5. Teams and Roles: create your own roles and teams

Besides the teams shipped with the app, you can create your own: standalone roles, kept in a catalog to reuse them, and complete teams, with their roles, their flow and their documents. Everything happens in Teams and Roles, in the Home sidebar, with an agent that interviews you and writes with you. In short:

1. On the Roles tab, create or import the roles you need (or let the agent write them with you).
2. On the AI Teams tab, create a team (empty or as a copy of an app team), add its roles, define its flow (which can be conditional) and write its documents.
3. Click ▶ Try the team to see it work in a test project, and export it as a .tacteam to share it.

It needs a license with the roles and teams features (see Licenses).

The roles and teams gallery

A single screen with a shared header and three tabs: AI Teams (AI Teams tab; the one you see when you arrive), Roles (Roles tab) and Catalog (The downloadable catalog). The header has, from left to right, the title Teams and Roles, with “Browse, create and edit roles and AI teams.” below it; the tabs, each with its counter (how many teams or roles there are in total, ignoring filters); the Search by name box, a single one for both tabs (the text is kept when you switch tabs); Import ▾, with the entries of the visible tab, and the main button, which changes with the tab: + New team or + New role. Switch tabs with a click or with the ← and → keys (Home and End go to the first and the last); each tab keeps its position and whatever you had open. Notices appear under the header on a single line, with ✕ to close them: green with ✓ for something done (for example “Lyra saved. Version 1.0.0 → 1.1.0.”), blue for imported roles and red for errors. It is a gallery: it lists the roles and the teams with their actions. Opening a role’s detail or one of its documents fills the whole screen (without the header) until you go back to the list (← Roles). The All teams → link in Projects opens this screen at the AI Teams tab. In a narrow window, the header wraps onto several rows and the cards go from 4 columns down to 3, 2 or 1.

The dedicated screen of a role or a team

When you create a role or a team, or choose to edit it (or click the name of an app team, or View team on its expanded card), the window leaves Home and its side menu and shows a full screen dedicated to that role or team, just like opening a project (the window takes its working size and, while it changes, shows “Opening…”):

• At the top, the bar of the project screen: ← Teams and Roles, the type (Role or Team), the name, the version (“v1.2.0 → v1.2.1” when there are unsaved changes: the version it will have when you save) and the status with a coloured dot: Ready (green), The agent is writing…, Unsaved changes (orange), Saving…, Saved or Changes discarded (for a moment, after saving or resetting), Error (red) or Read-only (grey). For a team, also its validation (✔ Complete, ⚠ N warnings or ✖ N errors; clicking it opens the Validation tab) and, if the catalog has newer versions of its roles, “N roles with a new version” (opens the Roles tab). On the right, the editor’s agent and model, the agent’s session, the app language and the theme. New window is the first button on the left, before Back (opens another app window on the home screen; this one stays as it is, with whatever you are editing; without a license that includes multiple windows it shows a 🔒).
• Below, the role or team on the left and the chat with the agent on the right, with a divider you can drag (or move with the arrow keys). In a narrow window, one half at a time, with the Role / Team and Chat tabs.
• Save, Cancel and Reset. On the right of the bar, before the language and the theme, the Reset · Cancel · Save group, always visible (in a narrow window, in its own row under the bar). No change is saved until you press Save, whatever it is: a role’s text, metadata and variants; a team’s data, roles, flow and documents; and whatever the chat agent applies or writes. While there are unsaved changes, the bar says Unsaved changes and Save is enabled; with none, it is off (“No changes to save”). The content is compared: if you type something and delete it, nothing is left pending.
  • Save (or ⌘S / Ctrl+S, also with the cursor in the chat or in the text) saves everything at once and keeps the editor open; the bar says “Saved” for a moment. If it can’t be saved, you see “Couldn’t save: …” and the changes stay pending. It is the only thing that raises the version (see Version).
  • Reset asks you to confirm (“Discard all unsaved changes and go back to the last saved version?” · Reset / Keep my changes) and leaves everything as it was at the last Save (or when you opened it, if you haven’t saved yet), without closing the editor. Chat cards already applied turn into “↩ Undone” and the chat notes it with a line (“Unsaved changes were discarded; …”). The conversation with the agent is not deleted.
  • Cancel with no changes closes the editor; with changes it asks you to confirm (“You have unsaved changes. Discard them and close?” · Discard and close / Keep editing) and goes back to the gallery without saving anything. If the role or team is new and is empty without what was discarded, it then asks Keep it or Delete it.
  • While the agent is writing, Save and Reset are off (“Wait for the agent to finish”); Cancel stays on and its confirmation warns that the agent will be stopped.
  • Changing step, view or document inside the editor doesn’t ask: the changes stay pending. If the app closes abruptly, whatever you didn’t save is lost.
• ← Teams and Roles takes you back to the gallery, in the same tab and at the same place in the list, with the role’s or team’s card highlighted for a moment and, if the version went up, the notice “… saved. Version … → …” (with the team’s name; for a role, “Saved. Version … → …”). It always warns you if something is pending, in a strip under the bar: if the agent is writing, Stop and leave or Keep waiting; with unsaved changes, Save and leave, Discard and leave or Keep editing (which just closes the strip); if the role or team you just created is still empty, Keep it or Delete it. With the agent writing and unsaved changes, Stop and leave stops the agent and then asks what to do with the changes. With nothing pending, it goes back without asking. The same happens if you close the window, open a project (⌘O / Ctrl+O) or quit the app (⌘Q) with a dedicated screen open. In the strips, Esc works as Keep editing.

