Appendix A — The Action Catalogue

This guide teaches the common journeys; the Action Catalogue is the complete, role-by-role reference for every action in Aegis — and this appendix shows you how to read it.

The chapters in this guide walk you through the tasks most people do most of the time — drafting a policy, logging an incident, handling a data-subject request — in the order you are likely to do them. They are deliberately selective. They do not list every screen, button and permission for every role, because that would make a reference, not a guide.

That exhaustive list is the Action Catalogue (a structured reference of every action in the application, recorded role by role). Where this guide says “a Contributor can upload evidence”, the catalogue is where you confirm the full set of evidence actions for every role, down to the individual permission and the exact control it lives on.

This appendix points at a document, not a screen

The Action Catalogue is not a page inside Aegis. It is a long reference document that ships with the product, so there is no screenshot to annotate here. What follows orientates you to the document’s layout and walks you through looking an answer up in it.

Where to find it

The catalogue lives in the Aegis repository at docs/USER_ACTIONS_BY_ROLE.md. It is a single Markdown file of roughly 19,400 lines, so open it in an editor or a Markdown viewer that gives you a search box and a table of contents rather than reading it top to bottom.

It is generated from the application’s own source — the route files under src/app/(app)/, the API permission checks under src/app/api/, and the role-to-permission map in src/lib/permissions/index.js — and is regenerated after large interface or permission changes rather than continuously. Always read the dated header line before relying on a detail. The current edition records seven milestones: a base generation on 1 July 2026, a delta refresh on 16 July 2026, a drift reconciliation on 21 July 2026, a preservation reconciliation on 22 August 2026 against the develop branch, a coverage reconciliation on 2 September 2026, a second drift reconciliation on 3 October 2026, and a drift fix on 6 November 2026.

Since the September reconciliation, an automated check in the product’s own test suite holds the catalogue to the code. It fails when a live route area or a registered AI action has no section, when a page that ships is not named in the file, when an entry points at a page that does not exist, or when the per-domain count table disagrees with the numbered actions in the sections below it. That keeps the catalogue’s coverage honest between regenerations; it does not check that every sentence of every entry is still current.

What’s in the document

The file runs in a fixed order, from broad to narrow. Knowing that order is most of the work of using it.

On scale, the main figures now largely agree. The header line, the count table and the detailed catalogue each give 75 domains. The count table totals 1,659 actions, while the header records roughly 1,669 rows plus 278 reconciled notes. Read them as a sense of scale and find your domain by searching, not by counting.

Looking up whether a role may do something

This is the question the catalogue answers best. Work from the domain, not from the role — domain sections carry the full detail, while the role maps are summaries.

  1. Open docs/USER_ACTIONS_BY_ROLE.md and check the dated header line at the top. If the date is older than a release you know changed the screen you care about, treat what follows as indicative and confirm in the application.
  2. Jump to the Contents index and follow the link for the domain you are asking about — Policies, Incidents, GDPR and so on. The section opens with a paragraph describing the whole area and the routes it covers. If the domain has no detailed section, as with Coach and Data Loss, search the file for its name instead.
  3. Search within that section for the action in the words you would use for it — “approve”, “export”, “delete”. Each match is a numbered entry with its own heading, so you land on a complete record rather than a fragment.
  4. Read the Roles line of that entry. It lists every role permitted to perform the action, resolved against the live permission map, and names the exact permission constant behind it.
  5. Read the Notes line before you conclude anything. This is where the catalogue records the qualifications — a licence feature that must be active, a module that must be switched on, a confirmation dialogue, or a page whose on-screen guard is stricter than the permission itself.
Holding the permission is not the same as reaching the screen

Some pages add a stricter guard than the underlying API permission. A Viewer may hold the read permission for a record and still be unable to open the page that shows it — /analytics and /ai/agents, for example, require Contributor to open at all and send a Viewer to an unauthorised page. Most other read pages, including the DORA and EU AI Act pages, now check the read permission itself, so a Viewer does reach them — apart from a few EU AI Act sub-pages (registering a new AI system, the per-system documentation wizard and the conformity assessment) that still require Contributor. The catalogue flags these cases in Notes, and it is the single most common reason a permission table and the real screen appear to disagree.

