HomeSkillsSkills for Strategy and FinanceAndrew PershSeptember 5, 202611 min read

50 GPT-6 Astra Skills for Strategy and Finance Teams

50 free GPT-6 Astra skills, one folder each, that run strategy and finance as one system across six stages: diagnose the decision, map the market, choose the option, execute the plan, govern the risks and communicate the numbers. Built for ChatGPT, Projects and Codex. Download the pack and run the engagement yourself.

AI Academy for Professionals

Learn More2,500+ consultants, bankers, private equity professionals

50 GPT-6 Astra skills that run strategy and finance as one system

This pack is 50 GPT-6 Astra skills for strategy and finance teams, shipped as 50 SKILL.md folders across six stages: Diagnose, Map, Choose, Execute, Govern and Communicate. Half of the skills are finance-native or blend finance into a strategy method, so every strategic choice arrives with its economics attached: a baseline, a horizon, a currency and a labelled evidence trail. Each skill teaches the model one named method for one job, in the format OpenAI documents for ChatGPT and Codex.

The pack was rebuilt for GPT-6 Astra rather than ported. OpenAI describes the model as its most capable option for difficult end-to-end work across research, software and professional tools. Its prompting guide also identifies three habits the skills have to manage: Astra defaults to lists, tables and heavy Markdown, asks more clarifying questions than prior versions, and is more sensitive to instructions inside skills and files. So every skill pins its output shape, sets an autonomy rule and states which instruction wins when two collide.

It is built for the people who own the number as well as the story: strategy and corporate development teams, FP&A, private equity operating partners, and consultants running an engagement without a bench behind them. The same six-stage thinking, packaged for a different assistant, sits in our 50 Tier-1 strategy skills for Claude and our finance agents for Claude. This page is the version that speaks GPT-6 Astra's dialect.

Download all 50 skills

One zip, 50 folders, six stages, free, no signup

Download the zip
A colleague at a whiteboard walking a meeting room of professionals through an AI workflow

Be the person your team asks about AI

Learn it once, then set the standard the rest of the team follows

Go to Academy

What changed with GPT-6 Astra, and what the skills do about it

OpenAI documents each of these behaviors for GPT-6 Astra on its own pages. Each card pairs one documented behavior with the rule every skill in the pack applies to it. None of the rules is a workaround for a weak model. They are the settings that turn a capable model into a repeatable analyst.

Astra behavior

Defaults to lists, tables and heavy Markdown, according to OpenAI's prompting guide

What the skill does

The Output block pins concise paragraphs first, permits compact tables only when comparison is clearer, and bans trailing summaries

Astra behavior

Asks more clarifying questions than prior versions, per the same guide

What the skill does

An Autonomy rule: bias to action, ask only when the answer would change the outcome, otherwise state the assumption and proceed

Astra behavior

More sensitive to contextual instructions inside skills and files

What the skill does

An instruction priority line at the top: your chat instructions beat the skill, and any pause must name the rule that caused it

Astra behavior

Supports web search, file search and code execution for grounded professional work

What the skill does

The Tool use block routes current facts to opened primary sources and all material calculations to executable code

Astra behavior

Supports async tool calls and multi-agent orchestration for long, multistep workflows

What the skill does

Independent research and modelling can run in parallel, while shared assumptions and the model of record keep one owner

Astra behavior

The API model supports a 1,050,000-token context window

What the skill does

A source ledger and evidence inventory still control what enters the answer, so a large context does not become an unreviewed data dump

The rules come straight from OpenAI's prompting guide for GPT-6 Astra, which is worth one careful read before you run the first skill. The guide describes a more capable model, but no model result removes the need to check a number that enters a decision. Evidence labels, executable calculations and reconciliations stay in every deliverable.

One operating system: Diagnose, Map, Choose, Execute, Govern, Communicate

Six stages, each handing a numbered output to the next. Diagnose turns the mandate into a decision question, sizes the value at stake and sets the momentum baseline. Map builds the market, competitor and profit-pool picture on the same numbers. Choose compares every option against that baseline with its returns, risks and exit conditions. Execute sequences and funds the plan so early value pays for later moves. Govern stresses the case, watches the assumptions and proves the value reached the P&L. Communicate turns the approved numbers into one message for the board, the committee or the investor. Install every stage, or pull the one skill you need today.