The role catalog (Roles tab)

The app’s role catalog: here you keep standalone roles (their document and their data) to reuse them, without opening a team. It is not the same as the Roles panel of an open project (the one for models per role). The catalog starts empty.

Header (the shared one): with this tab, the main button is + New role and Import ▾ offers A .tacrole file… and From a team…. Below, the area filter as a row of round chips (All and one per area of the catalog’s roles, with a dot in its color; the selected one has a dark border and bold text) and, on the right, the language filter as a segmented selector (All languages, Spanish, English; one is always selected) and how many roles are shown. With many areas, the chips wrap onto another line. If nothing matches: “No role matches.”.

Empty catalog: a centered box with a dashed border, a “+”, “There are no roles in the catalog yet”, “Roles are the pieces teams are built from. Create one with the agent or import it.” and the buttons + New role, From a .tacrole file and From a team.

Role card: like a person’s card, centered: a large round avatar with the role’s initials, in its area’s color and with a ring of the same color; below it the name and the area (in its color); the summary (up to three lines); a line with the version (for example “v1.2.0”), the origin (Own or App), the languages as two colored chips, always ES (warm) and EN (blue) in that order (a language the role doesn’t have looks faded, struck through and with a dashed border; hover to see, for example, “Spanish — not available”) and the number of variants if it has more than one (hover to see their names). Below, a teams tag: “Not in any team yet” or, if your teams carry it, the icons of those teams (by category; up to 3 and a “+N”), and on hover which teams they are and at which version. In the footer, Edit (own roles only), Details and the ··· menu with Duplicate, Export and Delete (for an app role, Duplicate and Export). Details expands the card to the full width, as in Projects, without opening any editor: the area, version and origin (for example “Development · v1.0.0 · Shipped with the app”), the full summary, the Variants (if it has more than one), Used in with each of your teams that carries it and its version (or “No own team uses it”), the date of the last change and the View the full role link, which opens the role’s page (like clicking its name); Less collapses it again. Only one card is expanded at a time: expanding another collapses the previous one. A role folder the app cannot read shows as “Could not read “folder”“, with the reason and Delete.

