market adaptation I've seen, and India's a market a lot of people ask about. Would you consider sharing it here? Just reply with your fork link + a line on what it covers, and it goes into the list and the per-portal table. (Aside: someone once proposed a One protective heads-up for both of you: |
|
Verified and added, thanks Thomas 🇳🇴 nav-search and nav-feed-sweep are in the index and the portal table now. And I appreciate the finn.no call, that's exactly the right instinct: leaning on arbeidsplassen.nav.no (which republishes most of finn's ads and links back) instead of crawling a board whose robots.txt says no. First Norwegian adaptation in the list, genuinely useful. |
|
Thanks for the shout-out! Sharing here: ai-job-search-india. Adds India market portal skills on top of upstream: |
|
Thanks for sharing it, and sorry for the slow reply. That's the most complete single-market set anyone has built on top of this so far, ten portals including the remote boards, and I checked: they're all really there on Before I add the index row, one thing I want to flag, and I'd rather say it here quickly than sit on it. Your fork is public, and three of the framework files on The good news is that it's contained. I looked, and the rest is clean: no real CV or cover letter committed, no Two separate decisions here, and I'd treat them separately: Whether you want that public at all. If not, be aware that deleting it in a new commit isn't enough, the data stays in the history and stays reachable. GitHub's guide for this is removing sensitive data from a repository. Your fork is only a few days old, so the rewrite is about as painless as it ever gets. Entirely your call, plenty of people put their real CV online on purpose. Whether I list it here. This is the part I do need to hold. A row in this thread sends a steady stream of people to your repo to clone it, and I'm not comfortable pointing them at profile files carrying someone's personal details, both for your sake and because people copy what they see and assume it's the intended pattern. So: push a branch with the placeholders restored, tell me which one, and I'll point the index at that. Farley did exactly this for the Colombia fork and it works fine. Honestly, some of this is on the framework rather than on you. Once there's a clean branch I'll do the usual verification pass on the portal CLIs and get the index row and the portal-table entries up. The work deserves to be listed. |
|
@ashutoshgh one practical addition to the above, so you don't waste time looking for a setting that isn't there. If you decide you'd rather the profile files weren't public, the obvious move is to flip the repo to private, and that option doesn't exist for a fork. GitHub's docs put it plainly: "You cannot change the visibility of a fork by itself." So the realistic routes are a history rewrite on the existing fork (the sensitive data removal guide covers it, and yours is only days old so it's about as cheap as that ever gets), or pushing the work into a fresh private repo and keeping the public fork for the portal skills alone. That second option is arguably the nicer split anyway: the India portal skills are the part worth sharing, and they don't need your profile data to be useful to anyone else. Nothing else changes from my earlier message, the index row is still yours whenever you've got a clean branch to point it at. |
|
@MadsLorentzen thanks for flagging this, and for being careful about not quoting it back. Fixed: rewrote history ( Also went back through every file that differs from upstream (not just the three you named) to check nothing else slipped in - tracker CSV, cover letter
One heads-up: since it is a history rewrite, old commit SHAs (including the one I referenced fixing the Outlook scope earlier) are gone - happy to re-confirm anything against the current |
|
@FARLEY-PIEDRAHITA-OROZCO - going through forks I came across ai-job-search-opencode, and you've built Would you share it in this thread? A reply with your fork link + a one-line "what it adapts" is all it takes, and it goes into the index and the portal-skill table so people in Colombia and neighbouring markets can find it. Same protective heads-up as above: point people at a branch where |
|
🇬🇧 UK adaptation — ai-job-search-uk Repo: https://github.com/anjolok1997/ai-job-search-uk Thanks for building this, Mads — I forked it and adapted the job boards and evaluation gates for the UK market. The core /setup → /scrape → /apply workflow is unchanged; everything below is additive. What's different:
All portal skills follow the same search/detail contract as upstream, with tests, so they're drop-in for anyone who wants just one UK board. Full credit for the framework goes to you —happy for others to borrow from it. |
|
Verified and listed - index row plus portal-table rows for reed-search, totaljobs-search, and wttj-search are up. As with every #78 listing, I cloned the fork and screened it before answering: lineage, the full delta vs upstream, a per-CLI code read (dependencies, endpoints, env reads, file writes), robots.txt posture per portal, a personal-data scan at HEAD and across history, placeholder integrity, and your guards and lint run in-tree. Clean across the board: zero-dependency CLIs, endpoints exactly the claimed portal hosts plus gov.uk for the sponsor register, no personal data anywhere in any branch's history, placeholders intact, and the attribution in your README is the nicest I've seen in this thread. The Sponsorship Gate is genuinely well designed - opt-in, "silence is not refusal" stated in the gate itself, graceful fallback when the register is unreachable - and you even followed this repo's framework_version convention on the profile files you touched, which almost nobody does. The cherry-picked upstream fixes tell me you'll keep tracking; appreciated. Three asks, none of which held up the listing - the row carries the first two as caveats until they're fixed, and I'll drop the caveats when you ping:
Welcome to the index - first UK adaptation, and a strong one. |
|
