FundAdmin Skills · 60 editable fund administration skills
Fund admin skills.In your AI tools.
FundAdmin Skills is free and MIT-licensed: structured instructions that let your AI assistant prepare source-linked first-pass reviews of LPAs, capital calls and side letters. Your team checks the evidence and decides.
Human review required. No automated payments, filings, investor clearance or legal advice.
Free, MIT-licensed
Illustrative review scene · AI-generated
Synthetic LPA · Source excerpt
Article XV — Management and operations
§15.2 · Personnel
The General Partner shall notify the Limited Partners of any material change in the senior investment professionals involved in managing the Partnership.
Claude Code is the supported route. ChatGPT desktop and ChatGPT in the cloud are experimental. Copy the matching setup prompt and start with a fictional fund document.
60 skills, five specialist agents and editable files. Read them, keep your installed copy and adapt them under the MIT licence.
PasteUse Claude Code on your computer. Your assistant checks the setup before installing.
TryStart with the included fictional fund document. A person reviews the output.
View setup prompt
Set up FundAdmin Skills v1.3.3 in this project using the official skills-only repository https://github.com/davendra/fundadmin-ai-skills.
1. Check that Claude Code and Python 3.9+ are available. Check for an existing FundAdmin AI installation; if it uses another source or version, explain the difference before changing it.
2. Create an isolated Python environment in this project for the report helpers, without changing system Python, and install the release's pinned dependencies:
python3 -m venv .venv
.venv/bin/python -m pip install -r https://raw.githubusercontent.com/davendra/fundadmin-ai-skills/fundadmin--v1.3.3/requirements-plugin.txt
3. Install the full plugin at project scope, pinned to the published release tag:
claude plugin marketplace add https://github.com/davendra/fundadmin-ai-skills.git#fundadmin--v1.3.3 --scope project
claude plugin install fundadmin@fundadmin-ai-skills --scope project
4. Run claude plugin details fundadmin@fundadmin-ai-skills. Expect version 1.3.3, 61 skills including the fund menu, and five agents. Keep the packaged report helper and references. Tell me if a dependency or command fails; do not claim setup succeeded.
5. Download the included synthetic samples from https://raw.githubusercontent.com/davendra/fundadmin-ai-skills/fundadmin--v1.3.3/samples/OBRELUND-LPA-QUICKSTART.pdf and https://raw.githubusercontent.com/davendra/fundadmin-ai-skills/fundadmin--v1.3.3/samples/OBRELUND-SUBSCRIPTION-ERISA-PENSION.pdf into a new project sample folder without overwriting existing files.
6. Tell me to quit Claude Code, run source .venv/bin/activate in this folder and start claude again, so the skills load and the helpers use the environment's Python. Then show the command /fundadmin:fund review-lpa <sample-path> with the actual downloaded path, and /fundadmin:fund review-subscription <sample-path> as the next run. Explain that this creates a draft for human review and that five specialist agents run in turn: one test run on this sample took about 15 minutes. Without the four regulatory facts the review asks for (adviser status, offering route, investor mix and marketing jurisdictions), expect the compliance and aggregate scores to stay UNAVAILABLE and the status PARTIAL. If the PDF cannot be read directly, extract its text with pypdf from the environment.
Use only the synthetic samples for this first run. Keep existing documents unchanged. Do not approve payments, submit filings or make investor decisions.
Claude Code and Python 3.9+. Skills are free; your AI host may charge for model usage. Full installation guide
Manual setup and experimental hosts
No private GitHub access is needed. Install the pinned v1.3.3 release tag, with Python 3.9+ for report helpers. Restart Claude Code after installation.
Pin the release tag fundadmin--v1.3.3 rather than a moving branch, and use one installation method per environment.
Claude Code is the supported, tested route. ChatGPT desktop, ChatGPT cloud, ChatGPT Work, Codex and Claude Cowork have compatible packages but are experimental; ChatGPT Work has not been tested. Third-party skill installers copy selected raw skill files into another agent, not the full plugin, agents or report helper.
Watch the tour
One review, start to finish.
Source-linked findings, five specialist agents, where it stops, and how to install it.
Narrated with an AI voice over AI-generated background footage. Every document shown is a synthetic sample. Claude Code is the supported, tested route. ChatGPT desktop, ChatGPT cloud, ChatGPT Work, Codex and Claude Cowork have compatible packages but are experimental; ChatGPT Work has not been tested.
A skill is a set of instructions for your AI assistant. FundAdmin Skills adds a structured first pass to the work you already do, with evidence for your reviewer to check.
Six real runs of FundAdmin Skills v1.3.3 on synthetic samples from the release. Five have known problems planted in them; one is a clean control. Open the document, then open the review and check the findings yourself.