Viewing a role: the formatted document, with the language selector (only the existing ones) and, if the role has variants, the Variant selector (you see the common text plus that variant’s). Beside it, the summary, area, suggested code, the modification date, the origin (“Written in the editor”, “Copy of … v…”, “Taken from the team … v…”, “Imported from a package”, “Shipped with the app”) and the teams it is in use in. Below, the actions: Edit, Duplicate, Export, Delete and, if it was updated from a package, Restore the previous version (v…).

Editing the document and metadata: Edit opens the role’s editor (see The role editor with the agent, below), where you change the document in the Definition view (View or Text, with a live check in Checks: the document needs a title (# …), an intro paragraph and the nine sections in this order: Description, Area, Identity, Main responsibility, Skills required for the job, What it can do, What it cannot do, Quality of its work and Artifacts it produces; in Spanish, their Spanish names) and the metadata in Metadata. There are no separate “Edit document” or “Edit metadata” buttons any more.

Metadata view: the name, area and summary in each of the role’s languages; the code suggested to teams; the variants (add, remove, name them; the standard one cannot be removed; inside the document, a variant’s text goes between <!-- variant: id --> and <!-- /variant -->), and the version: changing the metadata does not raise it, unless you tick Set the version by hand and type another one (x.y.z format; it warns you if it is lower than the current one). It has no Save button of its own: what you type stays pending until the editor’s Save. If a field is not valid (for example, an empty name), it is marked and you can’t go on until you fix it.

New role: opens a dialog with a “+”, the title New role and “Start with the basics. The agent will help you define the rest in the editor.” (× closes it, like Cancel and Esc). It asks for the Name (required); below it you see the Identifier derived from the name, with Edit to change it by hand (if its format is invalid or it already exists, the field opens by itself with the warning). The Area (required) is picked from chips: Product, Analysis, Design, Architecture, Implementation, Quality, Operations, the areas already in the catalog and Other, which opens a field to type it. Then the Initial language (Español or English; “You can translate it later.”) and Which teams it is for (optional: only the agent uses it, to tailor its questions; it isn’t saved in the role). In the footer, “Next: 9 steps with the agent”, Cancel and Create and open editor →, dimmed until there is a name, a valid identifier and an area. It creates the role at version 1.0.0, with the title and the nine sections empty, and opens its dedicated screen with the agent.

The role editor with the agent

Edit opens the role’s dedicated screen (see The dedicated screen of a role or a team), in three columns: the steps panel on the left, the document in the middle and the chat on the right (you can resize the chat by dragging the divider, or with the arrow keys; it starts at 430 px). At the top, the role bar: New window (opens another app window; without a license that includes multiple windows it shows a 🔒), ‹ Teams and Roles, the role’s name with its badges and, below it, the save status (“Unsaved changes”, “Saving…”, “Saved”, “Changes discarded”, “The agent is writing…” or “Read-only”) and the version (“v1.2.0”, or “v1.2.0 → v1.2.1” if it will go up when you save); on the right, the document language (a small ES | EN pair, like the app’s: the chosen one is in the accent colour and the language the role doesn’t have yet is dimmed and struck through, and can’t be picked; hovering shows “Translate to …”), which is repeated inside the document area, to the right of the file path (it is the same selector: picking a language in one picks it in the other), Translate to English / Translate to Spanish (or Translate again to … if the other language already exists), the agent’s session, the app language and the theme, with the Reset · Cancel · Save group in front (see The dedicated screen of a role or a team). What you type stays pending until Save (⌘S / Ctrl+S); it no longer saves itself.