Thanks for the review and for adding the fork to the index! Both flags were fair — I've fixed them:
All covered by regression tests. I also added a small weekly GitHub Action on my fork that triages upstream commits I haven't picked up into a rolling issue — report-only, it never auto-merges, so I stay in control of what gets ported. |
|
Verified and both caveats are now dropped from your index rows. What I checked on Fairness note: the one remaining browser-style UA in your fork is Your upstream-watch tooling is a genuinely nice build, by the way - see my note on #300 about pointing its |
|
Thanks! #300's sorted: root cause was that it authenticated with a cross-repo PAT, so I've switched it back to the built-in GITHUB_TOKEN (plus pinned the target repo). It can't write outside my own fork now. Sorry for the noise. And I took you up on the triage-tool idea — opened a discussion to settle the shape before I build anything: |
|
🇺🇸 US new-grad adaptation — verification-first + referral sourcing ( Repo: https://github.com/ayushsawnani/job-search-project TL;DR: I reworked it from an on-demand "tailor my CV" tool into an autonomous, verification-first What I added
Guardrails I kept strict: never auto-submits, never auto-connects/messages, never fabricates a Stack added: Claude in Chrome (browser automation), scheduled tasks, per-posting subagents. Happy to break any of these down if useful to others building US/new-grad variants. |
|
Thanks for sharing this, Ayush - I read through the repo properly before replying (I do that for everything announced here, since a listing in this thread is an implicit endorsement). The engineering is real: the verification gate, the four-agent council, the referral drafting, the backlog-promoting daily loop - all of it exists as described, the guardrails you list are actually hardcoded where they claim to be, and the security screen came back clean (no dependencies, no auto-send code anywhere, the only outbound call is your opt-in Telegram notify with the token strictly from env). The cost-ordered verification pipeline and the "one read per page, isolate each posting in a subagent" discipline are genuinely good ideas. Three things before I add it to the index, two of them just wording, one worth a decision on your side:
Ping me here once you've decided on 3 (either way - "I'm fine with it staying" is a valid answer for your own data) and I'll add the row with the adaptation framing. Nice work - the dead-posting/hallucinated-question failure mode you're attacking is real, and catching a prompt-injection bot-trap in the wild is the best validation of the verification-first design anyone could ask for. |
|
Thank you for checking it out, Mads! After looking at my git history, I also thought the best approach was to delete my history and re push everything, which I have just done (you can double check if you'd like). Appreciate the endorsement! |
|
Appreciated the fast action - and since you invited a double-check, I did one, and I have to be straight with you: the squash cleared
To be clear on scope: everything at your current HEAD (the resume, the tracker) reads as deliberate self-publication and is your call - this is only about the pre-scrub history you already decided you wanted gone. On listing: I'll add the index row - framed as an independent adaptation with a browser-automation note, per my earlier comment - as soon as the branch in point 1 is deleted. Ping me here when it's gone. The Support ticket in point 2 is worth doing but I won't hold the listing on GitHub's queue. |
|
Ah, thanks for that catch! Point 1 is deleted. I'll look more into Point 2, but for now it doesn't matter that much. |
|
Verified (only |
|
Runtime fork: MiMo Code + Qwen3-Coder-Next Added support for running this framework on MiMo Code (Xiaomi's OpenCode fork) with Qwen3-Coder-Next as the coding agent, for anyone who wants a fully local/self-hosted alternative to Claude Code. Following the thin-pointer approach from CONTRIBUTING.md: the .mimocode/ directory holds MiMo Code's own agents/commands, but they point back to the canonical specs in .claude/ and AGENTS.md rather than duplicating them — so upstream methodology changes stay reachable via check_upstream_updates.py like any other fork. What's added: .mimocode/commands/ — command definitions mirroring the core lifecycle (/setup, /scrape, /rank, /apply, etc.) for the MiMo Code runtime Fork: https://github.com/CoNFiDeNTHaKaN/ai-job-search (branch: mimocode-integration, set as default) Happy to answer questions if anyone's running local LLMs (Qwen, GLM, etc.) via llama.cpp and wants to try this route instead of Claude Code. |
|
Verified and added to the index, thanks @CoNFiDeNTHaKaN ⚙️ - first local-model runtime in the list. For the record, the pre-listing screen held up unusually well: the delta is exactly what the post claims (2 commits ahead of current master, 0 behind - the only fork announce so far that arrived fully up to date), |
|
Thanks for the thorough verification, means a lot coming from someone who actually checks rather than takes the claim at face value! Happy to be the local-model reference point if anyone else wants to go that route. |
|
Makes sense - "nothing MiMo-specific to write because the skills format is already portable" is a stronger property than a port, and worth stating outright, so yes to that AGENTS.md line. And noted on the reference-point offer: anyone landing in this thread asking about local models via llama.cpp, this fork is the working example. |
|
My comment isn't about my fork, but rather about bunch of your forks and inspired projects. I've been building an index of repos you fork and work inside, like this one. I call them agentic workspaces. agentic-workspaces.com This project has one of the biggest families in the index (other repos that have this one in its core), 124 copies and adaptations: https://agentic-workspaces.com/family/madslorentzen-ai-job-search. Each row shows what that copy added on top of the original. 58 of the 124 built something the original doesn't have, and 32 added support for a different agent. It finds them by structure rather than the fork graph, so it catches copies that were never forked from here. The loose matches at the bottom are guesses. |