Six stages, one operating system

Diagnose · 8 skills

Frame the decision and size what is at stake

Turn a vague mandate into a decision question with the money attached

Decision Question Framing

decision-question-framing

Use when: a request arrives as a topic or a symptom rather than a decision someone must make by a date

Output: A one-page decision statement with the decision owner, options in play, deadline, constraints and the value at stake range

Evidence Inventory

evidence-inventory

Use when: before analysis starts to know which data exists, which is missing and which proxies are acceptable

Output: A source register table listing each evidence item, owner, date, quality grade, gap and the proxy plan for each gap

Business Model Teardown

business-model-teardown

Use when: you need to explain how the business makes and keeps money before you judge any option

Output: A business model map covering revenue logic, cost logic, key assets, choke points and the economic flywheel with weak links flagged

Value at Stake Sizing

value-at-stake-sizing

Use when: at the start of any engagement to size the prize before choosing where to spend analytical effort

Output: A full-potential gap table by lever with current, achievable and stretch EBITDA and the implied enterprise value uplift at a stated multiple

Economic Issue Tree

economic-issue-tree

Use when: you need a MECE breakdown of the question where every branch ends in a number someone can measure

Output: A MECE issue tree with each leaf tagged by driver, current value, data source, hypothesis and the analysis that will test it

Margin Bridge Diagnosis

margin-bridge-diagnosis

Use when: profit has moved between two periods and leadership needs to know exactly why, in numbers

Output: A price, volume, mix and cost bridge from period one to period two that reconciles to reported figures with a one-line cause per bar

Unit Economics Diagnosis

unit-economics-diagnosis

Use when: growth is being funded and nobody can show whether each additional unit or customer earns its keep

Output: A unit economics table by segment with contribution margin, acquisition cost, lifetime value, payback period and the break-even volume

Momentum Case Baseline

momentum-case-baseline

Use when: before evaluating any option so every alternative is measured against what happens if nothing changes

Output: A three to five year do-nothing P&L and cash forecast with known headwinds, contract roll-offs and cost inflation, plus the assumption list

Map · 9 skills

Map the market and where the money sits

See customers, competitors and profit pools as one economic picture

Addressable Revenue Model

addressable-revenue-model

Use when: a market size exists but nobody has turned it into revenue the company can actually win

Output: A bottom-up revenue build from addressable market to obtainable revenue by year, with share ramp, price and penetration assumptions sourced

Competitor Economics Teardown

competitor-economics-teardown

Use when: you need a competitor's cost structure and returns rather than a description of its strategy

Output: A reconstructed competitor P&L and ROIC decomposition with each line marked as reported, derived or estimated and the confidence level

Profit Pool Migration

profit-pool-migration

Use when: the question is where margin will sit across the value chain in five years, not where it sits today

Output: A value chain profit pool map for today and a projected future state, with the forces moving each pool and the implied position to take

Segment Profitability Map

segment-profitability-map

Use when: segments are defined by needs or size and nobody knows which ones actually make money

Output: A segment P&L with cost-to-serve allocated, a whale curve of cumulative profit by customer, and the serve, fix or exit call per segment

Cost Structure Benchmarking

cost-structure-benchmarking

Use when: you need to know whether the cost base is out of line with peers and by how much

Output: A cost ratio comparison by function against named sourced peers, normalised for scale and mix, with the gap-to-best in currency

Strategic Group Map

strategic-group-map

Use when: the competitor list is long and you need to see who competes on the same basis as you

Output: A two-axis strategic group map with each group's basis of competition, mobility barriers and the margin range observed in each group

Moat Assessment

moat-assessment

Use when: you must judge whether an advantage will hold long enough to earn back the investment behind it

Output: A moat scorecard by source of advantage with switching cost evidence, durability rating, erosion threats and the implied margin defensibility

Disruption Economics Scan

disruption-economics-scan

Use when: a technology or regulatory shift is visible and you need its effect on cost curves and prices

