Reading room · riskmandate.ai

On this page

The memo queue
The four steps
What happened to each of them
An article on the risk propagation visualiser: user stories first, then the story, business analysis, and the technical analysis last, with captures of the figure at every step, each connection alone and combined, each control alone and all of them
How it technically works: prompts first, then hope is not a control, then fit, integrate, graph, and the vault as provenance
UK support, in the open: consolidate what the UK offers a startup at go-to-market, ask people what is missing, and let other founders use it
Pilots do not stay in production: the business sees the gap between mandate and reach, at machine speed, and declines to sign for it
Calendar edits cannot be undone: integrity risk, and why the edit permission is the dangerous one
Business cases by risk reduced: the register without a security product and with it, from the operator to the board, starting with our own
OWASP and open source first: a semantic graph of OWASP, business cases for the open-source projects that reduce risk, and the companies built on open source
A behaviour policy for everybody the lead talks to: a brief and a zip for a new agent, three vaults, and controls so nothing leaks across
An interview page, and a ChatGPT voice prompt to run it: a reusable pattern, and the first page, for a founder who knows UK events and marketing
Ask the ABP questions in the interview, and answer a LinkedIn role map as a stand-alone article: a generic framework, adjusted in every company, with accountability that holds on the way up
Graph visualisations for the role-ownership article: the blast radius as the employee connects and disconnects, the flows played out, the evidence, and the settings a mail scope cannot narrow
Risks always flow upwards: the roles above carry the aggregate, click a role to see what it holds, list every risk that holds now, and show the level of risk the business already accepts against the gap outside it
Red travels up the path, and a control does not make a risk zero: it leaves a green one. Enough controls, including a proxy in the middle, to make every path green; risks for the CFO and Legal
The named professional, and the assignment of the individual who does the £1,500 review
Voice memo: the next phase is users. The MVP is nailed and the vaults are more than most first users need, so find many smaller examples where a policy adds value now. The abp.sgit.ai/gmail prompt workflow - four steps, thirteen prompts, about twenty minutes - is the shareable artefact, already sent to a real group. KPIs and OKRs are the number of users who run the prompts and the number who create a policy; the pricing ladder exists, what is missing is a bigger free funnel above it. Nobody buys a full policy until they have made a smaller one. And the second frontier, beyond technical permissions: consequences in the user's own terms (relabelling destroys a mailbox's information architecture, marking read things unread, too many sends, drafts or moves), which messages should and should not be read or replied to, prompt injection, and rules of engagement for a single session.
Voice memo: the multi-layer grant, and the vault opens on the audience. Map three storeys — what the Google OAuth permission allows, what Anthropic's connector exposes to the chat (code, which can have bugs and can change at any minute's notice), and what normal usage and Claude's own guardrails allow — then the layer above, which is what becomes possible with those mechanical capabilities. Use this vault as the case study and change whatever we want on it. Find good ways to visualise it, because that is where the gaps are and the question is whether people are aware. The vendor's own restriction is a control we can reverse engineer and make the business case for. On the vault: three audience folders (executive, operator, risk) as the main section of the left menu, the executive view being the whole thing in one line, a couple of diagrams and a couple of paragraphs; keep the packs and the how-to-use, but not centre stage.
Build a specific vault for Claude chat connected to a Gmail inbox; capture the connection screens (address obscured) in a folder for the connector; then map the whole customer workflow — the vault, its home page, the settings, the permissions, the prompts given to Claude — as the purchase workflow. Do not use the agent's name.
Where is the new Claude + Gmail section; publish the vault (the read key makes the publish automatic); create a page in the admin section for the Gmail workflow brief, in a way that takes many briefs; and refactor the admin section to the structure, layout and capabilities of store.sgit.ai/admin/ and pt.newsroom.sgit.ai/newsroom/.
Make the Gmail-connector vault oc433z3m the first MVP vault, with a solid end-to-end experience and the design template every other vault will reuse; global changes to the code vault are fine, and backwards compatibility with vaults nobody uses is not worth buying. Start from the store's V3 marketplace mock-up (the vault panel with its left navigation, the positioning of the vault) and the store's comparison table of what each level gets you, including the dual licence.
Voice memo: the material to add to the Gmail-connector vault. Go back to first principles on the grant and map its side effects: a grant is the union of capabilities, and the reader needs the consequences, each explicit and each tied to the asset that makes it real (secrets in mail, reset links, mail from others); count the routes out; the authorisation to read is not the authorisation to forward; harvesting, mass send and what makes a platform suspend an account; mass change to the inbox's filing; realistic scenarios on the mandate; standards as mini-graphs in the vault; the vault navigated as a website with materials per audience.
Risk Mandate Website Repositioning Strategy
ABP Graph and Stakeholder Views — the policy is a graph, and every stakeholder gets a projection of it
Use Case Driven ABP Policy Strategy — a policy per use case, and the £500 level is a prompt the customer runs
The offer is built and the button is not — each level is the level below plus one thing, and the post-sale page does not exist yet
Add a section answering the questions people actually ask in public — with the standing rule that the site never names who asked. Answers that outgrow a section get their own page.
The Grant Is User Shaped And Not Data Shaped: Start With The Connectors, And The Template Vault Is The Product
This Is The Tier One Application Nobody Could Find: The Connector List Alone Is Twenty Bits, So Compute Locally And Submit Banded
The Commit Author Is A Free Text Field: A Prompt Shifts The Odds, And Every Documented Fix Was Architectural
The Urgency Is Not A Deadline But A State: You Already Connected It, And A Distributed Skill Cannot Carry A Control
Startup Summit 2026 — exhibitor booth guide
Every Routable Address Is In The Grant: Do Not Attack Anyone Is The One Rule Everybody Signs, And It Is The One With No Barrier
Voice memo: put the Agent Behaviour Policy at the centre of the site. The ABP is the fundamental primitive; the grant is calculated from reality via digital twins; RiskMandate drives the sale of ABPs; corporate users, investors and founders are the audiences; and a security vendor whose controls reduce the delta has a business case we can make for them.
Hire a freelancer who also works through agents; give them a page and a first prompt focused on sales online and at Lisbon, and make their first task being a power user and tester of ABPs.
Preserve the Lab's thinking as it changes, and publish it as PDFs that can be sent through a chat app rather than linked.
The Startup Summit exhibitor pack and the event site, for a strategy document and the materials to submit.

Reading room / riskmandate.ai · raw text · live ↗

From riskmandate.ai, the page as fetched on 2026-09-25 · open the live page ↗Everything on this sheet is the source site's own text; the newsroom's chrome is outside it.