|
That's a genuinely interesting index - structure-matching rather than the fork graph explains why it sees copies the network view misses. 124 is a bigger family than I knew about. Leaving the link here for readers; as with everything in this thread, entries in the index aren't individually vetted by me, so apply the usual judgment before running someone else's workspace. |
|
Typst CV + cover-letter templates (bring-your-own; bundled LaTeX stays) The pair #351 asked for, published separately — not a request to replace moderncv. • Clean copy (placeholders only): https://github.com/sdrarunvarshan/ai-job-search (typst-modern/) /add-template typst-modern/cv/template.typ Compile stays typst compile .typ .pdf. /add-template --use default restores LaTeX. • Tracks upstream: not a framework fork — independent MIT templates After the ATS PR is open, reply on #367 that you took those two doors and left FCC and portals out and also i have combined the waleedtarbosh/ai-job-search Universal Edition (OpenCode / Gemini / Codex pointers) and 🇮🇳 ai-job-search-india by @ashutoshgh India + remote portal skills in my workflow |
|
Verified and happy to point people here: both templates are placeholder-only (no personal data, no external imports), so this is exactly the pair #351 was missing. Two transparency notes for readers: I haven't compile-tested them myself - |
|
🇨🇴 Colombia / Latin America adaptation — oscarbol09/ai-job-search (Computrabajo)
|
|
Thanks for announcing this, and for following the portal-skill contract closely. I've verified the branch before listing (standard practice for #78): all announced files exist under 🇨🇴 Colombia / LATAM users: this is the door |
|
Hey @MadsLorentzen! A quick update on the Colombia / LATAM adaptation branch ( We have expanded the suite to cover the entire Colombian job ecosystem, now providing native portal skills across the public, corporate, and tech sectors:
All 5 portal skills maintain zero runtime dependencies, no lifecycle scripts, conform strictly to the Everything is live, documented in |
|
Read the branch at
One correction to the post: the branch is ahead 6 and behind 3 of master as of this afternoon, not "up to date" - three merges landed today. Harmless, but the row pins a commit, so rebase before I pin it, or I pin |
|
Thanks @MadsLorentzen! Addressed both points and rebased the branch:
The branch is now up to date with |
|
ticjob is in. The robots note matches the live |
|
Hi everyone — I’m Louis, an undergraduate student at The Ohio State University with an interest in AI tools, developer workflows, and open-source software. I built this Codex port while exploring how mature Claude Code workflows can be adapted cleanly to other agent runtimes without compromising traceability or user privacy. Project: ai-job-search-codex-port What it adapts: An unofficial Codex plugin port with an installable plugin manifest and marketplace layout, Codex-native orchestration over the upstream canonical workflows, and a private workspace initializer that keeps candidate data outside the installed plugin. LinkedIn and FreeHire are enabled by default, while the four Danish portals remain opt-in. Implementation: Portal launchers execute the checked-in TypeScript source directly, with no prebuilt bundles or binaries in the repository. Upstream tracking: The port follows the upstream canonical workflows closely and preserves the MIT license and NOTICE attribution. Thanks @MadsLorentzen for the thorough review and for adding the port to the community index! Contact: chen.14774@osu.edu |
|
🎨 Into Design Systems - a portal indexed by specialization, not by market Repo: https://github.com/SimonDiff/into-design-systems-mcp-adapter What it adapts: one portal skill, Transport: the board publishes a public, read-only MCP server, so the CLI speaks JSON-RPC to it rather than parsing HTML. No account, no API key, and no robots.txt question: the endpoint is the documented interface. Direct MCP is the better choice if you just want to chat with the board; this exists so the results land in How closely it tracks upstream: not a fork. A standalone MIT repo holding just the skill folder, so the Path below points at Nothing to scrub: the repo has never held profile data. There's no Two deviations worth knowing about when you look:
Suggested row for the portal index:
Credit where it's due: the board and its MCP server are built and maintained by Sil Bormüller of Into Design Systems. I'm not affiliated with them, just a user who wanted the board in the |
|
Verified and listed - welcome to the index, and with it a first: this is the index's first portal scoped by specialization rather than market, so your suggested "global (Design System roles)" phrasing goes into the Market column as-is. If more specialization-scoped boards arrive, the section may grow its own shape; one row doesn't need a new taxonomy. What I checked (the standing caveat applies: I read every file but did not run the code, and the row endorses the reviewed state, commit Three small notes, for you and for anyone copying the skill:
Thanks for the cleanest external adapter submission this thread has had - the honest deviation notes and the "nothing to scrub" framing are exactly what the format above asks for. |
This comment was marked as spam.
This comment was marked as spam.
|
@MadsLorentzen Thank you for the detailed review. All three notes have been addressed:
All three are For the record, the commit you reviewed ( One update that may interest other fork authors, since you highlighted the transport choice. While building this I found two bugs in the board's date filters, I reported them to the MCP's owner, who fixed them immediately, so this is all reflected in the docs. Thanks again for building the index and the contract. Having a spec to conform to is most of the reason this was straightforward to write. |
This comment was marked as spam.
This comment was marked as spam.