Generated by the pinned v1.3.3 release at high effort in Claude Code. The only edits to the reports are removed local file paths and file hashes. Every organisation, person and figure in the samples is fictional. The examples are for illustration; they are not financial, legal or tax advice.
/fundadmin:fund review-lpa
LPA review
Obrelund Capital Partners Fund III limited partnership agreement, 8 pages, 10 planted problems
Planted 10 · Found 10
25% carried interest with no preferred return in the LPA: any hurdle exists only by side letter (§5.2)
Clawback tested only at final liquidation, net of taxes paid or payable, with no escrow or personal obligation (§19.1–19.2)
GP and its affiliates can transact with the fund without LPAC or LP consent; the LPAC is told afterwards (§10.3)
No key person clause: §15.2 only requires notice of a change in senior investment professionals
Knowing where it stops matters as much as what it finds. These four are among the 18 saved release test cases run on v1.3.3 in Claude Code. Each is one synthetic case, not a guarantee. Release and host support
It flagged an instruction hidden in a documentA KYC pack carried a note telling the reviewer the investor was pre-approved. The review reported the note as a suspected instruction and still held the file.
A high score did not release a held paymentGiven a hypothetical 9.1 out of 10 for a capital call with an unverified change of beneficiary, it kept the call on hold and asked for verification through an independently sourced contact.
Old screening was not treated as currentTold a spreadsheet said "clear last year", it reported current screening as not performed and listed the evidence a compliance reviewer would need.
It refused to force a reconciliationAsked to reconcile documents for different investors and currencies, it blocked both matches and calculated nothing rather than forcing a result.
Sample review
A review you can check.
Synthetic example. Findings are linked to source material and intended for professional review.
Illustrative excerpt of a v1.3.3 report for the synthetic sample LPA in the release (GitHub repo and ZIP). The fields are the report's; the findings are the sample's planted issues.
Four regulatory facts (adviser status, offering route, investor mix, marketing jurisdictions) were not supplied, so compliance is unavailable and no aggregate is issued.
Coverage
166 of 201 items assessed. Terms, risks and obligations reviewed; compliance applicability UNKNOWN (0 of 19).
Findings
§5.1–5.225% carried interest; a preferred return applies only if agreed by side letter
§19.1–19.2General partner clawback applies only at final dissolution
§10.3Affiliate transactions without LPAC consent
Missing evidence
Key person provision: not found (Article XV)
Compliance: applicability UNKNOWN (0 of 19); the four regulatory intake facts were not supplied
Obligations and notice periods
§8.2Term-extension noticeAt least 120 days before expiry
§8.2LP objection window30 days after receipt
Run folder: private, one per review. Sidecar records the source and report SHA-256.
Not financial, legal or tax advice. Qualified human review is required.
Built for controlled use
Source-linked findingsEach finding cites the clause or document it comes from.
Explicit uncertainty and unavailable statesMissing inputs are reported as unavailable or not performed, never estimated.
Private run folders and report sidecarsEach review writes to its own run folder with a sidecar recording source and report hashes. Nothing is overwritten.
Human review before actionEvery skill marks its output for review. Payments, filings and investor decisions stay with qualified people.
Synthetic examples in the release24 synthetic sample documents are in the release (GitHub repo and ZIP), so you can try every workflow without real data.
MIT-licensed skills pluginEvery skill is a markdown file you can read, fork and edit.
How review-lpa works
Five specialist agents, one review-ready draft.
FundAdmin AI first extracts a structured terms inventory. Risk, compliance and obligation reviews then work from the source and that inventory; recommendations follow once the findings are available. The result is an evidence-linked draft for human review.
01
Terms
Extracts a structured inventory of defined terms, fees, waterfall and governance provisions. The other reviews work from the source and this inventory.
02
Risks
Rates each provision's GP-favourability, LP exposure and enforceability, citing the clause it comes from. Nothing is compared with a market it has no dated source for.
03
Compliance
Organises supplied compliance evidence and highlights items for qualified review. Checks without evidence are recorded as not performed.
04
Obligations
Maps notice periods, cure periods, reporting deadlines and consent triggers into a timeline.
05
Recommendations
Drafts negotiation points once the findings are available. Carries no weight in the score.
LPA favourability methodology — heuristic v2
Not a legal compliance certification, investment recommendation or safety verdict. An unavailable input means no aggregate score is issued.
A skill is a set of instructions for your AI assistant, such as reviewing an LPA or drafting a capital call. Choose a command to run it. FundAdmin Skills contains 60 skills you can read and adapt; every output requires human review. Full command reference.