The memo queue

Every document, memo and instruction riskmandate.ai was built from, in the order it arrived, with what it produced and what it still owes — read off briefs-register.json. Source: https://riskmandate.ai/admin/memos/↗ · noindex · written by scripts/site/build-admin.mjs

2received, nothing built yet*Archived and read, and nothing has been built from it yet.*

25partly worked*read in full; a named part is not built*

8processed*something on the site exists because of it*

0superseded*kept because the reasoning is the record*

The queue. A document or a memo from the project lead arrives — a dev brief as a file, a voice memo as a transcript, an instruction in the thread — and is archived exactly as received: files by SHA-256 in the register ↗, so an identical file arriving again is recognised before anybody reads it; speech and chat as an entry with no digest. Then it is read into a brief ↗, and the brief is broken into units of work ↗. What was said and what we made of it are kept apart on purpose.

An article on the risk propagation visualiser: user stories first, then the story, business analysis, and the technical analysis last, with captures of the figure at every step, each connection alone and combined, each control alone and all of them

spoken brief, transcribed · 1 produced · 2 not done25 September 2026 · partly

How it technically works: prompts first, then hope is not a control, then fit, integrate, graph, and the vault as provenance

spoken brief, transcribed · 1 produced · 3 not done24 September 2026 · partly

UK support, in the open: consolidate what the UK offers a startup at go-to-market, ask people what is missing, and let other founders use it

spoken brief, transcribed · 2 produced · 3 not done24 September 2026 · partly

Pilots do not stay in production: the business sees the gap between mandate and reach, at machine speed, and declines to sign for it

spoken brief, transcribed · 1 produced · 3 not done24 September 2026 · partly

Calendar edits cannot be undone: integrity risk, and why the edit permission is the dangerous one

spoken brief, transcribed · 1 produced · 3 not done24 September 2026 · partly

Business cases by risk reduced: the register without a security product and with it, from the operator to the board, starting with our own

spoken brief, transcribed · 3 produced · 3 not done24 September 2026 · partly

OWASP and open source first: a semantic graph of OWASP, business cases for the open-source projects that reduce risk, and the companies built on open source

spoken brief, transcribed · 3 produced · 3 not done24 September 2026 · partly

A behaviour policy for everybody the lead talks to: a brief and a zip for a new agent, three vaults, and controls so nothing leaks across

spoken brief, transcribed · 2 produced · 3 not done24 September 2026 · partly

An interview page, and a ChatGPT voice prompt to run it: a reusable pattern, and the first page, for a founder who knows UK events and marketing

build brief from the sgit.ai site team, the page as served, kept byte for byte · 2 produced · 3 not done24 September 2026 · partly

Ask the ABP questions in the interview, and answer a LinkedIn role map as a stand-alone article: a generic framework, adjusted in every company, with accountability that holds on the way up

spoken brief, transcribed · 2 produced · 3 not done24 September 2026 · partly

Graph visualisations for the role-ownership article: the blast radius as the employee connects and disconnects, the flows played out, the evidence, and the settings a mail scope cannot narrow

spoken brief, transcribed · 2 produced · 3 not done24 September 2026 · partly

Risks always flow upwards: the roles above carry the aggregate, click a role to see what it holds, list every risk that holds now, and show the level of risk the business already accepts against the gap outside it

spoken brief, transcribed · 3 produced · 2 not done24 September 2026 · partly

Red travels up the path, and a control does not make a risk zero: it leaves a green one. Enough controls, including a proxy in the middle, to make every path green; risks for the CFO and Legal

spoken brief, transcribed · 2 produced · 2 not done24 September 2026 · partly

The named professional, and the assignment of the individual who does the £1,500 review

spoken brief, transcribed · 5 produced · 5 not done22 September 2026 · partly

Voice memo: the next phase is users. The MVP is nailed and the vaults are more than most first users need, so find many smaller examples where a policy adds value now. The abp.sgit.ai/gmail prompt workflow - four steps, thirteen prompts, about twenty minutes - is the shareable artefact, already sent to a real group. KPIs and OKRs are the number of users who run the prompts and the number who create a policy; the pricing ladder exists, what is missing is a bigger free funnel above it. Nobody buys a full policy until they have made a smaller one. And the second frontier, beyond technical permissions: consequences in the user's own terms (relabelling destroys a mailbox's information architecture, marking read things unread, too many sends, drafts or moves), which messages should and should not be read or replied to, prompt injection, and rules of engagement for a single session.

voice memo, transcribed in the thread, with a screenshot of the Agentics Foundation London group, 21 Sept · 2 produced · 6 not done21 September 2026 · received

Voice memo: the multi-layer grant, and the vault opens on the audience. Map three storeys — what the Google OAuth permission allows, what Anthropic's connector exposes to the chat (code, which can have bugs and can change at any minute's notice), and what normal usage and Claude's own guardrails allow — then the layer above, which is what becomes possible with those mechanical capabilities. Use this vault as the case study and change whatever we want on it. Find good ways to visualise it, because that is where the gaps are and the question is whether people are aware. The vendor's own restriction is a control we can reverse engineer and make the business case for. On the vault: three audience folders (executive, operator, risk) as the main section of the left menu, the executive view being the whole thing in one line, a couple of diagrams and a couple of paragraphs; keep the packs and the how-to-use, but not centre stage.

voice memo, transcribed in the thread, 16 Sept · 1 produced · 5 not done16 September 2026 · received

Build a specific vault for Claude chat connected to a Gmail inbox; capture the connection screens (address obscured) in a folder for the connector; then map the whole customer workflow — the vault, its home page, the settings, the permissions, the prompts given to Claude — as the purchase workflow. Do not use the agent's name.

chat instruction with eight screenshots, 16 Sept · 5 produced · 2 not done16 September 2026 · partly

Where is the new Claude + Gmail section; publish the vault (the read key makes the publish automatic); create a page in the admin section for the Gmail workflow brief, in a way that takes many briefs; and refactor the admin section to the structure, layout and capabilities of store.sgit.ai/admin/ and pt.newsroom.sgit.ai/newsroom/.

chat instruction, 16 Sept · 3 produced · 0 not done16 September 2026 · processed

Make the Gmail-connector vault oc433z3m the first MVP vault, with a solid end-to-end experience and the design template every other vault will reuse; global changes to the code vault are fine, and backwards compatibility with vaults nobody uses is not worth buying. Start from the store's V3 marketplace mock-up (the vault panel with its left navigation, the positioning of the vault) and the store's comparison table of what each level gets you, including the dual licence.

