publish
Skill2 starsChanged 3 months agoMITView on GitHub
Re-publish the public job-hunt-kit snapshot from this repo — dry-run the extraction, review what changed, cold-clone check, then push. Use when asked to publish, re-extract, update the public repo, or ship changes to job-hunt-kit.
- Installs packages
- Commits and pushes
67 other projects use a version of this.
Adopted by
Most recently taken up by
- filecoin-project/filecoin-skillstheir own version, first seen Nov 16, 2026
- noelceta/content-enginetheir own version, first seen Nov 16, 2026
- vincentventalon/claude-code-video-editing-skilltheir own version, first seen Nov 15, 2026
- vincentventalon/my-claude-skillstheir own version, first seen Nov 15, 2026
- foster-systems/shelftheir own version, first seen Nov 15, 2026
- carlos2219/web-kittheir own version, first seen Nov 15, 2026
Used with
Projects with it also use clips, outliers and 9 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
MelvynxSkillclips
Turns one long video (YouTube link, podcast, webinar, livestream, or a local file) into finished vertical shorts. Use when the user says long to short, repurpose, long video to shorts, clips from my video, podcast clips, cut my YouTube video into shorts, or gives a long video link. Finds clips that stand alone (one idea, 20 to 60 seconds), scores them, cuts them and edits each one with short-editing (split-screen paper look, captions, cuts, hook, boards, 1080x1920), and writes new scripts in the studio-script format from the ideas too spread out to clip. Reads YouTube captions and chapters with yt-dlp, or transcribes a local file with the short-editing engine (whisper.cpp, free, offline).
Also in 3 projects that use this
MelvynxSkilloutliers
Finds the outlier reels of Instagram accounts or videos of TikTok accounts (3x their own account's median views or more) through Treg and turns the 5 best into remake scripts in the creator's voice. Two modes - /outliers @handle digs one account; /outliers with no account starts from the creator's own reels, uses and grows a competitor watchlist (video-agent/competitors.json, 70 % main language / 30 % second language when they watch two), ranks every account's outliers against its own median and writes 5 scripts into video-agent/scripts/. Use when the user says outliers, find outliers, what blew up for my competitors, what works in my niche, find me videos to remake, ideas from @someone, what blew up on TikTok.
Also in 3 projects that use this
khadinakbarlabsSkillkeywords
Research and prioritize SEO keywords, search intent, topic clusters, competitor keyword gaps and keyword-to-page mappings. Use when asked what to rank for, keyword opportunities, long-tail keywords, a content calendar, or a keyword plan. Uses owned Apify keyword/SERP Actors with approved budgets, or supplied exports offline. Does not invent volume or organic difficulty.
Also in 3 projects that use this
MicrosoftSkillcreate
Also in 4 projects that use this
justinseger015-ctrlSkilledit
Edit a video using the video-editing orchestrator. Detects format, routes to correct editing skill, and executes the full process. Use when user says: /edit, edit video, edit clip, edit this, make it polished, finish editing, polish this video.
Also in 3 projects that use this
LangChainSkillrender
Turn an SVG design into a PNG and look at it. Use after you write or change a design, but only if execute is one of your tools.
Also in 3 projects that use this
Other versions67 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 67
- code-yeongyu/oh-my-openagent★ 71ksame name
- code-yeongyu/oh-my-opencode★ 71ksame name
- garrytan/gbrain★ 33ksame name
- TencentCloud/Octop★ 12ksame name
- harbor-framework/harbor★ 7.1ksame name
- laude-institute/harbor★ 7.1ksame name
- Q00/ouroboros★ 6.4ksame name
- paean-ai/deeptide★ 1.1ksame name
- ZimoLiao/scholaraio★ 589same name
- ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool★ 432same name
57 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- apply-formSkill
- cover-letterSkill
- evaluate-roleSkill
- job-triageSkill
- research-companySkill
- setupSkill
- sync-appliedSkill
- tailor-cv-batchSkill
- tailor-cvSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 56 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 51
Must read passed and skipped, never failed — a stranger's fresh clone showing red lines is the thing the public suite exists to prevent.
- Very specific: line 73
And the standing rule this step now carries: if you find another value duplicated between the two files, do not fix the copy — move it into …
- Distinctive: line 83
Read the classification summary in full.
- Distinctive: line 157
Do not ask Ben a question step 0's four files answer: Read them, then state the answer flatly.
- Distinctive: line 180
step 0 exists because of this: Read the four files, then speak.
