onboard
Skill1 starsChanged 4 months agoMITView on GitHub
Build the evidence bank from the user's raw career material (resumes, CVs, notes; pdf/md/txt/tex or pasted text). Ingests everything in bank/sources/ (or given paths / a paste), structures it into experience_bank.md + profile.json, interviews the user with a few high-value questions, and finishes with the evidence map. Cold-start only; it builds a bank where none exists.
172 other projects use a version of this.
Adopted by
Most recently taken up by
- TsukiNi22/skillstheir own version, first seen Nov 16, 2026
- TsukiNi22/ai-utilstheir own version, first seen Nov 16, 2026
- akovfoj-lang/plainframetheir own version, first seen Nov 16, 2026
- chuckplayer/claude-agent-packtheir own version, first seen Nov 16, 2026
- jestatsio/codebase-mentortheir own version, first seen Nov 15, 2026
- chuanglibs/Claude-Code-Game-Studios-Codextheir own version, first seen Nov 15, 2026
Used with
Projects with it also use soak-test, team-live-ops and 10 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.
lilangMaxSkillsoak-test
Generate a soak test protocol for extended play sessions. Defines what to observe, measure, and log during long play sessions to surface slow leaks, fatigue effects, and edge cases that only appear after sustained play. Primarily used in Polish and Release phases.
Also in 9 projects that use this
lilangMaxSkillteam-live-ops
Orchestrate the live-ops team for post-launch content planning: coordinates live-ops-designer, economy-designer, analytics-engineer, community-manager, writer, and narrative-director to design and plan a season, event, or live content update.
Also in 9 projects that use this
HK-hubSkillteach-impeccable
One-time setup that gathers design context for your project and saves it to your AI config file. Run once to establish persistent design guidelines.
Also in 19 projects that use this
CoralGameSkillproject-stage-detect
Automatically analyze project state, detect stage, identify gaps, and recommend next steps based on existing artifacts.
Also in 15 projects that use this
CoralGameSkillteam-level
Orchestrate level design team: level-designer + narrative-director + world-builder + art-director + systems-designer + qa-tester for complete area/level creation.
Also in 11 projects that use this
- +5 more from lilangMax
- +2 more from CoralGame
Other versions172 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 172
- Donchitos/Claude-Code-Game-Studios★ 39ksame name
- parcadei/Continuous-Claude★ 3.9ksame name
- parcadei/Continuous-Claude-v2★ 3.9ksame name
- parcadei/Continuous-Claude-v3★ 3.9ksame name
- nateherkai/AIS-OS★ 1.4ksame name
- poshan0126/dotclaude★ 871same name
- talesofai/cohub★ 585same name
- me2resh/apexyard★ 542same name
- nexmoe/DayBee★ 424same name
- nexmoe/eve★ 424same name
162 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.
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.
0
Common
Most of the 160 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
If bank/experience_bank.md already exists, stop: this command builds a bank where none exists, and running it against a populated bank …
- Very specific: line 68
Read bank/.ingest/gaps.md.
- Distinctive: line 74
Ask via the AskUserQuestion tool, in batches of up to 4 questions per call, Tier-A first.
- Very specific: line 115
If the interview left profile.json without an application block (the sources may have volunteered one), make ONE offer: saving …
- Very specific: line 125
Run python3 ops/bank_lint.py bank.
“Common” is measured: kept by at least 6 in 10 of the 160 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
/onboard: build the bank from the user's raw material
Turn a pile of raw sources into the structured bank the tailoring pipeline runs on. Code ingests, an agent structures, and you host the one conversational part: a short interview. The user's answers and their sources are the only places facts come from.
Voice. Describe what is happening to the user's career material in their terms: "I've read your three files", "your two resumes disagree on when you left Acme", "saved, your experience bank is ready". Give a file path where the user acts on it (where their bank lives, where to drop files). One short progress line per stage.
0. Cold-start guard
If bank/experience_bank.md already exists, stop: this command builds a bank where none
exists, and running it against a populated bank would clobber curated work. Tell the user
what exists and ask how they want to proceed (they can move the current bank aside first
if a rebuild is really intended).
1. Resolve the sources
bank/sources/ is the drop box AND the permanent home: everything the user gives us lands
there, forever, unedited. From $ARGUMENTS (and, if empty, the conversation):
- Path arguments (files or a folder) → expand a folder to the supported files inside
it (pdf, md, txt, tex), then:
python3 ops/bank_ingest.py bank <file ...> - No arguments → ingest whatever is already in the drop box:
Exit 4 with nothing ingested → the box is empty: tell the user their options (drop files intopython3 ops/bank_ingest.py bankbank/sources/and rerun, give paths, or paste their resume text right here). - Pasted text (in the argument or earlier in this conversation) → Write it VERBATIM to
bank/sources/pasted-<YYYY-MM-DD>.md(one copy, change nothing), then run the no-arguments form.