chat instruction with a PDF and a link, 16 Sept · 2 produced · 1 not done16 September 2026 · partly

Voice memo: the material to add to the Gmail-connector vault. Go back to first principles on the grant and map its side effects: a grant is the union of capabilities, and the reader needs the consequences, each explicit and each tied to the asset that makes it real (secrets in mail, reset links, mail from others); count the routes out; the authorisation to read is not the authorisation to forward; harvesting, mass send and what makes a platform suspend an account; mass change to the inbox's filing; realistic scenarios on the mandate; standards as mini-graphs in the vault; the vault navigated as a website with materials per audience.

voice memo (Otter transcript, about nine minutes), 16 Sept · 3 produced · 2 not done16 September 2026 · partly

Risk Mandate Website Repositioning Strategy

spoken brief, transcribed · 7 produced · 4 not done15 September 2026 · partly

ABP Graph and Stakeholder Views — the policy is a graph, and every stakeholder gets a projection of it

voice memo (Otter transcript) · 2 produced · 5 not done15 September 2026 · partly

Use Case Driven ABP Policy Strategy — a policy per use case, and the £500 level is a prompt the customer runs

voice memo (Otter transcript) · 3 produced · 5 not done15 September 2026 · partly

The offer is built and the button is not — each level is the level below plus one thing, and the post-sale page does not exist yet

strategy brief · 6 produced · 3 not done15 September 2026 · partly

Add a section answering the questions people actually ask in public — with the standing rule that the site never names who asked. Answers that outgrow a section get their own page.

chat instruction, with two public comments supplied as the first two questions · 1 produced · 1 not done13 September 2026 · processed

The Grant Is User Shaped And Not Data Shaped: Start With The Connectors, And The Template Vault Is The Product

dev brief · 3 produced · 3 not done12 September 2026 · processed

This Is The Tier One Application Nobody Could Find: The Connector List Alone Is Twenty Bits, So Compute Locally And Submit Banded

dev brief · 1 produced · 1 not done12 September 2026 · processed

The Commit Author Is A Free Text Field: A Prompt Shifts The Odds, And Every Documented Fix Was Architectural

dev brief · 2 produced · 3 not done12 September 2026 · partly

The Urgency Is Not A Deadline But A State: You Already Connected It, And A Distributed Skill Cannot Carry A Control

strategy brief · 2 produced · 3 not done12 September 2026 · partly

Startup Summit 2026 — exhibitor booth guide

organiser document · 2 produced · 2 not done12 September 2026 · processed

Every Routable Address Is In The Grant: Do Not Attack Anyone Is The One Rule Everybody Signs, And It Is The One With No Barrier

dev brief · 1 produced · 4 not done12 September 2026 · partly

Voice memo: put the Agent Behaviour Policy at the centre of the site. The ABP is the fundamental primitive; the grant is calculated from reality via digital twins; RiskMandate drives the sale of ABPs; corporate users, investors and founders are the audiences; and a security vendor whose controls reduce the delta has a business case we can make for them.

voice memo, transcribed in the thread · 3 produced · 1 not done12 September 2026 · partly

Hire a freelancer who also works through agents; give them a page and a first prompt focused on sales online and at Lisbon, and make their first task being a power user and tester of ABPs.

chat instruction · 2 produced · 1 not done12 September 2026 · processed

Preserve the Lab's thinking as it changes, and publish it as PDFs that can be sent through a chat app rather than linked.

chat instruction · 2 produced · 0 not done12 September 2026 · processed

The Startup Summit exhibitor pack and the event site, for a strategy document and the materials to submit.

link, in the thread · 2 produced · 0 not done11 September 2026 · processed

The four steps

1 · Capture

The file goes into site/assets/briefs/ byte for byte and its digest into briefs-register.json the moment it arrives. Speech is transcribed and archived the same way; nothing is trimmed.

2 · Read

A brief under docs/briefs/: what was asked, what it changes, what it contradicts, and what is not yet decidable. One page each, here ↗.

3 · Break

Task briefs under .claude/briefs/, each sized for one agent and naming the files it touches, and a row in the queue.

4 · Track

The row moves on the board ↗; the register's status and not done say what the memo still owes.

What happened to each of them

D24 · spoken brief, transcribed · 25 September 2026

An article on the risk propagation visualiser: user stories first, then the story, business analysis, and the technical analysis last, with captures of the figure at every step, each connection alone and combined, each control alone and all of them

partly

the file as received ↗ · f3849c17c9fb… · 1 KB

Produced. The article, in the order asked: seven user stories; the six weeks as a story; how to read the six bands, the two colours and a click; each connection alone (CRM, Mail, Calendar), then combined, where the trifecta appears; eight controls one at a time and then all of them, with a table of what each ends and leaves; the business analysis and five questions for a room; the technical analysis down to the one-way walk. Thirty-four captures, each taken by script in the state its caption names ↗

Not done.

Captures of a real agent. Every picture is of the invented example; running the figure on a published ABP vault is the next step the article names.
A capture per role for all thirteen of the map's roles: the article shows the ones the argument needs (the rep, Sales, Legal, the CIO, the CEO, the board), not each one.

D12 · spoken brief, transcribed · 24 September 2026

How it technically works: prompts first, then hope is not a control, then fit, integrate, graph, and the vault as provenance

partly

the file as received ↗ · dadd026e61c5… · 2 KB

Produced. How it works, rebuilt in the order the memo gives: six steps with the status of each on the step ↗

Not done.

Connectors to a customer's own control planes (identity provider, cloud IAM, proxy, CI). Described as built per engagement and productised as they mature; none exists as a product and the page says so.
The graph's edges upward to risks and the board, and sideways to a customer's internal policies and documents. Standards edges run today; these are marked in design.
Execution logs and evidence held in the vault beside the policy, as the memo's provenance store. Marked as where this goes, not as something that exists.

D13 · spoken brief, transcribed · 24 September 2026

UK support, in the open: consolidate what the UK offers a startup at go-to-market, ask people what is missing, and let other founders use it

partly

the file as received ↗ · 0c528b8c5736… · 3 KB

Produced. UK support, in the open: 79 programmes, events, schemes and networks, each read on its official page and dated, with our own status on every row ↗ · The same register as data, for other founders to copy ↗

Not done.