|
🇮🇳 India fork, blue-collar focus: https://github.com/smartbriefai/ai-job-search-india-bluecollar Working: ✅ shine-search - Shine.com (HT Media), mainstream white-collar, all major cities. 27 tests Blocked: NCS (API needs encrypted payloads, returns Invalid or unencrypted request payload), Naukri (Cloudflare + fingerprinting), Indeed India (robots.txt names ClaudeBot on /jobs), Internshala (robots.txt blocks Anthropic bots by name, technically allowed under * but skipped), Cutshort (public API is recruiter-side only, no candidate job search exists), Quikr (robots.txt prose prohibits automation), Freshersworld (Disallow:/jobs/jobsearch/), Foundit / TimesJobs / HerKey (client-rendered), Unstop (CloudFront 403 on the whole domain). Deviations, both in workindia-search. Its robots.txt disallows every URL containing ? or &:
Worth flagging: an unrecognised city slug 301s to WorkIndia's nationwide page, so "lakshadweep" quietly returned twenty jobs in Delhi and Hyderabad. The skill now compares the post-redirect URL and fails loud. Caught in a pre-publication audit, not by tests. Tracks upstream: merged onto current master (b959d6a). Danish portals stay disabled, linkedin and freehire stay enabled. @ashutoshgh's India fork is already listed and this isn't a replacement for it. Zero runtime dependencies, honest UA tokens, fixtures scrubbed of recruiter contact details before the branch was ever pushed. Adds two lines to the Bash allowlist and the matching entries in security_guards.py. Thanks for the nudge to build a portal skill rather than take a micro-task; the investigation taught me far more. |
|
Verified, and this is a strong one. What I checked, read-only - I read the code and audited the repo, I didn't execute the CLIs:
Two caveats for anyone borrowing, neither of them a fault of yours:
Listed. Blue-collar and grey-collar hiring genuinely wasn't represented here, and the investigation section - ten portals ruled out, with the reason for each - is worth more to the next India fork author than the two skills are. |
|
🇨🇳 China-market adaptation: offer-harvester A China-market adaptation, restructured as a standalone repo (fresh history, no shared git lineage with upstream, so update triage via What it adapts
Honestly disclosed limits
Tracks upstream: additively, with intentional divergence — Privacy: no personal data in the repo — Thanks for the framework — the drafter-reviewer |
|
Read the repo. Starting with the one thing that needs your attention today, because it is your data, not mine. Your real name is committed in five CV files. Worth knowing: this is the exact failure mode upstream's placeholder CI check was rewritten to catch. The old sentinel sat in a header comment, which a personalization pass never touches, so a fully personalized CV passed the check. The fix was moving the sentinel into the The good news is the timing. Your repo is one commit old. Amending that commit and force-pushing genuinely removes the content, which would not be true in a fork, where objects stay reachable by SHA in the fork network. If you would rather be certain, deleting and re-creating the repo is the belt-and-braces version. I would do it before this gets more traffic. On the rest, and I mean this: the submit-gate engineering is the most careful I have reviewed here.
Three things to fix or restate before I would list it:
None of that is a judgement on the work, which is careful. Scrub the name first, tell me when it is done, and we can settle the listing question then. For what it is worth, a fork is allowed to go further than the template does, so form-filling with a real human gate is not disqualifying here the way it would be in an upstream PR. |
|
Done — thank you for the careful review, and especially for catching the name.
The read on the submit gate means a lot — it was the design center of the pipeline. |
|
A quick follow-up: the repo went private again briefly after your review. A second pass found identity fingerprints beyond the name — education-city trail, publication counts, bare award names — hiding in the CV template bodies, so the personalized files were replaced wholesale with pristine upstream versions rather than redacted ones (your round-1 lesson, applied properly this time: partial redaction of personal files keeps the fingerprint). The upstream test suite came back with them, It is public again at the same URL, rebuilt as a brand-new repository with a single-commit history — the earlier commit SHAs no longer resolve. Happy for a re-verification whenever convenient. Both of your catches have permanently changed how this repo scans for personal data; the fragment-level patterns from this round are now part of its pre-push checklist. |
|
Re-verified at Listed, with the caveat #78's new criterion 5 asks for. Thank you for taking a hard first review this well. |
|
Thank you for the re-verification and the listing — and for a review process that genuinely made the repo better; both rounds caught things my own scans had missed or rationalized. "The gate in the code is the bar" is the right standard to build against, and it is now written into the project's own CI as a permanent guard. Honored to be in the index — and thanks for screening the two later commits too. |
|
🇵🇱 Polish job boards: justjoin.it and nofluffjobs.com Hi @MadsLorentzen, First of all, thanks for the great software you're building — I use it myself and it's already helped me a lot. I'd like to contribute something I've been using that might be useful for others: support for two Polish job boards, justjoin.it and nofluffjobs.com. Each board is implemented as other job portals, no dependencies, works out of the box. The previous Polish implementation I found here (pandawidkur/ai-job-search-pl) returns a 404 now, so mine can be used instead: janlukowicz/ai-job-search Best regards, |
|
Verified, and listed. Both skills read as first-class portal skills: zero runtime dependencies, no lifecycle scripts, honest identifying UAs, You are right about the previous Polish fork: |
|
Here is the Docker Sandbox environment https://github.com/methanoya/MadsLorentzen_ai-job-search in case someone is looking for that |
|
Verified and listed - thanks for taking the fork route. What I checked at 4a10e72: the fork is upstream master plus the |
|
AIPayList portal skill (
|
|
Verified at The row carries three caveats so readers know what they are picking up: first-party skill from the board's operator, listings monetised by referral links on the site (not in the feed), and the roles are AI-training and evaluation contractor gigs priced "up to $X/hr" rather than employment. Your |
|
🚀 Google Antigravity & Global Tech Edition (ai-job-search-global) Fork: https://github.com/Faisa1khan/ai-job-search-global Thank you @MadsLorentzen — the workflow specs are the reason this port was mechanical rather than speculative. Replying in the format the thread asks for, with the details underneath:
What is in it
|
|
Updated the entry above to match the format the thread asks for: the two automation surfaces and where their human gate sits, what the fork adapts, and how closely it tracks upstream (c7bd494, 2026-09-10 — 12 commits behind, divergence measured file by file). Fork: https://github.com/Faisa1khan/ai-job-search-global |
and others
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
This template is deliberately kept universal: market-specific portals, language adaptations, and alternative-harness ports live in forks rather than upstream (see the policy discussions in #31, #39, #49, #67). This pinned thread is where those forks get discovered.