• On the left, the steps panel shows how far the role’s definition has got: “Role definition”, “N of 9” with a bar (it counts closed sections) and the nine sections of the language you are viewing, numbered: green ✓ closed (with This is good, or because it already had content when the editor opened or arrived with a translation or a variant), blue with its number and in bold the current one (the first not closed, the one to work on now), amber circle written but not closed (it has content but hasn’t been closed), empty circle empty, and red “!” if the heading is missing. What is closed lasts for the session: it isn’t saved in the role, and when you open the editor again everything with content counts as closed. Click a step to go to its section: it is marked (blue background) and the document scrolls to its card (in Text, the cursor goes to that section’s heading). Picking a step doesn’t change the current section or send anything to the agent; if you are in another view, it goes back to Definition and takes the document to that step. At the bottom, below a line, four views that open the same way, one at a time, in the middle (the chosen one is marked with a blue background): Definition (the document; it’s the one you see when the editor opens), Metadata (the usual form, with the version, which can be set by hand), Variants (each with how many blocks it has in the visible language, Add variant and Remove) and Checks, with a badge showing how many checks are pending (amber; red if there is any error). Switching view loses nothing: the document, the View/Text choice and the conversation are kept. The panel updates itself when each agent turn ends, when you edit by hand, when you apply a proposal and when you switch language.
• In the middle, the chosen view. In Definition, the document: at the top, the View / Text toggle (the left and right arrow keys switch options), the file path (for example en/full-stack-developer.md) and, when needed, an amber notice: “N empty sections” (empty or without a heading) or, if there are none but there are checks, “N checks” (clicking it opens the Checks view). With View (selected when the editor opens) you see the role’s title and introduction and one card per section, with its number (01…09), its title, a Done, In progress, Not closed, Pending or Missing heading tag, and its text (or “This will be filled in when you reach this point in the chat” if it’s empty); the current card has a soft blue background and the one picked in the panel, a blue border. Anything in the document outside the nine sections is shown as is, without a card. With Text you see the Markdown, which you can edit by hand: it stays pending until Save, and the view reflects it when you go back to it. Before each message to the agent, your hand edits reach what the agent sees, even though they are still unsaved. While the agent works, the document is read-only and, when it finishes, shows what it wrote. Metadata, Variants and Checks take the whole middle, under their title. Checks shows “Checks: N errors, M warnings” or “All good.” and the list: those of the document’s structure and those of the variants (a variant the document uses and the role doesn’t declare, or an unclosed marker, is an error; a declared variant without any block, a warning), with each one’s language if the role has several. Errors inform but don’t prevent closing.
• On the right, the chat with the agent. At the top, “Assistant” with the agent and model button (“Claude Code · Default ▾”), the “Chat in EN” badge and New conversation (deletes that role’s conversation after confirming; the document doesn’t change; the chosen agent and model stay). Choosing the agent: the button opens a small dropdown; at the top, Agent, with Claude Code and Codex (one that isn’t installed shows as “… (not installed)” and can’t be chosen). Switching agents starts a new conversation: if there already is one, the chat asks you to confirm (“Switching to {agent} starts a new conversation: the current one will be deleted. The document doesn’t change.” · Switch to {agent} / Cancel). The chosen agent applies to all roles (and, in the team editor, to all teams), and each agent remembers its own model. You can’t switch while the agent is working. Choosing the model: below, Model, just like a project’s settings: Default (the agent’s own; “The agent’s default model.” says so), with Claude Code Fable, Opus, Sonnet and Haiku, and Custom… to type a model id or alias; below it, “using {model}” with the one the agent reported. The change applies to your next message, not to the reply in progress. The choice is remembered when you close the editor and restart the app: one for all roles and another for all teams. If your plan doesn’t include the chosen model, the projects’ notice appears with Use {another} and Choose another…; after switching, send your message again. The agent talks in the role’s main language (the first of its languages), even if the app is in another language or you are editing the other language’s version; “Chat in ES/EN” reminds you. The app’s notices and buttons stay in the app’s language. The agent interviews you (in a new role it starts) and asks about what is red in the checks.