Applying to anything. Every row says not started except Web Summit; which doors to try, starting with the Sovereign AI procurement challenge, is the lead's decision.
A matching page on sgit.ai's partnerships section. That site is not in this repository; it can link here.
Pages that refused an automated read (British Business Bank, london.gov.uk, blackhat.com, rsaconference.com, the Turing, the ABI) are marked unclear rather than filled from search results.

D14 · spoken brief, transcribed · 24 September 2026

Pilots do not stay in production: the business sees the gap between mandate and reach, at machine speed, and declines to sign for it

partly

the file as received ↗ · 762fd1ea85c2… · 7 KB

Produced. The pilot worked. Then somebody asked what else it could do: fourteen surveys and forecasts with the causes each names, seven cases, five variables the lethal trifecta leaves out, and what the data does not show ↗

Not done.

Data that tests the hypothesis. No survey found asks whether the gap between reach and mandate stopped sign-off; the article says so and asks readers for cases.
The credit approval at two in the morning has no public source. It is told as an anecdote the author was told, and nothing more.
The operating limit, headroom and hard maximum the memo recalls from an insurance vault were not found in the published demo vaults; the article presents the shape with invented numbers, labelled as such.

D15 · spoken brief, transcribed · 24 September 2026

Calendar edits cannot be undone: integrity risk, and why the edit permission is the dangerous one

partly

the file as received ↗ · 0349fc94e7df… · 5 KB

Produced. A deleted meeting comes back. An edited one does not: what Google keeps after each action, in its own words, for personal and Workspace accounts; three contradictions in Google's pages; and the draft Calendar rows of a behaviour policy ↗

Not done.

The Calendar behaviour policies themselves. They are in draft; the article quotes their rows and says so.
A restore for edited events. Google documents none for users; Workspace with Vault keeps earlier versions an admin can export, and the article says the memo's claim holds for personal accounts and needs that qualification for Workspace.
The early-user conversation (calendar over email) is reported as a handful of conversations, not a survey.

D16 · spoken brief, transcribed · 24 September 2026

Business cases by risk reduced: the register without a security product and with it, from the operator to the board, starting with our own

partly

the file as received ↗ · 1dabf7f55b1c… · 6 KB

Produced. Business cases, by the risk they change: the method, the rules, and twelve categories computed against the typical deployment ↗ · The first case, our own: an Agent Behaviour Policy retires two risks, names one hidden one, and lists three expectations as reductions, never retirements ↗ · The engine and a copy of the RiskGraph Explorer model with its provenance, so every register is computed rather than written ↗

Not done.

The two named cases (agentgateway; Auth0 for AI Agents) are built as drafts, unlisted and marked noindex, until the lead decides to send them to the vendors.
A semantic graph per product beyond the model's sixteen questions. A case is written as the answers a product changes; products that act on things the questions do not ask about cannot yet be expressed.
Outreach to the vendors. It is the lead's, after review.

D17 · spoken brief, transcribed · 24 September 2026

OWASP and open source first: a semantic graph of OWASP, business cases for the open-source projects that reduce risk, and the companies built on open source

partly

the file as received ↗ · 3686c72fb430… · 3 KB

Produced. OWASP, as a graph: the foundation, four families, 52 projects and documents, 110 items of eleven lists by title, and the relationships OWASP states, joined to the risk model through the Agentic Top 10 ↗ · Eighteen open-source business cases, OWASP first, each change backed by the project's own words and each with what adopting it takes ↗ · Eighteen companies built on or beside an open-source project, with only the quotes that were checked word for word ↗

Not done.

Business cases for OWASP documents such as the Top 10s, ASVS and SAMM. They change what a team knows, not what an agent can reach, so they sit in the graph rather than as cases; the bridge shows which cases change the answers each Agentic Top 10 item depends on.
Three Agentic Top 10 items touch nothing in the model: supply chain, memory and context, and agents talking to agents. The model has to grow for those.
Outreach to OWASP projects, maintainers and the companies. The lead's, after review.

D18 · spoken brief, transcribed · 24 September 2026

A behaviour policy for everybody the lead talks to: a brief and a zip for a new agent, three vaults, and controls so nothing leaks across

partly

the file as received ↗ · db4b98aef049… · 2 KB

Produced. The brief: three vaults (the existing app vault as the UI, a private keys vault, one vault per person), the controls against leaking across, the workflow, and what the feedback loop is meant to learn ↗ · The pack for the new agent: instructions, templates, a scaffolder, a leak gate proved against six planted leaks, the site's own builder and every catalogue deployment, and one fictional example built and checked; kept in the repository at packs/dist/, not on the site ↗

Not done.

The keys vault itself. The first session creates it and gives its key to the lead, so the steps are exercised once with the lead watching.
A way for a person to send corrections from inside their vault. Today they reply to the lead; a form is a question for the app vault.
Any person's vault. The pack has been run end to end on a fictional organisation only.

D19 · build brief from the sgit.ai site team, the page as served, kept byte for byte · 24 September 2026

An interview page, and a ChatGPT voice prompt to run it: a reusable pattern, and the first page, for a founder who knows UK events and marketing

partly

the file as received ↗ · c4734574a343… · 25 KB · sgit.ai v0.6.7

Produced. The first interview page, with the six parts in the brief's order, a copy button that copies the prompt exactly, and nothing loaded or sent ↗ · The pattern as a template: site/interviews/_template.json and scripts/site/build-interview-pages.mjs, so the next page is a JSON file ↗

Not done.

Running the prompt once in ChatGPT voice mode. This agent has no ChatGPT account; the lead runs it and checks the summary has all ten sections.
The prompt is the brief's, with two phrases corrected to what the site states: the home page names the CEO, CTO and CISO, not an insurer; and the Licence to Operate page makes the behaviour policy the instrument, not the evidence.
Sending the summary back into a vault through a write-only lane, which the brief marks as later.

D20 · spoken brief, transcribed · 24 September 2026

Ask the ABP questions in the interview, and answer a LinkedIn role map as a stand-alone article: a generic framework, adjusted in every company, with accountability that holds on the way up

partly

the file as received ↗ · 2219d9a05179… · 1 KB

Produced. The founder interview's first part: six questions on the Agent Behaviour Policy (its name, explaining it back after one hearing, what it adds, whether the market understands the problem, its value to the people who would use it, whether it should sell); thirty minutes, sixteen summary sections ↗ · The article: the infographic's thirteen rows quoted and credited; what each company sets and what stays fixed; the middle column as a routing table and the organisation chart as the escalation path; one invented agent from eight ABP rows to the board; seven rules; six weeks; the model as nodes, edges with inverses and computations ↗

Not done.