📣 Built a market or language adaptation? Add it here. The list below is just the forks that found this thread - there are surely more out there. If you've adapted this for your country, language, or job boards, drop a reply (format below) and it gets added to the index so other job-seekers in your market can find it.
Before you post: what gets checked. A fork link here is an implicit recommendation to everyone who reads this thread, so each one gets its contents opened and read before it is added. Four things:
A clean, shareable branch. This one is for your protection, not mine:
/setupfillsCLAUDE.mdand the profile files with your real name, contact details, CV content and salary expectations. Point at a branch where those are back to the[YOUR_*]placeholders, and check thatseen_jobs, the tracker and any real CVs aren't committed. Don't lean on CI for this - theplaceholder-integrityjob is deliberately upstream-only, since it would fail on every personalised fork by design. History counts too, not just the tip. The line that gets missed is inside the data, not the contact block: a CV whose email and phone are back to placeholders but still carries a real name iname{}, with a real education and work history around it, identifies you just as well. Before you post, rungit grep -n "<your surname>"on the branch and read every hit, CVs included. And if real data ever made it into a commit on a fork, a stripped tip is not enough - objects stay reachable by SHA through the fork network even after a force-push - so build the public branch from a history that never held it, and keep the personalised workspace on a private remote (SETUP.md section 8). A standalone repo is the one case where amending the commit genuinely removes it.The portal skills exist and follow the contract.
search/detailcommands,--format json|table|plain, JSON errors on stderr with exit 1, backoff on 429/5xx, zero runtime dependencies, and their own tests. Deviations are fine, just declare them in your post - "this board has no pagination, so--pageisn't supported" reads much better than me finding it later.A safety pass. No install or lifecycle scripts, no surprise dependencies, no requests to anything beyond the portals you declare, nothing reading environment variables or the filesystem beyond what the skill needs. Boring in practice, and the reason a listing isn't instant.
Honest scoping. Licence and attribution intact; a personal-use note where a portal's ToS calls for it; anything metered or needing an API key shipped
enabled: false; and a line on how closely you track upstream, so readers know what they are picking up.Any automation that fills or submits applications. The template itself stops one step short of the send and always will. A fork may go further, and gets listed with a caveat in its row when the last touch is provably human in the code: an unconditional confirmation before every submit, with no flag, environment variable or batch mode that skips it; single-recipient sends only, never bulk; and no captcha, QR-code or bot-protection bypass anywhere. "The user reviews before it sends" in a README is not the bar - the gate in the code is. Say plainly in your post what the automation does and where that gate sits, and it gets read first.
None of this is an "is the code good enough" bar. It is about you not publishing your own data by accident, and this index not recommending something nobody read.
Maintaining a fork? Reply below with:
A link to your fork
Any automation beyond search (form filling, sending), and where its human gate sits
What it adapts (market/country, language, job boards, agent harness)
How closely it tracks upstream
Known adaptations so far:
🇨🇳 ai-job-search-cn by @cypggs - Chinese-language adaptation
🇨🇳 youngledo/ai-job-search by @youngledo - China-mainland market overlay (Chinese templates and workflows; from PR Add China market job-search overlay #67)
🇻🇳 ai-job-search-vn by @jamesng16 - Vietnam adaptation with an ITviec portal skill
🇪🇸 erikpr1994/ai-job-search by @erikpr1994 - Spanish market edition (InfoJobs, Tecnoempleo, GetManfred portal skills; localized README/SETUP)
🇧🇷 davicalvarenga/ai-job-search by @davicalvarenga - Brazil adaptation (Gupy and Vagas.com portal skills, full PT-BR guide)
⚙️ trikto/ai-job-search by @trikto - Codex harness compatibility layer (command-named
$skills pointing at the Claude workflow; currently on theagent/codex-command-skillsbranch)🇵🇱 ai-job-search-pl by @pandawidkur - Poland adaptation (Pracuj.pl, JustJoin.it); working snapshot that modifies core files rather than tracking upstream Repository no longer exists (checked 2026-09-07); see the Polish fork below for a maintained alternative.