A role line describes the role, not the person

An Admin can give an individual user extra permissions through an active permission group, and both the screens and the API honour those grants. A Viewer in a group that grants policy editing sees and can use the policy edit control, even though the catalogue lists that action for Contributor and above. Some permissions can never be granted this way — settings and user management, the two personal risk mandates (accepting a risk and signing off its treatment), and full access — and pages guarded by role alone stay role-only.

How to read one entry

Every action in the detailed catalogue follows the same shape, so once you have read one you can scan the rest quickly.

Part What it tells you
Roles Which roles may perform the action, resolved against the live permission map, together with the action’s category (view, create, execute and so on) and the exact permission constant it requires. Admin holds every permission and is listed explicitly rather than implied.
Where The screen and the control — the /route plus the button, menu item, filter or form field, often with the test identifier used to target it.
Flow The full path from navigation to result, in the same shape as the numbered steps in this guide, including which request the screen makes and what the loading, empty and error states look like.
Notes Anything that gates or qualifies the action — a licence feature, a module switch, a rate limit, a confirmation dialogue, or a place where the screen and the underlying permission differ.
Expected What a correct run looks like, and what a refused one looks like — which control is hidden, and which error the request returns. This line is what makes the catalogue usable as a test script.

The four roles it maps

The catalogue resolves every action against the four-role hierarchy. Each higher role inherits everything the role below it can do.

Role What it can do
Admin Every action, plus the administration surface that no other role reaches — users, roles and permission groups, single sign-on, API keys, webhooks, licence, framework enablement, and Trust Center publication. Issuing the EU Declaration of Conformity is also reserved to Admin.
Manager Manage records across the whole compliance surface, approve policies, procedures and documents, and configure connectors and governance. A Manager edits Trust Center content but cannot switch it on or off or change its public address. No tenant administration.
Contributor The day-to-day operator: create, edit and progress policies, risks, incidents, evidence, vendors and procedures, and run AI assists. No final approval, no administration.
Viewer Read, search and open records, and export reports from data it can already see. It can manage its own account, acknowledge policies, use the free AI chat and read saved AI output, but it authors nothing and cannot run a billable AI action.

One result of that hierarchy is worth knowing, because it looks like a mistake and is not: there are twelve actions an Admin cannot perform. They are the external-collaborator and public-share actions — viewing, commenting on or reviewing a shared document as an outside magic-link collaborator, and redeeming an invitation link. Those belong to people who hold no role in Aegis at all, so no in-app role covers them.

The AI actions section

AI-assisted actions are gathered into their own section as well as appearing in their domain, because they share one pattern. The current edition records 72 registered AI actions, followed by four shared entries for what you do with a result. Each action passes the same ordered set of checks: the module must be on, the caller must hold the action’s permission and also the general permission to run an AI action, a per-user rate limit applies, the AI assist feature must be licensed, the transparency notice must have been acknowledged, and the monthly AI allowance must not be exhausted. If any one of those fails, the entry tells you exactly which response you get.

That general permission is held by Admin, Manager and Contributor, and never by Viewer. A Viewer can read every AI-backed screen, saved insight and action item, but cannot start an AI action or spend an AI credit, even where it holds the read permission for that domain; the buttons that start one are hidden from it. The Roles line of each AI entry lists only the roles that can actually run it.

The section also records what you can do with a result — copy it, run it again, save it as a record where the screen supports that, and turn it into action items. It is explicit that not every AI screen offers saving, and that creating action items is hidden from Viewer. In every case a person reads the result and decides what to do with it; no AI action changes a record on its own.

Tips and limits

Two documents, two jobs

Reach for this guide when you want to learn how to do a task. Reach for the Action Catalogue when you need to confirm, precisely, whether a particular role is allowed a particular action — for example, when deciding who to invite at which role.

Where this connects

For a plain-language explanation of the four roles, the permission matrix and how roles are assigned, see What each role can do. To change someone’s role once you have decided, see Settings. For the meaning of a term used in any catalogue entry, see Appendix B — Glossary. If the catalogue and the screen in front of you disagree and you cannot work out why, see Appendix C — Getting help.