The LinkedIn post's own address and date. The article credits the infographic by title and author and says when it reached us; the link is the lead's to add.
Running the longer interview prompt once in ChatGPT voice mode, and checking the summary has all sixteen sections.
The seven roles the article needs and the model lacks (COO, CIO / Data, Legal, HR, Marketing, Sales, all employees), authority as data, and an engine that runs the clocks. The article says so in its section 10.

D21 · spoken brief, transcribed · 24 September 2026

Graph visualisations for the role-ownership article: the blast radius as the employee connects and disconnects, the flows played out, the evidence, and the settings a mail scope cannot narrow

partly

the file as received ↗ · 524b59f30d44… · 2 KB

Produced. Three figures drawn by the page's own script from the article's data: the blast radius, which plays the six weeks or takes the reader's own changes (connect, disconnect, boundaries, instructions) and recomputes the rows, the risks, their holders and the counts; the six weeks as a strip; and the ontology, every verb readable from both ends ↗ · The instruction switch: the mandate becomes precise about the business process, every row in the gap gains an expectation, and unbounded excess does not move. The mail scope fact is cited from the Gmail record: three consent lines, each for the whole account ↗

Not done.

Blast radius as a number. The figure shows which roles a risk reaches and who holds it; it does not size the consequence, because the article scores nothing.
Real grants. The figure runs on the article's invented agent; wiring it to a published ABP vault, so a real grant drives the same picture, is the next step.
The flows of evidence downward. A ceased risk cites its ABP version in text; the figure does not yet animate the evidence travelling back to the version.

D22 · spoken brief, transcribed · 24 September 2026

Risks always flow upwards: the roles above carry the aggregate, click a role to see what it holds, list every risk that holds now, and show the level of risk the business already accepts against the gap outside it

partly

the file as received ↗ · 19de617c22e7… · 5 KB

Produced. The blast radius rebuilt: every live risk lights the path from its holder to the board, every role above carries a count, a role's panel lists what it holds, carries and is informed of, and a table lists every risk that holds at that moment with the chain it reaches, its consequence in words and whether it can be undone ↗ · The accepted level: row 1 is the rep's own mail and establishes R0, inside the rep's authority and accepted on connecting; a new row 9, the shared sales inbox, is outside the mandate and establishes R7. The article's tables, counts and six weeks follow: nine rows, five in the gap, unbounded excess five, then four, then three ↗ · Marketing, Product and HR leave the figure; Legal and the CFO are informed of every risk that touches customer data or a customer. A compound risk is drawn larger by the number of facts it needs ↗

Not done.

A quantity of impact or potential loss on a risk. The figure gives each risk a consequence in words and an undo property, and no number: the article's own rule is that kinds of consequence do the work of a score, and a loss figure is the lead's decision.
Other scenarios. The figure runs one agent; the lead said to keep it so for now.

D23 · spoken brief, transcribed · 24 September 2026

Red travels up the path, and a control does not make a risk zero: it leaves a green one. Enough controls, including a proxy in the middle, to make every path green; risks for the CFO and Legal

partly

the file as received ↗ · d47dbfda7dd3… · 2 KB

Produced. The colour travels up: while anything below a role is unaccepted, that role's path, halo and count are red. Four more controls, one of them an execution proxy that carries out every write under the ABP's rules and bounds every write row at once; two fact controls, processing terms with the provider and the insurer's written answer, which end Legal's and the CFO's risks. Every control has a residual risk, green, inside its holder's authority, accepted when it appears; with every control on, every path to the board is green and nine green risks remain. A button does it in one click ↗ · Two more held risks in the article: RL with Legal (customer mail goes to the model provider, and the notice does not say so) and RC with the CFO (nothing in the cover says whether an act of the agent as staff is covered). The tables, the six weeks and the strip follow: nine risks on day 0, nine open and none unaccepted on day 42 ↗

Not done.

The effectiveness of a control. The figure takes every control as working, as the lead said to for now; a control that is on and not working is a fact the ABP's evidence tiers exist to record, and it is not modelled here.
Residual risks are drawn from a fixed list. In a real record they would be written by each holder in their own words when the control lands.

D11 · spoken brief, transcribed · 22 September 2026

The named professional, and the assignment of the individual who does the £1,500 review

partly

the file as received ↗ · d376355f1747… · 2 KB

Produced. The reviewed level — the workflow, the expectations and the figures with their sources ↗ · Who runs your review — the list, and the rule about what gets published ↗ · The first reviewer's page, every line read off a published page ↗ · The shape of a reviewer page, labelled a placeholder ↗ · The manifest the store can build its chooser from ↗

Not done.

A second, named reviewer: nothing about them is published until they have read their own page and agreed to it. The placeholder stands in the meantime.
The chooser at checkout, which is the store's workflow rather than ours. The manifest is published for it; the arrangement is written up for the store team and nothing is agreed yet.
Comments and testimonials. Asked for; none exist, and none will be composed here.
Availability, a queue or a throughput number: none can be honoured yet, so the pages carry a status and the date it was confirmed instead.
What the reviewer is paid — agreed in the brief and deliberately not published. A commercial term, and the lead's call.

I11 · voice memo, transcribed in the thread, with a screenshot of the Agentics Foundation London group, 21 Sept · 21 September 2026

Voice memo: the next phase is users. The MVP is nailed and the vaults are more than most first users need, so find many smaller examples where a policy adds value now. The abp.sgit.ai/gmail prompt workflow - four steps, thirteen prompts, about twenty minutes - is the shareable artefact, already sent to a real group. KPIs and OKRs are the number of users who run the prompts and the number who create a policy; the pricing ladder exists, what is missing is a bigger free funnel above it. Nobody buys a full policy until they have made a smaller one. And the second frontier, beyond technical permissions: consequences in the user's own terms (relabelling destroys a mailbox's information architecture, marking read things unread, too many sends, drafts or moves), which messages should and should not be read or replied to, prompt injection, and rules of engagement for a single session.

received

voice memo, transcribed in the thread, with a screenshot of the Agentics Foundation London group, 21 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the free rung below the ladder, why a small policy is not a small vault, the six things the memo names beyond permissions mapped against the model, the build order and five decisions ↗ · Try it - the seventh top-level section: the four steps, what you end up with, what it honestly is not, the four layers, and where to say it broke ↗

Not done.