🇵🇱 janlukowicz/ai-job-search by @janlukowicz - Poland adaptation (
justjoin-searchover justjoin.it andnofluffjobs-searchover nofluffjobs.com; zero runtime dependencies, 40 and 47 test cases, honest UAs, both skills stay off the paths each site's robots.txt disallows and document which. Verified 2026-09-07 atdc79ebe)🇺🇸 chenyuan99/ai-job-search by @chenyuan99 - US market adaptation (swelist-search skill over SimplifyJobs new-grad/internship data; currently on a work branch, master pending)
⚙️ Affan-p/ai-job-search by @Affan-p - OpenCode harness compatibility layer (
opencode.jsonmapping all 11 commands to the Claude workflow specs, rootAGENTS.mdas instructions; curated permission allowlist)⚙️ J-Argus by @MadCkull - Google Antigravity harness compatibility layer (15 thin pointer skills in
.agents/skills/wrapping the 12 commands and 3 core workflows, plus aTOOL_GLOSSARY.mdtranslating Claude Code tool names to runtime-neutral equivalents; zero changes to.claude/, tracks upstream cleanly)🇩🇪 rasstamann/ai-job-search-germany by @rasstamann - Germany/DACH adaptation (xing-search portal skill; StepStone and Indeed investigated and documented as blocked, kept on WebSearch fallback)
🇳🇱 ai-job-search-nl by @Fabian-Schreuder - Netherlands adaptation (nationalevacaturebank-search over the board's public JSON API; Danish demo portals kept as references but disabled via the
enabled:toggle; OpenCode pointer)🇨🇴 FARLEY-PIEDRAHITA-OROZCO/ai-job-search by @FARLEY-PIEDRAHITA-OROZCO - Colombia adaptation (computrabajo-search and elempleo-search portal skills for the two largest Colombian boards; Spanish triggers, Colombian city queries; OpenCode harness port)
🇳🇴 ai-job-search-no by @ThomasHjartarson - Norway adaptation (nav-search over arbeidsplassen.nav.no plus nav-feed-sweep over NAV's official public job-vacancy feed API; locale retargeted to Norway, Danish demo portals dropped. No finn.no skill by design, finn's robots.txt disallows crawling, so nav-search leans on arbeidsplassen.nav.no which republishes much of finn's advert text and links back)
🌍 9mtm/ai-job-search by @9mtm - Flowxtra platform portal (
flowxtra-searchover the ATS's public cross-tenant API - live postings from every company hosted on the platform, multi-country, Europe-leaning; author-disclosed first-party: the author works at Flowxtra, so automated read access is operator-sanctioned)🇦🇺 ai-job-search-au-starter by @frJEN - Australia adaptation (adzuna-search over Adzuna's official API, free BYO key via env vars; read-only Gmail alert-email ingestion for boards that block scraping - Seek, Indeed, Jora - plus LinkedIn alert emails; docs in English and Mandarin with a beginner walkthrough. Heads-up: ships an optional
/platform-syncthat uses Claude in Chrome to edit profiles and set up job alerts on logged-in accounts - per-action confirmation and no submissions, but it is browser automation upstream deliberately doesn't ship (feat(submit): add /submit to file drafted applications through their … #214), and the fork itself notes the platforms' ToS ambiguity; a draft/outlook-syncrelies on an unvetted third-party MCP server)🌍 rakshith48/ai-job-search by @rakshith48 - Firecrawl fallback integration (
firecrawl-searchfor markets and boards without a dedicated portal skill, or when a portal parser is temporarily degraded; ships disabled -enabled: false- and/scrapemechanically skips it until a key is configured. Heads-up: Firecrawl is a metered hosted API billed per call, user-supplied key, env-only - the one paid-service exception in this index, which is why it lives in a fork (feat(firecrawl-search): add metered web job search #250); self-hosting Firecrawl is supported and needs no key)🇨🇦 ai-job-search-ca by @p3ji - Canada bilingual EN/FR adaptation (jobbank-ca-search over Job Bank / Guichet-Emplois and gcjobs-search over GC Jobs federal public-service postings, both zero-dependency and rate-limit-respecting; Canadian resume-conventions doc covering protected grounds, work-authorization phrasing, NOC codes and federal screening questions; Danish demo portals disabled. Sits on current upstream master; tracks via upstream merges)
⚙️ waleedtarbosh/ai-job-search by @waleedtarbosh - Universal Edition for Claude, OpenCode, Gemini/Antigravity, and Codex (centralized
AGENTS.mdcanon with thin per-harness pointer files and quickstart walkthroughs; verified placeholder-clean after a full history rewrite. Heads-up: the top layer is restructured -AGENTS.mdholds the canon andCLAUDE.mdbecomes a pointer, the inverse of upstream's Setup for opencode #45 shape - so it tracks upstream commands/skills, but core-file merges need care)🇨🇭 kblackma/ai-job-search by @kblackma - Switzerland/Romandie adaptation, French/English (job-room-search over the federal SECO job-room.ch board - Swiss Stellenmeldepflicht postings surface there before public boards - plus company-pages-search, an ATS-direct employer-watchlist skill over Greenhouse/Lever/SmartRecruiters/Oracle HCM, the named first candidate for core promotion per the fix(web-research): stop treating a WebFetch 403 as a dead posting #277 thread; jobup-search and englishjobsearch-search ship disabled by the author's own call - jobup's API path is robots-disallowed, englishjobsearch's categories proved noise; fetch layer identifies honestly and is gated by upstream's robots checker; rebased to current upstream master)
🇬🇧 ai-job-search-uk by @anjolok1997 - UK adaptation (reed-search, totaljobs-search, and wttj-search portal skills, UK city config, a cross-portal --graduate filter, and an opt-in Sponsorship Gate that checks employers against the gov.uk visa-sponsor register; Danish demo portals removed; tracks upstream via cherry-picks. Honest identifying User-Agent tokens and robots-compliant totaljobs pagination verified 2026-08-07)
🇺🇸 job-search-project by @ayushsawnani - US new-grad independent adaptation (inspired-by, shares no code) of the drafter-reviewer pattern for entry-level DS/ML/SWE, NY & Chicago (browser-verified posting liveness and real application questions via an ATS gate, a 4-agent review council, draft-only LinkedIn referral sourcing, an autonomous daily loop with CSV memory, and longitudinal skill-gap aggregation. Note: uses Claude in Chrome browser verification and scheduled unattended browsing - further than this template's own no-browser-automation posture - while keeping the last touch (submitting, sending) strictly human; never auto-submits or auto-connects. Screened 2026-08-06)