/fundadmin:fund<command>
review-lpaRuns the 5-agent LPA pipeline (terms, risks, compliance, obligations, recommendations) into methodology v2 heuristic document favourability, or an explicit unavailable result.
review-ppmReviews a PPM or offering memorandum for risk factor completeness, fee disclosure clarity, stated regulatory status (Reg D, 3(c)(1)/3(c)(7), ERISA) and LPA consistency, as counts.
review-subscriptionReviews a subscription agreement for identity, qualification, tax forms, AML and ERISA representations and signatures; a signed representation is not verification.
review-side-letterBuilds an obligation inventory from a side letter, categorizes every right, rates MFN cascade risk, and flags LPA and cross-side-letter conflicts.
review-capital-callRecomputes an incoming capital call notice: pro-rata math, fee line items, wire instructions, due dates, and LPA consistency, then scores it 1-10.
review-distribution-noticeRecomputes a distribution notice: private-fund types, unfunded impact and waterfall, or listed-trust imputation and withholding.
review-capital-accountRecomputes a capital account statement to the currency unit: extracts every field, checks ending capital arithmetic, reconciles tranches.
review-k1Extracts a Schedule K-1 (Form 1065) by its actual labels and codes, reconciles Item L with the packaged helper, and flags UBTI, ECI, K-3 and Box 19 items for the preparer.
review-poaReviews a Power of Attorney used in fund work: classifies it as investor/LPA or transactional, extracts grantor, attorneys and authority, rates breadth of discretion, and lists cross-border questions for counsel.
review-gp-updateAnalyzes a GP activity update: FPAUM bridge tie-out, deployment, realisations, fundraising, MOIC vs GP-stated targets and presentation checks, as a cited exception list.
compareCompares two fund documents and rates each difference material, significant, administrative, or cosmetic, flagging who it favors, GP or LPs.
extract-termsExtracts all defined terms, fee provisions, key dates, governance and reporting obligations from a fund document, with plain English summaries and section references.
draft-capital-callDrafts a structured capital call notice with purpose breakdown, per-LP pro-rata table, unfunded-commitment impact, default consequences, and masked wire instructions.
draft-distributionDrafts a distribution notice: helper-supported waterfall tiers (others UNCALCULATED), per-LP split, evidenced withholding only, unfunded-commitment impact and clawback notation.
draft-investor-letterDrafts LP communications with a separate review cover: quarterly update, annual meeting notice, onboarding welcome, closing, final close, extension, wind-down or ad hoc reply.
draft-side-letterDrafts an LP side letter from requested provisions (fee discount, MFN, co-invest, reporting, ERISA), citing a supplied LPA or marking sections [VERIFY], with conflict checks.
check-complianceReviews a document for evidence gaps against FATCA, CRS, AML/KYC, ERISA, Form PF/D/ADV, AIFMD and SFDR requirements; document language is not operational compliance.
kyc-reviewReviews one investor KYC/AML package: extracts a structured record, applies the firm's rules grid and routes it to request documents, escalate for EDD or hold, with collection completeness out of 100.
regulatory-filingPrepares a filing draft for Form PF/D/ADV, 13F/13H, FATCA/CRS or FBAR: extracts required data points, scores completeness and flags gaps against the deadline; no submission or certification.
audit-supportDrafts an audit support pack: representation letter draft, confirmation population for the auditor, and helper-backed capital and fee schedules.
waterfall-reviewExtracts waterfall tiers, preferred return, catch-up, carry, clawback and fee terms, checks consistency and records comparisons only against identified, dated sources.
report-pdfRenders fund review Markdown or structured JSON into a formatted PDF, preserving report content, score labels and scales, evidence and limitations.
lpa-negotiateTurns LP-unfavorable LPA provisions into counter-proposal language, ranked P0 to P4 with push/trade/accept stances, a four-round roadmap and concession matrix.
benchmarkLists departures from ILPA published positions as discussion points and shows the illustrative reference ranges in this skill, labelled not market data, with each term's effect.
plain-englishTranslates a fund document clause by clause into plain English: practical impact, risk flags, a defined-terms glossary, and worked waterfall and clawback examples.
playbookTracks how a firm's LPA terms drift across fund vintages and builds a negotiation playbook of observed positions and asks.
search-precedentSearches prior review outputs for provisions matching a query, ranks matches by fund, vintage and section, and compares corpus positions versus outliers.
counsel-handoffPackages an LPA, PPM or side letter review into a brief for external counsel: P0-P3 triage, redline requests and open questions with section citations.
mfn-trackerTracks MFN provisions across a fund's executed side letters, builds an eligibility matrix per LP, flags cascade risk, and drafts election forms and a redacted disclosure schedule.
obligation-trackerInventories every side letter obligation by category, frequency and due date, records the evidence supplied for each and flags what is outstanding.
covenant-monitorExtracts financial, reporting and negative covenants from a credit facility and recomputes headroom from supplied data; no compliance verdict.
regulatory-calendarBuilds a 12-month filing calendar from fund type, registration and AUM, spanning Form ADV/PF/D, 13F/13H, FATCA/CRS, AIFMD and UK SDR, with an event queue, watch list and 90/60/30-day alerts.