A prompt workflow per shape we already have (T13) - the Gmail one is published at abp.sgit.ai and the other fifteen have only MAP-A-GRANT.md, inside the vault
A feedback path from the workflow's last step that survives a stranger
What we count, and where it is published - the site collects nothing, so the honest measure needs the lead
Volume and instances on the consequence layer (T14): the relabelling, the mass send, the draft flood, the mailbox's state as an asset
The session-scoped mandate, and the instance-scoped mandate - proposed as a Lab ask rather than a change to the published vocabulary
Waiting on the lead: the seventh menu slot, what is counted, whether a user-made policy gets published, and how far the free rung goes

I10 · voice memo, transcribed in the thread, 16 Sept · 16 September 2026

Voice memo: the multi-layer grant, and the vault opens on the audience. Map three storeys — what the Google OAuth permission allows, what Anthropic's connector exposes to the chat (code, which can have bugs and can change at any minute's notice), and what normal usage and Claude's own guardrails allow — then the layer above, which is what becomes possible with those mechanical capabilities. Use this vault as the case study and change whatever we want on it. Find good ways to visualise it, because that is where the gaps are and the question is whether people are aware. The vendor's own restriction is a control we can reverse engineer and make the business case for. On the vault: three audience folders (executive, operator, risk) as the main section of the left menu, the executive view being the whole thing in one line, a couple of diagrams and a couple of paragraphs; keep the packs and the how-to-use, but not centre stage.

received

voice memo, transcribed in the thread, 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the four storeys (permitted, exposed, practised, consequences), latent capability as the gap between the first two, the barrier gaining a holder and whether it moves without you, the audience spine, the build order and four decisions ↗

Not done.

Storey one for this shape (data/permitted.json, the scopes and the methods they unlock)
held_by and moves_without_you on every barrier; splitting not_reachable into bounded and not exposed
The audience spine in the reading app, with the executive view authored
The storeys drawn as one diagram
Waiting on the lead: whether grant keeps naming storey two, and the audience names

I6 · chat instruction with eight screenshots, 16 Sept · 16 September 2026

Build a specific vault for Claude chat connected to a Gmail inbox; capture the connection screens (address obscured) in a folder for the connector; then map the whole customer workflow — the vault, its home page, the settings, the permissions, the prompts given to Claude — as the purchase workflow. Do not use the agent's name.

partly

chat instruction with eight screenshots, 16 Sept — no digest; given as speech, chat text or a link

Produced. The template vault claude-gmail-connector: six rows, four measured on the deployer's own account, the evidence transcribed with the address redacted; pushed as vault oc433z3m and read live on its own page ↗ · The customer's draft instance, anonymised, with a question on every mandate line (vaults-instances/) ↗ · The purchase workflow, run once: the brief ↗ · The tile in the library, in Mail & files connectors, measured 4 of 6 ↗ · The customer's draft instance pushed as a private vault (xjir6m0c); the key handed to the lead in the session, never on a page ↗

Not done.

The twelve screenshots as redacted image files in evidence/: they arrived inline and could not be edited from the session; the transcriptions stand in for them
The customer's correction of the mandate, and the signed licence

I7 · chat instruction, 16 Sept · 16 September 2026

Where is the new Claude + Gmail section; publish the vault (the read key makes the publish automatic); create a page in the admin section for the Gmail workflow brief, in a way that takes many briefs; and refactor the admin section to the structure, layout and capabilities of store.sgit.ai/admin/ and pt.newsroom.sgit.ai/newsroom/.

processed

chat instruction, 16 Sept — no digest; given as speech, chat text or a link

Produced. The Gmail-connector vault pushed (oc433z3m) and read live on its page and in the library ↗ · The admin console: a rail with counts, what needs the lead, the board, the memo queue, every brief as a page, the vaults, the records, the tooling — written by build-admin.mjs from the repository ↗ · The Gmail workflow brief as a console page, one of every document under docs/↗

I8 · chat instruction with a PDF and a link, 16 Sept · 16 September 2026

Make the Gmail-connector vault oc433z3m the first MVP vault, with a solid end-to-end experience and the design template every other vault will reuse; global changes to the code vault are fine, and backwards compatibility with vaults nobody uses is not worth buying. Start from the store's V3 marketplace mock-up (the vault panel with its left navigation, the positioning of the vault) and the store's comparison table of what each level gets you, including the dual licence.

partly

chat instruction with a PDF and a link, 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: what the mock-up says about the vault, the feature list read off the store's table, the dual licence as a file and a data block, the redesign of the reading app with a left navigation, what changes globally, four decisions for the lead, and the plan ↗ · Built: the dual licence in the build and LICENCE.md; the v5 left-navigation reading app in a new app vault (vbhmlulo); oc433z3m rebuilt and re-pushed; the vault page repositioned (v1.24.0)↗

Not done.

The live host view of oc433z3m was still resolving the new sub-vault mount at merge; the vault data is verified correct and the renderer proven via a static preview

I9 · voice memo (Otter transcript, about nine minutes), 16 Sept · 16 September 2026

Voice memo: the material to add to the Gmail-connector vault. Go back to first principles on the grant and map its side effects: a grant is the union of capabilities, and the reader needs the consequences, each explicit and each tied to the asset that makes it real (secrets in mail, reset links, mail from others); count the routes out; the authorisation to read is not the authorisation to forward; harvesting, mass send and what makes a platform suspend an account; mass change to the inbox's filing; realistic scenarios on the mandate; standards as mini-graphs in the vault; the vault navigated as a website with materials per audience.

partly

voice memo (Otter transcript, about nine minutes), 16 Sept — no digest; given as speech, chat text or a link

Produced. The direction brief: the memo in its own order, where the vault already is, the consequence layer (assets, consequences, standards mini-graphs, the triple), eleven first consequences for oc433z3m, what it settles for the MVP brief, the build order, what needs the lead ↗ · Task briefs T11 (consequences and assets) and T12 (standards mini-graphs); T04 amended with Who are you? ↗ · Built: assets.json and consequences.json for oc433z3m (eleven consequences, six assets, two routes out), CONSEQUENCES.md derived, the What follows view, the standards mini-graphs, two scenarios (v1.24.0)↗

Not done.

The research list documented from Google's pages — sending limits, suspension, the modify tools — is still open (T05/research-vault)
Who are you? ships as a live filter; the authored per-audience views and projections remain T04

D10 · spoken brief, transcribed · 15 September 2026

Risk Mandate Website Repositioning Strategy

partly

the file as received ↗ · ab6f9486ecae… · 3 KB

Produced. The home page rebuilt around who is arriving and what is sold, with the insurance argument moved off it ↗ · Make agents insurable — the whole insurance argument kept in full on a page of its own ↗ · Licence to Operate — the middle rung, given the page the brief asked for ↗ · You run agents today ↗ · You are a founder ↗ · You are a startup ↗ · The menu restructured to seven top-level entries with Behaviour policy, Who it's for and Insurance as groups ↗

Not done.