Invite more than one source: old resume versions, a CV, notes; the union holds more true material than any single file, and this is the moment to collect it.
Then read bank/.ingest/manifest.json and act on per-source status: low_yield,
not_text, or extract_failed means that source is unusable as text; name the file,
say what happened, and offer the paste route for it. Proceed when at least one source is
usable.
2. Structure: spawn the extract agent
Spawn the bank-extract agent (Agent tool, subagent_type: "bank-extract") with a prompt
giving the concrete paths: the bank directory (bank/), the manifest
(bank/.ingest/manifest.json), and the format contract (spec/bank_format.md). It writes
bank/experience_bank.md, bank/profile.json, and bank/.ingest/gaps.md, and returns a
one-line summary. Relay that summary. Then log the operation:
## [YYYY-MM-DD] extract | <roles> roles, <units> units, <agenda size> gapsappended to bank/log.md.
3. The interview: interactive, every question skippable
Read bank/.ingest/gaps.md. Select what to ask: every Tier-A item (record blockers),
then Tier B, C, D in order until the total reaches 8; drop the rest. Lead with one
plain sentence of what you found and why a few questions are worth their time (e.g. "Your
resumes are in. A few things only you can answer would make the result stronger; skip
anything freely.").
Ask via the AskUserQuestion tool, in batches of up to 4 questions per call, Tier-A first. The lead-in message explains the mechanics once: answers are typed into the Other field; the listed choices are ways to skip, so each question's text is the question alone. How to build each question:
- A conflict between the user's own sources (two dates, two titles): the options ARE the conflicting values, each labeled with where it came from ("August 2019 (your 2022 resume)"). The user clicks the right one, or types the real answer under Other.
- A public-record fact (an institution's location, a credential's issuing body): research it (model knowledge or a quick web search) and offer the candidates as options to confirm, each labeled with what it refers to ("San Marcos, CA (Cal State San Marcos)"). Offer candidates only when confident; otherwise ask openly. The recorded fact's source is the user's confirmation, never the research itself, which is also why ambiguity (two campuses with the name) is presented as options, not resolved silently.
- Every personal question (a scale, an outcome, a decision, a date of their own employment): the answer arrives through the built-in Other free-text field. The two listed options are the two ways to skip, "Skip: no such number/detail exists" and "Skip for now: could dig it up later", because they mean different things for the bank (a dead end vs. a note to revisit). Personal facts never carry suggested values: a proposed number, outcome, or tool name in an option is putting words in the user's mouth.
Rules for the answers:
- The user's answer is testimony: fold it in verbatim, their numbers, their words. An answer that hedges ("maybe 30 or so services") is recorded with its hedge.
- A record-layer answer (a date, a title, a location) updates BOTH
bank/profile.jsonand the matching header line inbank/experience_bank.md; the two must round-trip. - A narrative answer (scale, outcome, decision reasoning) is a targeted Edit to the unit
the agenda item names: fill its thin field (
Scale,Results, a newDesign contributionsentry). Everything else in the unit stays byte-for-byte. - A skipped question stays a gap: leave the unit thin and the profile field null. Mark
each item in
gaps.mdasanswered,skipped(no such detail exists), ordeferred(worth revisiting) so the file records what remains and why. - One follow-up round is fine if an answer opens something clearly valuable; past that, stop; the bank grows later, this is not an interrogation.
Then log it: ## [YYYY-MM-DD] interview | <n> answered, <m> skipped appended to
bank/log.md.
4. Application defaults: one optional offer
If the interview left profile.json without an application block (the sources may have
volunteered one), make ONE offer: saving application-form defaults (work authorization /
sponsorship, salary expectation, notice period, relocation, remote/hybrid/onsite
preference) means application forms never have to re-ask. One AskUserQuestion batch at
most: closed items (sponsorship, work preference) as options, free-text items via Other,
recorded verbatim into the application block per spec/bank_format.md. Decline = move
on; ask nothing item-by-item.
5. Integrity check
Run python3 ops/bank_lint.py bank. An ERROR is yours to fix before moving on: after a
fold it is usually a header line that drifted from profile.json; correct the file and run
the check again until no errors remain. Warnings are known gaps, already the user's choice
(a skipped date, a unit with no stack): carry them into the closing summary in plain terms
rather than fixing anything.
6. Index and close
Invoke the bank-map skill to build bank/evidence_map.md from the new bank. Then close
with: where everything lives (bank/experience_bank.md, bank/profile.json, the sources
in bank/sources/), what stayed thin (the skipped items in bank/.ingest/gaps.md, as
things worth adding when the user has the numbers), and that /tailor <posting> is now
ready to run.
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.