“Common” is measured: kept by at least 6 in 10 of the 56 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Internal to jobs-db. This skill ships in the public snapshot with the rest of .claude/skills/,
but it is not part of the product and is deliberately absent from the public README's workflow table:
it drives scripts/, which is excluded whole, so every command in it names a file a stranger does not
have. Leave it out of that table when the table changes.
Re-publish job-hunt-kit — the public, one-way snapshot of this repo. scripts/EXTRACTION.md is the
first-time seeding procedure (create repo → private → flip to public) and is done. This skill is the
repeat path, which is the one that happens often enough to be worth a checklist.
Read this before running anything
The public repo is a snapshot, not a mirror. Re-running the extraction overwrites the whole public tree. It is not a merge, not an update, and there is no cadence — you publish when a stale snapshot starts bothering Ben, never because something told you to. Consequences that follow, and that you must state to Ben if he seems to expect otherwise:
- An upstream PR merged into
job-hunt-kitis destroyed by the next publish. That is why the repo has noCONTRIBUTING.md(andcore/test_licensing.pyfails if one appears). config/settings.yamlis substituted: the public tree shipsconfig/example/settings.yamlin its place. So an edit to the real settings file does not reach the public repo, and an edit that should be public has to be made in the example too. This is the single most common way the public tree goes quietly stale — check it every time (see step 2).profile/,matches/,applications/,data/never ship. They are the privacy seam.
0 · Required reading — four files, before you form an opinion about anything
Read these before step 1, every time, and do not ask Ben anything they answer. They are here because on 2026-08-06 an agent evaluating a publish told him the open-source product looked incomplete (it isn't) and asked him to decide how far to scrub his name (already decided, in writing). Both answers were in files that had not been opened. He is not in the weeds of this repo; the repo is, and a question it has already answered costs him time and reads as homework not done.
| read | it settles |
scripts/publish-denylist.txt | What "personal" means here — surname, email addresses, sheet and label ids, home-directory paths, secret values. Read them there; do not restate a denied value in this file, which ships — the guard caught exactly that while this step was being written. The first name is deliberately not on the list: the docs and skills are written as instructions about the person who runs the tool, and every publish so far has shipped it that way. Adding to the list is a real change; do not treat the current line as an oversight. |
docs/agents/tests.md | Why a stranger's clone skips tests, and what clean looks like (passed and skipped, never failed). A skip is the suite working. Also the rule a new skip must satisfy: say what is missing and how to get it. |
config/example/README.md | What a new user actually receives — a complete fictional-seeker configuration (profile, rubric, bullet bank, CV, settings) that seeds itself with python -m core.example. The public product is not missing a profile; profile/ is the owner's copy, and the example is the shipped one. |
| the last entry of §8 · Publish log at the bottom of this file | what went wrong the previous time |
1 · Pre-flight — the suite and the leak test
.venv/bin/python -m pytest -qMust read passed and skipped, never failed — a stranger's fresh clone showing red lines is the
thing the public suite exists to prevent. scripts/test_leaks.py runs as part of this and is the one that
matters most: it asserts no tracked path outside the four personal directories carries Ben's identifiers.
If it goes red, fix the file it names — never the test.
2 · The settings substitution — now a test, not a judgement call
The public tree ships config/example/settings.yaml as its config/settings.yaml, so anything stated
in the example becomes every new user's default. This step used to read "diff the two files and decide
which differences are facts about the tool rather than Ben's preferences." That is judgement, made from
memory, once per publish — and it failed exactly as you would expect: the model ids were duplicated in
both files, Ben moved his to Sonnet on cost on 2026-07-29, the example kept an older Opus, and for two
weeks every new user's default was a model nobody had chosen.
The fix was structural. Model ids now live in core.settings.DEFAULT_MODELS and any settings file
that stays silent inherits them; core/test_settings.py asserts every role has a default and
core/test_example.py asserts the example names none. Step 1 covers this.
What is left for a human, and it is small: skim the example's comments for a claim about the code that has gone stale — a count, a source list, a runtime. A comment saying "six staffing firms" when there are seven ships a lie to every reader. Never quote a measurement you have not just taken.
And the standing rule this step now carries: if you find another value duplicated between the two
files, do not fix the copy — move it into the code and delete the copy, the way models: was. A second
source of truth in a file nobody diffs is the failure this whole step exists to catch.
3 · Dry-run and read the summary
.venv/bin/python scripts/extract.py ../job-hunt-kit --dry-runRead the classification summary in full. An unclassified top-level path aborts the run — that is the
allowlist working, and the fix is to classify the path deliberately, never to loosen the guard. New skills
ship automatically (a skill is vendored iff its directory name is a key in skills-lock.json), so a new
product skill needs no allowlist edit but should be sanity-checked as "ours" in the summary.