MULTILINGUAL DELIVERY FOR TIER 3 AND TIER 4. The brief asks for the £500 correction and the £1,500 session to be deliverable in the buyer's own language, since both have a person in the loop. Nothing is built and nothing is claimed anywhere on the site. It is recorded here so it is not lost.
The four offerings are not yet listed on the site as offerings. The store publishes four tiers and two add-ons; pricing.html still describes the free/paid line rather than the ladder.
An investor page. The brief names three audiences; the printed sheets, the store and the Lisbon page all carry a fourth (investors, or 'you back companies that do'). Three pages were built and the fourth is flagged on the home page as missing.
The vault's per-audience views are described on the home page as the reason for audience pages, but no ABP vault is published on this site yet.

D7 · voice memo (Otter transcript) · 15 September 2026

ABP Graph and Stakeholder Views — the policy is a graph, and every stakeholder gets a projection of it

partly

the file as received ↗ · d93b0fdc381a… · 4 KB

Produced. The direction brief: the policy is a graph, and every stakeholder gets a projection of it (docs/briefs/direction__abp-as-a-graph-and-stakeholder-views.md) ↗ · The by-behaviour facet on the library: one click answers which policies can delete files ↗

Not done.

One page per behaviour, generated from the catalogue
Edges per path rather than per row, in every vault
Metrics (speed, volume, blast) and outward links (ATT&CK, GDPR) on the 23 primitives
Views per audience — CEO, CFO, CTO, investor, buyer, operator, engineer, project manager — with the prompt and the script shipped in the vault so the projection can be regenerated
The two asks to the model site, on Lab 03

Notes. The memo was worked from the transcript on 15 September before the file itself was registered; the file arrived the same day with D8. The transcriber writes the acronym as "ADP" throughout; it is the spoken "ABP", and the archived bytes are kept as received. The task briefs for the unbuilt items are in the repository under .claude/briefs/ (T01–T04, T07).

D8 · voice memo (Otter transcript) · 15 September 2026

Use Case Driven ABP Policy Strategy — a policy per use case, and the £500 level is a prompt the customer runs

partly

the file as received ↗ · 8457a6637e21… · 4 KB

Produced. The direction brief: a policy per use case, and the £500 level is a prompt the customer runs (docs/briefs/direction__use-case-driven-policies-and-the-prompt-workflow.md) ↗ · The pricing page rebuilt around the four levels, with the level-3 prompt workflow written out step by step and every level linked to the store ↗ · The homepage and every vault page point at viewing a policy and buying one at the store ↗

Not done.

The two Voice Debrief use-case vaults (the web flow through a model-routing service; the WhatsApp flow through n8n), written by the agent that knows those workflows and packaged here
A page of its own for the £500 workflow, once the store's side has a return address that is not a mailbox
A use-case group on the library page, and the combination of several vaults' grants into one
The per-shape header on MAP-A-GRANT.md naming the vocabulary and the order reference
Synchronising the level-3 wording with the store agent's page

Notes. Received with D7 on 15 September. The store's four levels went live the same day (store.sgit.ai v0.1.7); the memo's four price points match them.

D9 · strategy brief · 15 September 2026

The offer is built and the button is not — each level is the level below plus one thing, and the post-sale page does not exist yet

partly

the file as received ↗ · 99e75be33f09… · 31 KB · v0.33.71

Produced. Level 1 after payment: the page a £5 buyer lands on, where the zip of the template vault for the shape bought is downloaded, with its size and sha256 stamped from the file by the build and checked in CI; the page hashes the download in the browser on request. Unlisted and noindex; the payment link's success address, with ?shape=<slug>↗ · Level 2 after payment: a working vault; a person follows up within 24 hours ↗ · Level 3 after payment: the prompt step and where to send what it produced; a person follows up within 24 hours ↗ · Level 4 after payment: the first session booked by a follow-up within 24 hours ↗ · Pricing: the plus-one-thing rule stated on the page, and an after-you-pay section with a definition of done per level, linking the four pages ↗ · The homepage: a four-level strip in the behaviour-policy section, and the £10 that was still on two pages corrected to £5 ↗

Not done.

The payment links themselves, and the success address set on each of them (level 1 passes ?shape=<slug>) — the store's, not this site's
The level-3 what-you-do text on the store's own product page, and the opinion add-on page
The A5 and the physical stand

Notes. Received in chat on 15 September. The brief's site-side items are built; its store-side items are named and left to the store. The lead's follow-up the same day set the two things the brief left open: the £5 sale lands on a download, and levels 2–4 are a follow-up within 24 hours.

I5 · chat instruction, with two public comments supplied as the first two questions · 13 September 2026

Add a section answering the questions people actually ask in public — with the standing rule that the site never names who asked. Answers that outgrow a section get their own page.

processed

chat instruction, with two public comments supplied as the first two questions — no digest; given as speech, chat text or a link

Produced. Questions — real questions, answered with a date and no name attached ↗

Not done.

A third-party screenshot convention. The instruction anticipated detailed examples and screenshots; screenshots of our own demos are fine, and screenshots of anybody else's product run into the no-probing and no-verdicts rules, so the Lab's drawn mockups remain the pattern until that is decided.

D1 · dev brief · 12 September 2026

The Grant Is User Shaped And Not Data Shaped: Start With The Connectors, And The Template Vault Is The Product

processed

the file as received ↗ · 1b2dfa45d228… · 29 KB · v0.33.70

Produced. Lab 01 — the grant is user-shaped, not data-shaped ↗ · Lab 02 — what buying a behaviour policy would look like, including the twelve-stage flow ↗ · Lab 03 — the request list against abp.sgit.ai, including the proposed material property ↗

Not done.

The five first policies themselves. Lab 01 documents the grants; no policy document exists for any of the five shapes.
The template vault and the shape library — named in the brief as the actual product, and not started.
The instrumentation table timing the first five, which the brief calls the only pricing input anybody will have.

Notes. Uploaded twice, byte-identical. The second arrival was in the same message as D3 and D4 and was not reprocessed.