  • Questions. Its open questions are numbered (1., 2.…) and you can reply by number (“1: …; 2: …”). Closed questions (yes or no, choosing among 2–4 options) appear as an “Agent question” card in the chat, with the same options as in projects: pick and click Answer, or click Answer in the chat (or just type in the box) and what you write is the answer; the card then shows your answer, “Answered in the chat” or, if the turn was interrupted, “No answer”. The agent guides you from start to finish: every reply ends with the next step (the current section and, with all nine closed, the summary, a variant, the translation, and review and close).
  • Drafts. Each section is worked on as an improvement cycle: you describe what you want and the agent shows you a draft of the section in the chat (“Draft · {section} (EN)” card, with the text in a scrollable box). The document on the left doesn’t change until you approve it: with the Apply button or by typing in the chat that it’s fine (“looks good”, “apply it”…). If you ask for changes, the agent replies with a new draft and the previous one shows “Replaced by a newer draft”. If the section already had text, the card warns that it replaces it. Once applied (“✓ Applied”), the agent asks “Do you want me to review and improve it?”; if you say yes, it proposes an improved draft, as many times as you like. If you click This is good with a draft not applied, the point closes with what the document already had and the draft shows “Not applied”.
  • Online research: every time you ask it to propose a section, the agent first searches the web for how this role is defined in the market today (the chat shows a “Web search” line with the query), says so in its message and the card lists the sources under Based on (up to 5, with links that open in the browser). If it couldn’t research, the card says so and the draft is based on the agent’s own knowledge. A change you ask for on a draft that was already researched doesn’t repeat the search. Only the role editor’s agent can search the web.
  • Installable skills (“Skills” section). When proposing the Skills section, the agent also searches the skill marketplaces of Claude Code (plugins and skills.sh) and of Codex, and proposes only the ones that fit the role’s description, area and responsibility already written (if it finds few or none, it says so). Its message says where it searched. The draft card shows, below the competencies, the Installable skills subsection, with one row per skill: the name, the label of the agent it exists for (Claude Code, Codex or Claude Code · Codex), one sentence on what it brings to this role, the Source (marketplace and address, which opens in the browser) and, if there is one, the Equivalent for the other agent. Once applied, the role document shows the same subsection (in Spanish, Skills instalables). The role is not tied to any agent: the agent only appears in those rows. If the marketplaces couldn’t be reached, the card says “Couldn’t reach the marketplaces: the proposal only includes the competencies.” and the section is proposed without the subsection. An older role without the subsection looks the same as always; proposing Skills again adds it.
  • Closing a point. The point is only closed with the This is good button, which appears above the shortcuts, in green, next to the line “Step N of 9 · {section}”, once the current section has content (disabled while the agent works); writing “ok” or “let’s go on” in the chat doesn’t close it. While the point is open, the agent talks only about that section: it doesn’t propose, ask about or write anything of the following ones until you click This is good (if you explicitly ask for another section, it does handle it). When it asks you to choose among several options (for example, which bullets to keep or remove), it gives them as a numbered list and you reply with the numbers in the chat box (“1, 3”, “all”), without typing the lines. When you click it, the chat shows Point "…" closed. Moving on to the next one. and the agent asks about the next section (after the ninth, “… Moving on to the next step.” and it goes on with the summary). A closed section isn’t reopened from the interface; if you ask in the chat to change it, the agent changes it. If you ask for something else, it does it and then asks whether to go on with what came next; if you tell it you’re done, it doesn’t insist.
  • Variants and summary. If it proposes a variant or a summary, a card with a button appears (Add variant “…”, Use this summary (EN)); clicking it applies it, it turns into “Applied” and the agent moves on by itself (the chat shows the line “Summary applied (EN). Moving on to the next step.”).