Output: A ranked list of shifts with the cost curve or price level each one moves, timing evidence, and the exposure of each revenue line

Demand Driver Map

demand-driver-map

Use when: volume forecasts are extrapolated lines and nobody can say what actually moves demand

Output: A driver map linking demand to macro, customer and substitution factors, with a leading indicator and data source for each driver

Choose · 9 skills

Choose the option and prove its economics

Every option compared on the same baseline with its returns, risks and exit conditions

Option Economics Comparison

option-economics-comparison

Use when: several strategic options are on the table and each one has been costed differently, if at all

Output: A side-by-side option table with incremental NPV against the momentum case, payback, capital required, key risks, reversibility and the recommended option

Break-Even Case

break-even-case

Use when: the board asks what must be true for this to work rather than what the forecast says

Output: A what-must-be-true statement with break-even volume, price and cost thresholds, payback, margin of safety and a reverse DCF of the implied assumptions

Valuation Triangulation

valuation-triangulation

Use when: a business, unit or target needs a value range that survives challenge from more than one method

Output: A football field chart of value ranges from DCF, trading multiples and precedent transactions, with the drivers behind each range and a reconciled view

Price Architecture Economics

price-architecture-economics

Use when: the pricing model itself is in question and the margin impact of each design has never been quantified

Output: A comparison of price models with pocket price waterfall today, volume response by model, and the margin and cash impact of each design

Deal Economics Case

deal-economics-case

Use when: an acquisition or merger is proposed and the premium must be justified by evidenced combination benefits

Output: A combination benefit build by type with timing and one-off costs, NPV of benefits against the premium, accretion or dilution, and the integration cost line

Portfolio Reallocation

portfolio-reallocation

Use when: capital follows last year's budget and units are never compared on the return they earn

Output: A unit-by-unit ROIC versus cost of capital table with growth outlook, the reallocation frontier, and a funding shift proposal with the value impact

Option Kill Criteria

option-kill-criteria

Use when: an option is approved and nobody has written down what would make you stop it

Output: A kill criteria sheet per option with measurable triggers, the review date for each, the decision owner and the exit cost if triggered

Timing Choice

timing-choice

Use when: the question is not whether to act but whether to act now, wait or follow a first mover

Output: A now versus wait comparison with the value of waiting, cost of delay, first-mover and follower economics, and the recommended timing with its conditions

Partnership Structure Choice

partnership-structure-choice

Use when: the path involves another company and the structure, from licensing to joint venture, is undecided

Output: A structure comparison across licence, alliance, minority stake, joint venture and acquisition with control, capital, risk share and economics for each

A professional at a desk working across a wall of monitors, each running an AI chat that drafts and answers work questions

Become Dangerously Productive with AI

For consultants, bankers and investors who want a compounding edge

Go to Academy

Execute · 8 skills

Execute with a plan the cash can carry

Sequence, fund and staff the roadmap so early value pays for later moves

Funding and Cash Runway

funding-cash-runway

Use when: a plan is approved on NPV and nobody has checked whether the company can fund the cash trough

Output: A monthly or quarterly cash bridge through the plan with the trough, covenant headroom, funding options ranked by cost and the funding recommendation

Self-Funding Sequence

self-funding-sequence

Use when: the roadmap has many initiatives and the order they run in decides whether the programme pays for itself

Output: An initiative sequence ordered by cash payback with a self-funding waterfall by quarter, dependencies, and the point where the programme turns cash positive

Zero-Based Cost Redesign

zero-based-cost-redesign

Use when: a function's cost must fall materially and incremental trimming has already been exhausted

Output: A zero-based budget for the function built from activities and service levels up, with the cost gap to today, the decisions required and a phased path

Span and Layer Review

span-and-layer-review

Use when: the organisation has grown layers and the cost and speed penalty has never been measured

Output: A layer and span analysis by function with target spans, layers to remove, role impacts, the annual cost effect and the transition risks

Transformation Office Setup

transformation-office-setup

Use when: a multi-workstream programme starts and nobody owns the cadence, tracking and escalation