⚙️ CoNFiDeNTHaKaN/ai-job-search by @CoNFiDeNTHaKaN - MiMo Code + Qwen3-Coder-Next runtime port, a fully local/self-hosted alternative to Claude Code (the 12 commands mirrored to
.mimocode/as thin ports with tool names remapped, skills read natively from.claude/skills/with nothing duplicated, reviewer/rank-scorer subagents run write/edit-disabled, runtime config points only at a local llama.cpp endpoint; on themimocode-integrationbranch, fully current with upstream master. Screened 2026-08-10)🇮🇳 ai-job-search-india by @ashutoshgh - India market adaptation (10 India portal skills - Apna, Hirist, IIMJobs, Instahyre, Indeed India, Randstad India, Cutshort, AIJobs, Foundit, ProtocolJobs - plus 5 remote-board skills - Himalayas, RemoteOK, WeWorkRemotely, Y Combinator, Otta; honest identifying User-Agent tokens across all 15 CLIs. Heads-up: indeed-india-search's detail command fetches a path Indeed's robots.txt disallows - documented honestly in the skill itself, author follow-up pending. History verified placeholder-clean across all branches 2026-08-12)
🇨🇴 oscarbol09/ai-job-search by @oscarbol09 - Colombia / LATAM adaptation, second independent Computrabajo implementation (computrabajo-search over co.computrabajo.com; zero runtime dependencies, 40-case test suite, robots.txt-honest page-1-only pagination with an explicit personal-use note; on the
feature/computrabajo-searchbranch, rebased on current upstream master. Verified 2026-08-27)⚙️ ai-job-search-codex-port by @Tomchen070814 - unofficial Codex plugin port (installable Codex plugin manifest + marketplace layout, a Codex-native orchestration skill over the upstream canonical workflows, and a private workspace initializer that keeps candidate data outside the installed plugin; linkedin-search and freehire-search enabled by default, the four Danish portals ship opt-in with upstream's
enabled: false; source-only portal launchers - no prebuilt bundles or binaries anywhere in the tree, verified 2026-08-27 after the author removed the original dist/ payloads on request. MIT + NOTICE attribution intact)🌍 into-design-systems-mcp-adapter by @SimonDiff - Into Design Systems portal adapter (
into-design-systems-searchover the board's own public read-only MCP server via JSON-RPC - the index's first specialization-scoped portal: global Design System roles rather than a market. Standalone MIT repo, not a fork - one skill folder plus docs, zero runtime dependencies, offline stubbed tests; declared deviations: no--page(date-window walking via--jobage/--beforeinstead) andid: nulllistings kept and counted viameta.withoutDetail. Verified 2026-09-02 ate0e0382)🇮🇳 ai-job-search-india-bluecollar by @smartbriefai - India adaptation, second independent India fork, blue-collar focus (
shine-searchover Shine.com for mainstream corporate roles,workindia-searchover WorkIndia for blue- and grey-collar roles with tier-2/tier-3 strength; zero runtime dependencies, 27 test cases each, honest UA tokens, fixture contact details scrubbed before the first push. Declared deviations in workindia-search, both forced by a robots.txt that disallows every query-string URL: category+city path slugs instead of keyword search, and page 1 only. Fails loud when an unrecognized city slug 301s to the nationwide page - caught in the author's own pre-publication audit. Not a replacement for @ashutoshgh's India fork; the two cover different halves of the market. Verified 2026-09-03 atd71ae47)🇩🇰 job-agent by @mikkelkrogsholm - Danish job search as an MCP server rather than a skill folder: four standalone read-only servers (Jobnet, Akademikernes Jobbank, Jobindex, Jobdanmark) plus one combined server that searches all four concurrently, hosted for anyone at
https://job-agent.dk/mcpand self-hostable from the repo (bun >= 1.4). The same boards this template ships scrapers for, reachable from any MCP client with no code to maintain. Read-only by construction and by instruction: the server's own MCP instructions say "Never log in, change profiles, contact employers, or submit applications; the jobseeker approves and sends everything", every tool marks advertisement text as untrusted third-party content, honest identifying UAs, only the four boards as outbound hosts, no lifecycle scripts, anonymous fair-use rate limiting, 76 tests. Two things to know before you point a client at it: it is a hosted endpoint, so your search queries (not your profile) transit the author's server - self-host if that matters to you - and it is not a*-searchskill, so/scrapewill not pick it up; use it interactively through your MCP client, or wrap it with/add-portalif you want it in the scrape loop. Also ships an optional WebMCP surface, off by default. Verified 2026-09-03 at12f8133)🇨🇳 offer-harvester by @JiemsLBJ - China-market adaptation, standalone repo (fresh history, not a git fork; upstream-update triage is manual). Three campus and corporate portal skills following the contract (
shixiseng-searchover Shixiseng,tencent-searchover Tencent careers' public JSON,hotjob-searchover Wecruit/Hotjob employer sites) pluscompany-careers-search(Meituan, Feishu Jobs, Moka, Greenhouse, Lever, Ashby; adapted from career-ops, MIT attribution kept); zero runtime deps, honest UAs, path-scoped permissions per security: narrow Bash(bun run:*) to the six shipped portal CLIs #396. Ships form-filling automation, listed under criterion 5: a Playwright pipeline for campus-application forms with an unconditional confirmation before every submit (no flag, env var or batch mode skips it, verified in code), email applications that send one message only after an exact per-draft token, no captcha or bot-protection bypass anywhere, and a local-only dashboard. That is further than the template goes and further than most readers here will want; readautomation/before you run it. Profile files placeholder-clean with a CI identity guard on the CV templates; personal data found in the first review was removed by rebuilding the repo, verified gone. Portal smoke tests are live by design; the 26-case Python suite is offline. README in Chinese. Verified 2026-09-06 atcca0d4d)⚙️ methanoya/MadsLorentzen_ai-job-search by @methanoya - Docker Sandboxes environment harness: a