  • Shortcuts and sending. Above the message box, three shortcuts: Suggest {current section} (for example, Suggest Description: the agent researches and shows you the draft of that section; the name changes when the current section changes and, with all nine closed, it becomes Suggest summary), Review the role and Continue (asks the agent to pick up the guidance again). The agent’s messages have no bubble and yours do. With the nine sections closed, instead of the step you see what comes next (“Next: …” or “All set: you can close the editor.”) with the optional suggestions (a variant, the translation). The ↑ button sends (Enter too) and Stop halts the agent while it works; if an agent question is waiting, you can type and send the answer even while the agent is working.
  • The main button (Sprint 100). At any moment only one button stands out, filled with colour: the one to press now. With the current section empty, the Suggest {section} shortcut stands out. Once it has content, the step line adds “· ✓ Has content”, This is good stands out (it closes the section and moves on to the next one) and the shortcut becomes discreet, named Suggest {section} again. With the nine sections closed, “✓ All nine sections are closed” shows above the next-step line; Suggest summary stands out while a summary is missing, and then Review the role if checks are pending (the summary shortcut becomes Suggest summary again). When everything is ready, no button stands out. It’s worked out from the content, also when you open an existing role, and nothing is saved. While the agent works, the highlighted button looks dimmed and can’t be pressed. In a narrow window, the highlighted shortcut stays visible and the Shortcuts menu holds the rest.
• Translate: the agent creates the other language’s document from the one you are viewing; when it finishes, the editor switches to that language and shows the strip “Translated by the agent. Review it.”. Translate again asks you to confirm before replacing the other language’s text.
• A single agent: the one of the role’s saved conversation; if there is none, the one you chose in the assistant’s dropdown; and if you haven’t chosen one, the one that is installed or, with both installed, the one you are signed in to (signed in to both, or unknown, Claude Code). So if you only use Codex, the editor works with Codex even if Claude Code is installed without an account. It works in the role’s folder, with no project open, and can only write there (never the metadata: if it touched them, the app puts them back). If it isn’t available (not installed, not signed in or failing), the chat says so in a red box with Try again and you can keep editing by hand.
• One conversation per role: when you open that role’s editor again, the conversation is still there.
• Role ready: when the nine sections are closed, the summary exists in every language, there are no pending checks, the agent isn’t working and nothing is unsaved, a green band with ✓ appears at the top of the editor: Role ready (“Every step is complete and saved. You can finish or keep editing.”). Its buttons: Finish and back to the catalog (leaves like ← Teams and Roles), View the role (leaves the same way and opens the role’s page in the catalog) and Keep editing (hides the band; the editor stays open). If you change something or the agent works again, the band goes away and comes back when everything is ready again; it also shows when you open a role that is already complete. It doesn’t appear in read-only mode.
• Version: it goes up when you press Save, if the document changed: the last number, or the middle one if “What it can do”, “What it cannot do” or “Artifacts it produces” change. Everything changed since you opened the editor counts, so saving several times doesn’t raise it further (1.2.0 → you save a small edit: 1.2.1 → you also save a change in “What it can do”: 1.3.0, not 1.3.1). If you set the version by hand, it is kept. Without changes, it doesn’t go up; Cancel and Reset don’t touch it.
• Leaving (← Teams and Roles): if you saved something, the gallery says so: “Saved. Version 1.2.0 → 1.2.1.”. If the agent is still working, it asks Stop and leave or Keep waiting; with unsaved changes, Save and leave, Discard and leave or Keep editing; if a role created in the editor is still empty, Keep it or Delete it. If the app closes abruptly, whatever you didn’t save is discarded: when you open the role again, it is as it was at the last Save.
• In a narrow window, the editor shows one part at a time, with the Role and Chat tabs (a dot on “Chat” flags a new reply). On the “Role” tab, the steps panel is folded above the document (“Role definition · N of 9”; unfolded, the steps and the Definition, Metadata, Variants and Checks views; picking a view folds it) and the role bar takes two lines; the This is good button is on the “Chat” tab (below 640 px it takes the full width).