Output: A programme office charter with workstream structure, meeting cadence, tracking templates, escalation paths and the value tracking link

Pilot Design

pilot-design

Use when: an initiative should be tested before scaling and the pilot must produce a number, not a feeling

Output: A pilot design with scope, control group or counterfactual, success metrics with thresholds, duration, cost and the scale or stop rule

Integration Plan

integration-plan

Use when: after a deal closes or is signed and the deal case must become a day-one and hundred-day plan

Output: A day-one readiness list and a hundred-day plan by workstream with benefit owners, milestones, decision dates and the value tracked against the deal case

Capacity Plan

capacity-plan

Use when: the roadmap assumes people and skills that the organisation may not have when it needs them

Output: A demand versus supply view of people and skills by quarter with the gaps, the hire, contract or reassign choice for each gap and its cost

Govern · 8 skills

Govern the risks and track the value

Stress the case, watch the assumptions and prove the value reached the P&L

Scenario Stress Test

scenario-stress-test

Use when: the case has one forecast and leadership needs to see what a recession or price war does to it

Output: Base, downside and severe scenarios with a tornado of the drivers that matter most, the covenant or cash breach point and either an evidenced expected value or an unweighted range

Run-Rate Value Tracking

run-rate-value-tracking

Use when: initiatives claim savings or growth and finance cannot find them in the reported numbers

Output: A value ladder by initiative from identified to validated to implemented to realised, run-rate versus in-year values, and the reconciliation to the P&L

Leading Indicator Design

leading-indicator-design

Use when: the scorecard reports last quarter's results and gives no warning before the financials move

Output: A paired indicator set linking each financial outcome to two or three leading indicators with thresholds, cadence, owner and data source

Competitor Reaction Matrix

competitor-reaction-matrix

Use when: a move is about to be announced and its economics assume competitors stand still

Output: A reaction matrix by competitor with likely responses, their probability and timing, the impact on your case for each and the pre-planned counter

Strategy Review Scorecard

strategy-review-scorecard

Use when: for the quarterly strategy review so the conversation covers assumptions and decisions, not only results

Output: A one-page scorecard with delivery versus plan, which assumptions changed, value realised to date, decisions required and the items to stop

Delivery Risk Register

delivery-risk-register

Use when: a programme is under way and risks are discussed in meetings but never quantified or owned

Output: A risk register with likelihood, value exposure in currency, trigger, owner, mitigation cost and the residual exposure after mitigation

Post-Decision Review

post-decision-review

Use when: six to eighteen months after a major decision to learn what you believed, what happened and why

Output: An after-action review comparing the decision case to outcomes, the assumptions that held or failed, the calibration lesson and the process changes to adopt

Assumption Watchlist

assumption-watchlist

Use when: after a decision to keep the load-bearing assumptions under watch with triggers that force a re-look

Output: A watchlist of the assumptions that carry the case, each with a current reading, tolerance band, data source, owner and the action when it breaches

Communicate · 8 skills

Communicate the numbers so the decision gets made

Boards, committees and investors hear one message backed by numbers they can trust

Board Numbers Narrative

board-numbers-narrative

Use when: the analysis is done and the board needs one story where every number carries a message

Output: A pyramid-structured board narrative with the answer first, a value bridge exhibit, three supporting arguments with their numbers and the ask

Investment Committee Paper

investment-committee-paper

Use when: a capital request goes to a committee and must survive finance scrutiny on its first reading

Output: A committee paper with the ask, the case against the momentum baseline, returns, risks and mitigations, terms, conditions of approval and the recommendation

Equity Story

equity-story

Use when: the strategy must be explained to investors or lenders as a value creation case rather than a plan

Output: An equity story with the value creation levers, the evidence behind each, the financial trajectory in ranges, the proof points to watch and the risks stated plainly

Decision Meeting Design

decision-meeting-design

Use when: a meeting is meant to produce a decision and previous ones produced discussion instead

Output: A meeting design with the decision to be taken, pre-read contents, agenda with timings, the options presented, the decision rule and the follow-up owner