D2 · dev brief · 12 September 2026

This Is The Tier One Application Nobody Could Find: The Connector List Alone Is Twenty Bits, So Compute Locally And Submit Banded

processed

the file as received ↗ · 11ff8f5f9eeb… · 31 KB · v0.33.70

Produced. Lab 04 — seven things to build, and one word we have not earned ↗

Not done.

All seven items on Lab 04's own build list. The page is the specification; none of it is built.

Notes. One correction to the brief is stated on the page and marked as ours: banding is the necessary first move and not the whole answer, because a banded submission still carries roughly the entropy of the fingerprint study the brief benchmarks against.

D3 · dev brief · 12 September 2026

The Commit Author Is A Free Text Field: A Prompt Shifts The Odds, And Every Documented Fix Was Architectural

partly

the file as received ↗ · b54eddae92bc… · 34 KB · v0.33.70

Produced. Lab 05 — your agent can commit as you, and no instruction stops it ↗ · The free/paid line on the pricing page, which now sends a reader to Lab 05's four free settings before asking for money ↗

Not done.

The comparison experiment. Lab 05 specifies it — three arms, twenty runs each, violations counted as repository queries — and nobody has set up a repository to run it against.
A signed-commits rule on this site's own repository. Lab 05 says in as many words that the argument is demonstrated and not adopted until that is on.
The provider and connector pages the brief maps onto the four layers, and the community incident repository behind them.

Notes. Read in full; the finding and the prompt are built, the experiment is not. All six load-bearing quotations were fetched and checked against their sources rather than relayed, which produced two corrections to the brief that are stated on the page: the partially-verified state additionally requires the author to have enabled vigilant mode, and the claim that the attribution renders a profile picture and link could not be found on the page cited.

D4 · strategy brief · 12 September 2026

The Urgency Is Not A Deadline But A State: You Already Connected It, And A Distributed Skill Cannot Carry A Control

partly

the file as received ↗ · af73131b6a72… · 29 KB · v0.33.70

Produced. The urgency section on the front page — three vendor sentences with dates and no adjective, and the two dated changes of this year ↗ · The free/paid line on the pricing page, with the mistakes-not-attacks label and the authorise question in place of the insurance word ↗

Not done.

The price experiment as redesigned — charge one price and count, with a certainty question after. The pricing page states the approach; no price is set and nothing is charged.
The generic prompt as a published, installable artefact. The pricing page says it is free and open; the file does not exist yet.
The early access group: a dozen people running one of the five shapes, used for objections rather than numbers. No list exists.

Notes. Read in full. Its two rulings are now on the site: the entry product says it reduces accidents and does not stop an attacker, and the behaviour policy is not sold as a skill — the skill is the free half, because the portable frontmatter cannot carry a constraint. The word for the narrowing cover does not appear on the page carrying a price; the authorise question stands in its place.

D5 · organiser document · 12 September 2026

Startup Summit 2026 — exhibitor booth guide

processed

the file as received ↗ · d13e3f08729f… · 2432 KB

Produced. The booth working page, with the guide embedded and the materials to hand over ↗ · Three corrections to the Lisbon messaging brief: no banner, the portal rather than email, and power must be requested ↗

Not done.

Logo and name into the exhibitor portal — the organisers' own deadline was 15 September.
The power request, which goes through the portal and not by email.

Notes. Third-party material, republished on a working page rather than a public one. Whether it stays fetchable under site/ is an open decision.

D6 · dev brief · 12 September 2026

Every Routable Address Is In The Grant: Do Not Attack Anyone Is The One Rule Everybody Signs, And It Is The One With No Barrier

partly

the file as received ↗ · b85fdbcb235b… · 68 KB · v0.33.70

Produced. Lab 06 — every routable address is in the grant, and the one rule everybody signs has no barrier ↗

Not done.

The two proposals to the model site — the barrier's companion fields (above whom, deterministic, what voids it, evidenceable) and the six composition rules. Lab 06 states them; they are not yet on the request list in Lab 03.
The provider comparison table, which the brief says writes itself from the sources: network on by default locally and hosted, allow list present, all egress paths or one, survives being written into a repository, fails open or closed, resolution covered, and a log the operator can read.
The reconciliation tool between an agent's several destination lists — open question 3, and possibly the smallest useful thing we could ship on this capability.
The generic pasteable document as an artefact. Lab 06 drafts it in a table; open question 4 is whether it should be published before the custom generator exists.

Notes. Seven load-bearing quotations were fetched and checked rather than relayed, including the reference container's firewall script. Two departures from the brief are stated on the page: the brief reports observing the two egress paths diverge in one session, and in ours they did not — both reached every host tried — so the finding rests on the vendor's own sentences rather than on that observation; and we published our own measurement of this machine's egress instead, which found a raw outbound socket to an arbitrary public address and working name resolution with no allow list in either path.

I1 · voice memo, transcribed in the thread · 12 September 2026

Voice memo: put the Agent Behaviour Policy at the centre of the site. The ABP is the fundamental primitive; the grant is calculated from reality via digital twins; RiskMandate drives the sale of ABPs; corporate users, investors and founders are the audiences; and a security vendor whose controls reduce the delta has a business case we can make for them.

partly

voice memo, transcribed in the thread — no digest; given as speech, chat text or a link

Produced. The direction brief ↗ · The Agent Behaviour Policy page ↗ · Pricing, rebuilt around the store's four levels and linked to it (v1.19.0)↗

Not done.

The homepage's second panel, which should become "the grant you did not enumerate" and still does not.

I3 · chat instruction · 12 September 2026

Hire a freelancer who also works through agents; give them a page and a first prompt focused on sales online and at Lisbon, and make their first task being a power user and tester of ABPs.

processed

chat instruction — no digest; given as speech, chat text or a link

Produced. Working with us ↗ · Brief B1 — power user and tester of Agent Behaviour Policies ↗

Not done.

Rate, hours, start date, escalation route, publication rules and repository access — deliberately absent because the page is public, and listed on it as owed.

I4 · chat instruction · 12 September 2026

Preserve the Lab's thinking as it changes, and publish it as PDFs that can be sent through a chat app rather than linked.

processed

chat instruction — no digest; given as speech, chat text or a link

Produced. Dated PDF editions of every Lab entry, and the whole Lab as one file ↗ · The edition register ↗

I2 · link, in the thread · 11 September 2026

The Startup Summit exhibitor pack and the event site, for a strategy document and the materials to submit.

processed

link, in the thread — no digest; given as speech, chat text or a link

Produced. The Lisbon 2026 page ↗ · The messaging and strategy briefs ↗