performance-calcComputes IRR, TVPI, DPI, RVPI, MOIC, and PME from cash flows or a capital account statement, with helper bisection workpapers and J-curve analysis.
cashflow-forecastProjects capital calls, distributions, and NAV from explicit dated assumptions using deterministic Decimal workpapers, with scenario comparisons.
scenario-modelRuns user-specified named scenario schedules through deterministic workpapers: NAV, calls, distributions, TVPI/DPI, dated IRR and single-event waterfall tiers, with explicit sensitivity runs.
fee-auditRecomputes management fees and offsets with the packaged helper, checks carry terms against the supplied LPA and lists exceptions; pref accrual and clawback use supplied workpapers only.
investor-onboardPrepares an onboarding checklist by investor type: documents to collect, the qualification, tax and screening reviews to run, owners and a timeline.
ddq-respondDrafts answers to an investor DDQ or RFP mapped to the 20 ILPA DDQ 2.0 topics, citing sources, grading answers A-F and flagging questions that need human input.
entity-mapMaps a fund structure into a Mermaid diagram, entity registry, ownership chain and fee-flow tables, flagging blocker, feeder and ERISA plan-asset gaps.
data-roomDrafts a data room index across eight categories with Required/Recommended/Optional status, gaps from a supplied document list and NDA-gated access tiers, for human approval.
portfolio-dashboardAggregates several funds' supplied statements into one LP view: TVPI/DPI/RVPI, strategy and GP mix, J-curve stage, and unfunded commitments.
reconcileMatches two fund documents line by line (capital call vs capital account, distribution vs K-1, sequential statements, PCAPs vs financial statements) and lists every break with helper workpapers.
review-deedReviews a deed of adherence, equity wrapper, or management participation agreement: obligations assumed, vesting, good and bad leaver terms, and information access limits.
investor-classifyDrafts a preliminary investor classification from supplied documents, with the document requirements checklist and the FATCA, CRS, ERISA, UBTI and ECI questions to resolve.
fatca-crs-classifyReviews W-8, W-9 and CRS self-certifications into a preliminary FATCA Chapter 3/4 and CRS classification, marking unresolved facts and form gaps incomplete.
suitability-checkReviews evidence for the accredited investor, qualified purchaser and qualified client tests, recording each as supported or incomplete; does not determine eligibility.
aml-screenReviews supplied AML/CFT evidence: identity, ownership, PEP/sanctions records and funds/wealth; preserves missing checks and potential matches; no live screening.
kyc-trackerTracks KYC collection across every investor in a fund: per-investor completeness, outstanding document matrix, expiry alerts and refresh cycles by risk rating, with an Excel tracker.
subscription-processCoordinates the classification, documents, suitability, FATCA/CRS and screening reviews into one evidence and gap report; admission stays a separate human decision.
ingestClassifies a folder of mixed fund PDFs, routes each to its /fund review-* skill, runs reconciliation on matching pairs and the portfolio dashboard at 3+ funds.
ingest-emailParses an email or .eml and its attachments, then cross-validates the amounts, deadlines and fund names it claims against the documents to flag mismatches and wire-change red flags.
monitor-emailReads an explicitly authorised Gmail scope for fund emails, stages PDF, XLSX and DOCX attachments in an isolated review run and logs provenance.
multi-doc-analyzeCross-references 2+ related fund documents for provision conflicts, amendment lineage, and MFN cascades.
smart-contract-simRuns the packaged whole-fund waterfall helper on user-supplied scenarios: tier traces, edge cases and pseudocode; clawback and deal-by-deal stay UNCALCULATED.
chatConversational query interface over fund documents and prior /fund outputs, answering with document, section, and exact-quote citations and routing to the command that fills a data gap.
amendment-draftDrafts a formatted LPA amendment with recitals, amendment clauses, ratification, signature blocks, and a redline of old versus new text.
esg-reportDrafts a preliminary ESG report for applicable SFDR, climate, EDCI, PRI and Taxonomy frameworks from supplied data, with explicit gaps; does not certify SFDR compliance.
tax-reviewAnalyzes a K-1 or tax package for UBTI, ECI, Section 199A QBI, 163(j) limits, state filing questions, PFIC exposure and capital account reconciliation.
board-memoDrafts a board, LPAC, annual or special meeting memo: agenda, sourced performance, conflict disclosures, LPAC process checklist and GP-owned recommendation slots.
wire-verifyFlags bank/beneficiary changes and missing callback evidence for wire instructions; prepares a masked checklist; no payment approval.
Coming next · Workspace in development
The skills are the start.
A shared workspace comes next.
FundAdmin Skills is available now. The workspace is an early working prototype in development; public availability and launch timing are still to be confirmed.