CFO Challenge Rehearsal

cfo-challenge-rehearsal

Use when: before presenting a case so the hardest finance questions are answered before they are asked

Output: A list of the toughest finance questions ranked by damage, the evidenced answer to each, the questions with no good answer yet and the fix for each

Stakeholder Value Map

stakeholder-value-map

Use when: a decision affects several groups differently and support depends on what each one gains or loses

Output: A stakeholder map with each group's economic gain or loss from the decision, their likely position, the influence they hold and the engagement move for each

Financial Exhibit Design

financial-exhibit-design

Use when: a number must go on a slide and the wrong chart would hide the message or invite doubt

Output: An exhibit specification naming the chart type, the data series, the action title, annotations, source note and the numbers that must be reconciled before it ships

Assumption Disclosure Note

assumption-disclosure-note

Use when: numbers leave the team so readers know what was assumed, estimated and left unknown

Output: A disclosure note listing each material assumption with its basis, each estimate with its range, the unknowns, the method limitations and the date of the data

A professional surrounded by small AI assistants handling analysis, documents, email, and presentations

You Bought Claude
Now Learn How to
10X Your Work

For consultants, bankers and investors who want a compounding edge

Go to Academy

Setup guide for ChatGPT and Codex

OpenAI documents the same skill format for ChatGPT and Codex. Standalone folders are available in the ChatGPT desktop app, Codex CLI and the Codex IDE extension. The interface changes, but each folder still contains one SKILL.md file with its trigger and workflow.

  1. Step 1

    Download the pack and unzip it

    Download all 50 skills (.zip)

    Unzip the pack and keep each of the 50 skills in its own folder with its SKILL.md file. The zip also carries PROJECT-INSTRUCTIONS.md, a work-order template and the shared quality gates.

ChatGPT desktop

OpenAI documents standalone skills in the ChatGPT desktop app. The same guide covers the shared format and invocation rules for ChatGPT and Codex. Read the build skills guide.

  1. Step 2

    Add the skill folders you need

    Open Skills in the ChatGPT desktop sidebar and add the selected folders. Start with one stage or the few leaf skills named in your work order. OpenAI notes that a very large installed set can crowd the initial discovery list, so explicit invocation is the reliable route for this 50-skill pack.

  2. Step 3

    Invoke a skill by name

    Type @ followed by the skill name, for example @margin-bridge-diagnosis, and add the brief and the files. Without the mention, ChatGPT matches a skill from its description, which is why every description names one compact method and decision job.

  3. Step 4

    Keep the installed set focused

    Add only the stages needed for the current decision. Keep the full zip as the library of record, then rotate leaf skills in as the work moves from Diagnose through Communicate. This keeps automatic matching clear and reduces the chance of running overlapping methods.

Shared operating system

Four shared files keep a multi-skill engagement coherent. Use them in your project or working folder alongside the leaf skills. They are guidance files, not a hidden 51st skill.

  1. Step 5

    Load the shared project instructions

    Add PROJECT-INSTRUCTIONS.md to the project or working folder. It tells Astra to select the minimum skill set, maintain one decision brief, one source ledger, one assumption ledger and one model of record, and carry them across all six stages.

  2. Step 6

    Complete the work order

    Fill in WORK-ORDER-TEMPLATE.md with the decision, owner, deadline, baseline, horizon, currency, approved sources, output and stop condition. Name independent research or modelling streams that Astra may run in parallel, and keep shared artifacts with one owner.

  3. Step 7

    Select Astra and run the quality gates

    Choose GPT-6 Astra in the model selector when it is available on your account. Use high reasoning for valuation, scenario, pricing and consequential choice work. Before one stage feeds the next, run the checks in QUALITY-GATES.md.

Codex

For analysts who work in a repository, Codex loads the same folders from disk. OpenAI documents the paths and the invocation syntax in its build skills guide.

  1. Step 8

    Copy the skills folder into place

    Copy the contents of skills/ into $HOME/.agents/skills/ for every project on the machine, or into <repo>/.agents/skills/ for one repository. Restart Codex, then run /skills to confirm all 50 are listed.