.sbx/directory (environment file plus a mixin kit installing poppler, pypdf, bun and TinyTeX with the moderncv package set, and PATH wiring for every shell) that runs Claude Code on this workspace inside an isolated microVM, plus a fork-side CI workflow that validates the kit. Tracks upstream master with the harness as the only change; profile files placeholder-clean. Read the kit before creating a sandbox: its network policy is allow-all and it provisions via piped shell installs, contained by the microVM. Screened 2026-09-10 at 4a10e72.🌍 alexmaiker007-bit/aipaylist-search by @alexmaiker007-bit - AIPayList portal skill as a standalone folder (paid AI training / RLHF / annotation contractor roles; first-party, referral-monetised board,
is_up_tokeeps "up to $X" honest; search and detail only; reviewed 2026-09-16 at8b5121d)Portal-skill index
Looking for a search skill for your job board? Find it below, open the fork, read the code, then copy the one folder into your own
.agents/skills/-/scrapepicks it up automatically. Discovery is centralized here; trust stays with you: review what you copy, exactly as you would any code that runs against your career data. (Idea credit: @rasstamann in #122.)Copy the skill folder and nothing else. In particular, never lift a fork's
.claude/settings.jsonortools/security_guards.py: upstream'sbun runallowlist is path-scoped per CLI on purpose (#396), and a fork may have widened it back toBash(bun run:*)for its own convenience. Borrowing a skill means approving its exact command once the first time you run it - expected, not a bug..agents/skills/jobindex-search.agents/skills/jobnet-search.agents/skills/jobbank-search.agents/skills/jobdanmark-search.agents/skills/linkedin-search.agents/skills/freehire-search.agents/skills/itviec-search.agents/skills/infojobs-search.agents/skills/tecnoempleo-search.agents/skills/getmanfred-search.agents/skills/gupy-search.agents/skills/vagas-com-search.agents/skills/justjoin-search(server-rendered listing pages; neverapi.justjoin.it, which robots.txt disallows; reviewed atdc79ebe).agents/skills/nofluffjobs-search(server-rendered listing pages; stays off/api/and/posting/per robots.txt; reviewed atdc79ebe).agents/skills/swelist-search(work branch).agents/skills/xing-search(personal use, low volume - see its ToS note).agents/skills/nationalevacaturebank-search(public JSON API; personal use, respect rate limits).agents/skills/computrabajo-search.agents/skills/elempleo-search.agents/skills/computrabajo-search(branchfeature/computrabajo-search; page-1 only per robots.txt - see its ToS note).agents/skills/elempleo-search(branchfeature/computrabajo-search; Schema.org JSON-LD extraction; reviewed atedcc9f9).agents/skills/torre-search(public REST API, no key; compensation in USD/COP; reviewed atedcc9f9).agents/skills/spe-search(public REST API of the national employment service; reviewed atedcc9f9).agents/skills/ticjob-search(IT and telecom vertical; robots-permitted listing pages, personal-use note; reviewed atde016b5).agents/skills/nav-search(public NAV board; fork also ships a companionnav-feed-sweepover NAV's feed API).agents/skills/flowxtra-search(public cross-tenant API; author-disclosed first-party skill).agents/skills/adzuna-search(official API; free personal key required, env-only).agents/skills/firecrawl-search(metered hosted API, user-supplied key, env-only; self-host supported; disabled by default viaenabled: false).agents/skills/jobbank-ca-search(bilingual EN/FR; honours the site's crawl-delay).agents/skills/gcjobs-search.agents/skills/job-room-search(federal board; Stellenmeldepflicht postings appear here before public boards).agents/skills/company-pages-search(Greenhouse/Lever/SmartRecruiters/Oracle HCM; user-curated watchlist in a gitignoredcompany_pages.json; honest UA, robots-gated escalation; core-promotion candidate per #277).agents/skills/reed-search(public pages, robots-compliant paths; no API key needed).agents/skills/totaljobs-search(robots-compliant path-only URLs; pagination beyond page 1 deliberately refused by the CLI).agents/skills/wttj-search(public JSON API, client-side GB filtering).agents/skills/<portal>-search, 10 skills (indeed-india-search'sdetailfetches a robots-disallowed path - see its own ToS note).agents/skills/<portal>-search, 5 skillsadapter/into-design-systems-search(board's own public read-only MCP server - no scraping, no key; reviewed ate0e0382).agents/skills/aipaylist-search(board's own public/jobs.json, no key; author-disclosed first-party, site monetised by referral links; pay is the advertised band top withis_up_toflagged; reviewed at8b5121d).agents/skills/shine-search(mainstream white-collar, HT Media group; robots-honest detail paths; reviewed atd71ae47).agents/skills/workindia-search(blue/grey-collar, tier-2/3 strength; category+city browsing and page-1 only per robots.txt, which disallows every query-string URL; reviewed atd71ae47)https://job-agent.dk/mcp- a hosted read-only MCP endpoint, not a skill folder: point your MCP client at it, or self-host from the repo; not picked up by/scrape; queries transit the author's server; reviewed at12f8133.agents/skills/shixiseng-search(internship board, Nuxt SSR parsing; live smoke tests; reviewed atcca0d4d).agents/skills/tencent-search(public JSON API, personal-use note; reviewed atcca0d4d).agents/skills/hotjob-search(defaults to Deloitte China, configurable; reviewed atcca0d4d).agents/skills/company-careers-search(adapted from career-ops, MIT; reviewed atcca0d4d)Fork authors: added a portal? Reply below and I'll add the row. China-fork portals welcome too - I only list paths I or the author can point to directly.
Forks that follow the portal-skill contract (
/add-portalgenerates compliant skills) stay compatible with upstream improvements -/scrapepicks up any installed*-searchskill automatically.All reactions