4 · Extract for real, then review the diff as a stranger
.venv/bin/python scripts/extract.py ../job-hunt-kit
cd ../job-hunt-kit && git diff HEAD~1 --stat && git log --oneline -3Then actually read the diff of anything under docs/, README.md or NOTICE. You are the last reader
before the internet. Two specific things to confirm:
NOTICEstill names every attributed market source.core/test_licensing.pyguards this, but the reader-facing half is prose and can drift.- No absolute
/Users/...path, inbox address, sheet id or personal filename appears anywhere in the diff. The leak test covers tracked paths; your eyes cover phrasing.
5 · The cold-clone check — the gate, and it runs BEFORE the push
This step used to be step 6, after the push, and that was the process bug in this file. The only
test of what a stranger actually receives ran after the irreversible act — a receipt rather than a
gate. On 2026-08-06 the repo's own suite was green while a fresh clone read 2 failed; it was found
only because the agent improvised a throwaway clone off-script. Run as written, that publish would
have shipped a broken first impression and then discovered it.
The public tree is the one thing this repo's own suite never runs against — it ships a substituted
settings file, and it has no profile/ at all. Both are why this cannot be skipped.
Extract into a throwaway clone of the public repo and run its suite there:
S=$(mktemp -d)
git clone -q ../job-hunt-kit "$S/cold" && .venv/bin/python scripts/extract.py "$S/cold"
cd "$S/cold" && python3 -m venv .venv && .venv/bin/pip install -q -r core/requirements.txt
.venv/bin/python -m pytest -qMust read passed and skipped, never failed. owner_only and needs_profile tests should
skip — that is them working. If anything fails: fix it here in jobs-db and re-extract. Never
patch the public clone; the next extraction overwrites it.
Two more things to confirm while you have a stranger's tree in front of you:
python -m core.examplethenpytest -qagain — a seeded clone must also be clean.- The shipped default run is sane for someone with no configuration. Do not quote a wall-clock number from this file; take the measurement or say nothing.
6 · Push
cd ../job-hunt-kit && git pushAsk Ben before pushing — this is the irreversible, outward-facing step, and approval on one publish does not carry to the next. Do not reach this step with a failing cold clone behind you.
7 · Report
Tell Ben: what shipped (a one-line summary of the diff, not a file list), that the cold clone passed before the push, and anything you found in step 2 that had gone stale. If you fixed a drift between the two settings files, say so — and say whether you moved the value into the code or merely synced it. Syncing it means it will drift again.
Rules
- Never push without asking. Steps 1–5 are safe and reversible; step 6 is not.
- Never loosen the allowlist to make a run pass. An unclassified path is a decision, not an obstacle.
- Never edit
scripts/EXTRACTION.mdto match a shortcut you took. If the procedure changed, change it deliberately and say why. - The public repo is written for a reader who is not Ben.
README.mdis the one tracked file whose audience is a stranger — read it as one. - Do not ask Ben a question step 0's four files answer. Read them, then state the answer flatly.
- A duplicated value is a bug, not a chore. Move it into the code; do not sync the copy.
8 · Publish log — append one entry per publish, before you report
Two or three lines: what went wrong, and what changed so it cannot recur. Step 0 requires reading the last entry, so this is the only part of the file that gets better on its own. Newest first.
2026-08-06
- Two tests failed on a fresh clone while the repo's own suite was green, and the check that would
have caught it ran after the push. One test (added 2026-07-30) called the scorer for real and needed
the owner's rubric, with no
needs_profilemarker. Another (added 2026-07-31) redirectedconfig.CORPUS_DIRbut not the writer, which was bound at import — so it wrote fixture ids into the owner's livedata/corpus/seen.jsonand created adata/directory that broke a third test on a clone. Fixed:triage/store.pyresolves the corpus path per call, soCORPUS_DIRis the single knob; the marker was added. The cold-clone check moved from step 6 to step 5, before the push. - The model ids were duplicated between
config/settings.yamland the example, and had drifted since 2026-07-29. Fixed structurally:core.settings.DEFAULT_MODELSowns them, the example names none, two tests enforce it, and the manual diff step is gone. - The agent asked Ben two questions the repo had already answered — whether the open-source product
was incomplete (it is not;
config/example/is a complete seeded configuration) and how far to scrub his name (scripts/publish-denylist.txtdefines it and deliberately excludes the first name). Fixed: step 0 exists because of this. Read the four files, then speak.
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.