  2. Step 9

    Invoke with the dollar sign

    In Codex a skill is called with $skill-name, for example $break-even-case, followed by the brief and the file paths. Codex also matches skills from their descriptions when you do not name one.

  3. Step 10

    Point Codex at GPT-6 Astra

    In ~/.codex/config.toml set model = "gpt-6-astra" and model_reasoning_effort = "high". If Astra is not listed, keep the skill installed and confirm model availability for your account before changing local configuration.

Try a skill

Name the skill, give it the brief, the files and the constraints, and state the baseline, horizon and currency once. Review every output before it feeds the next stage.

  • "Use decision-question-framing on this brief: the CEO wants a view on entering the adjacent segment before the October board"
  • "Run margin-bridge-diagnosis on the two years of segment P&L in this project and reconcile every bar to reported EBITDA"
  • "Use option-economics-comparison on the three options from the last chat, against the momentum case, five-year horizon, EUR"
  • "Use investment-committee-paper to turn the approved case into a paper for next week's committee"

Push GPT-6 Astra to the limit: the eight-slot work order

This pack turns OpenAI's Astra prompting guidance into an eight-part work order: Goal, Context, Instruction priority, Autonomy, Tools and delegation, Output, Verification and Stop condition. Every SKILL.md in the pack is written on that skeleton. So the fastest way to run a whole engagement step is to write your request the same way and name the skills in the order you want them run. The work order below routes four skills through one market-entry decision: Decision Question Framing, then Addressable Revenue Model, then Option Economics Comparison, then Board Numbers Narrative.

Goal
Decide whether to enter the mid-market segment in the adjacent country before the October board. Recommend enter, wait or stop, with the economics.

Context
Files in this project: three years of management accounts, the 2026 budget, the industry association market study, the competitor filings folder, assumptions.md. Today's date is 5 September 2026.

Instruction priority
My instructions in this chat take precedence over any skill file. If a skill rule makes you pause, name the skill and quote the rule.

Autonomy
Bias towards action and carry the task to completion. Ask only when the answer would change the recommendation. Otherwise state the assumption under Assumptions and proceed.

Tools and delegation
Run decision-question-framing, then addressable-revenue-model, then option-economics-comparison, then board-numbers-narrative. Do all arithmetic in Python and show the code. If subagents are available, run the revenue model and the competitor read in parallel.

Output
Concise paragraphs, answer first, one table per skill. Tag every figure Fact, Estimate, Inference, Management assumption or Unknown. No "Bottom line", no trailing summary.

Verification
Totals tie to the management accounts. Currency is EUR, horizon is five years, baseline is the momentum case. Every external figure has a URL and a date.

Stop condition
Deliver when the four outputs and the verification checklist are complete. Do not add unrequested analyses.

Each slot does one job. Goal and Stop condition bound the work. Instruction priority and Autonomy stop the model from either over-asking or wandering. Tools and delegation name the skills and force the math into code. Output and Verification make the result reviewable by someone who was not in the chat.

Running a whole engagement on Astra

  • Run independent evidence skills in parallel chats; Competitor Economics Teardown, Addressable Revenue Model and Unit Economics Diagnosis do not need each other's output
  • Keep one assumptions file in the project and make every skill read from it and write back to it, so the baseline never drifts between chats
  • Use Astra for the demanding reasoning stages; move routine formatting to a lower-cost model only when the handoff preserves the approved numbers and source ledger
  • Checkpoint the decision brief, assumption ledger, source ledger and model after every stage so a long engagement can continue without baseline drift
  • Put formulas, sources and checks in the deliverable itself; a polished recommendation is not an audit trail unless another reviewer can reproduce it

How to choose a skill

Match the question on your desk to the skill that answers it. The grid below picks one skill for each of eight common questions; the pack ships 50 across the six stages. For model build and audit work that lives in a workbook, our Claude skills for Excel modeling page covers the spreadsheet patterns that pair with the finance skills here.

Your question
Skill and stage
What decision are we actually making, and by when?
Decision Question Framing · Diagnose
Why did profit move between last year and this year?
Margin Bridge Diagnosis · Diagnose
How much of this market can we really win?
Addressable Revenue Model · Map
Which customer segments make money and which cost us?
Segment Profitability Map · Map
What must be true for this investment to pay back?
Break-Even Case · Choose
Can we fund the cash trough this plan creates?
Funding and Cash Runway · Execute
Where are the savings we claimed in the reported P&L?
Run-Rate Value Tracking · Govern
How do we get the committee to approve this on first reading?
Investment Committee Paper · Communicate

From approved numbers to slides with Oria

The Communicate skills do not end with a paragraph; they end with an exhibit specification. Financial Exhibit Design names the chart type, the data series, the action title, the annotations and the source line, and lists the numbers that must reconcile before the chart ships. Board Numbers Narrative and Investment Committee Paper produce the same specification for every exhibit they call for.

Oria turns that specification into editable PowerPoint slides in your own template. It runs inside PowerPoint as an add-in, so the chart, the action title and the source line land as native objects your team can edit, not as an image. The exhibit specification is how Astra's analysis carries into a deck your brand team has already approved. Oria also runs inside ChatGPT, and the install guide for Claude and ChatGPT covers both. For the full path from a chat transcript to a finished deck, read how to turn ChatGPT output into a board-ready deck.

Get the 50 skills

Six stages, one zip, ready for ChatGPT, Projects and Codex

Download the zip

The quality bar

Every skill holds its output to the standard a CFO applies before a number leaves the room. These eight checks sit in the verification block of every deliverable skill, and the model is told to fail the deliverable rather than skip one.

Every material calculation executed in Python or a spreadsheet, with formulas or code shown
Every number labelled Fact, Estimate, Inference, Management assumption or Unknown
One baseline, the momentum case, for every option
Horizon and currency stated on every table
Every external source carries a URL and a date
Unknowns listed, never filled with a plausible guess
Every exhibit carries an action title, not a topic title
Methods described by what they do, never by a firm's name

Frequently asked questions

What is in the GPT-6 Astra skills pack?

The pack holds 50 SKILL.md files, one folder each, organised into six stages: Diagnose, Map, Choose, Execute, Govern and Communicate. Half are finance-native or blend finance into a strategy method. Each skill pins its decision, evidence, tool, output and verification contract. The zip also includes shared project instructions, a work-order template and quality gates.

Is the pack for ChatGPT or for Codex?

Both. OpenAI documents standalone skills in the ChatGPT desktop app, Codex CLI and the Codex IDE extension. The same SKILL.md folders work across those surfaces. ChatGPT uses @ mentions and Codex uses $ mentions, while either product can also match a skill from its description.

Do I need to install all 50 skills?

No. Install the full library only when you want broad discovery. For a focused assignment, install one stage or a few leaf skills and invoke them by name. OpenAI notes that very large skill lists may shorten or omit descriptions from initial discovery, so explicit invocation is the reliable route for a 50-skill pack.

How is this different from the Claude skill packs on this site?

The frameworks overlap, but the operating rules are tailored to Astra. Each skill sets instruction priority, scopes autonomy, specifies tool use, identifies work that can run in parallel and pins a concise output shape. It also treats strategy and finance as one system rather than two libraries.

Do the skills do the math?

Yes, and never as unsupported mental arithmetic. Every finance skill instructs the model to run arithmetic, NPV, bridges, sensitivities and reconciliations in Python or a spreadsheet, show the formulas or executable logic and report the computed figure. Each skill ends with a reconciliation step: totals tie, units and currency match, and every figure carries an evidence label.

How do the six stages fit together?

Each stage hands a numbered output to the next. Diagnose produces the decision question and the momentum baseline. Map builds the market and profit picture. Choose compares every option against that same baseline. Execute turns the chosen option into a funded sequence. Govern stresses the case and tracks value to the P&L. Communicate turns the approved numbers into the board story and its exhibits.

Where does Oria fit?

Oria turns the approved numbers into slides. The Communicate skills end with an exhibit specification: chart type, data series, action title and source line. Oria reads that specification inside PowerPoint and produces editable native slides in your own corporate template, so the deck matches the house style without rebuilding charts by hand. It also runs inside ChatGPT.