Duplicating, deleting, exporting and importing roles

Duplicate: opens the Duplicate role dialog (with a “⧉”, “Creates a copy of “…”. You can change it afterwards with Edit.” and ×, Cancel and Esc to close it). It asks for just the Name of the copy; below it you see the Identifier derived from it. If the name is empty it says “Enter a name.”; if a role with that name exists, “A role with that name already exists.”; if it has no letters or digits, “The name must include letters or digits.” (Duplicate is disabled until it is valid). Confirming creates a new own role, at version 1.0.0, with the name you typed in every language and everything else (document in each language, metadata, variants and checks) equal to the original, and remembers where it came from. App roles can be duplicated too. The copy’s detail shows “Copy created. Want to polish it with the agent?” with Open with the agent and Not now (nothing is lost if you decline).

Delete: asks for confirmation and lists the own teams that carry it: they keep their copy, but they stop receiving updates. App roles cannot be deleted.

Export: saves the role to a .tacrole file (“id-version.tacrole”) to take it to another installation. It does not need any license feature.

Importing a .tacrole file: you choose the file and the app checks it before importing it: name, id, version, languages, variants, origin and integrity (SHA-256). A damaged or modified file is rejected with the reason. If the role is new, Import; if you already have it in an older version, Update (with the list of changes), and the previous version is kept so you can restore it; if you already have that same version, Reinstall; if yours is newer, Import anyway; if an app role uses the id, it is imported with another id (“id-2”).

Importing from a team: you choose an app team, one of Your teams (with its language, ES or EN) or a project folder, tick the roles you want to bring and, for each one, check what the app deduced (name, code, area and id) and write a one-line summary, which is required. If a role with that id already exists, you choose Import as id-2 or Skip. The document is copied as it is, in the team’s language. When it finishes, the app offers to open each imported role with the agent (“Open … with the agent”, for example to review the summary) or Not now.

App roles: the app can ship its own signed roles, marked App: you can view, duplicate and export them, but not edit or delete them (duplicate it to change it). The app ships eight: Business Analyst, Documentador, Full Stack Developer, Product Owner, Project Manager, QA, Software Architect and UI/UX Designer. Without a license you can view them and add them to your teams; duplicating or editing them needs a license, as with any other role. If one was modified outside the app (its signature does not match), its card and detail say “This app role was modified: its signature does not match. It is blocked: it can’t be added to a team or used. To fix it, reinstall the app.” In the team editor’s Roles step its checkbox is disabled with “⚠ App role modified: blocked. Reinstall the app to fix it.”, and a team’s copy can’t be updated from it either. A team that already carried it keeps its copy and can remove it.

License: creating a new role, importing and duplicating need Create roles; editing the document or the metadata, opening the editor with the agent and restoring the previous version, Modify roles; deleting, Delete roles. Without the feature, the button shows a 🔒. Viewing, searching and exporting are always allowed.

Roles are stored in the app’s data folder, in roles/<id>/ (a role.json and the document for each language); each role’s editor conversation, separately, in roles/.editor/ (it doesn’t travel in the .tacrole).

Your teams (AI Teams tab)

This is where teams are managed, in two groups: Your teams (“Created, imported or copied. You can edit them.”) and, below, the App teams (“They update with the app. Copy them to modify them.”; signed, only to view and copy them). With this tab, the header’s main button is + New team and Import ▾ offers A .tacteam file… or A folder…. Below the header, a row with three controls: the category selector (All, Asteroid, Nebula, Supernova, Galaxy, Moon and Comet, which groups your own teams; each with its icon), the language one (All languages, Spanish, English) and, on the right, the number of teams shown. The three filters and the search box work together and are not saved: when you reopen the app they go back to All. If a category has no teams: “There are no teams in this category.”. Under the row, the group titles YOUR TEAMS and APP TEAMS in capitals, each with its description; a group with no matching team is hidden. Cards go up to 4 per row (3, 2 or 1 in a narrow window).

Team card: its top is tinted with its category color (yours, in violet, like Comet): the category with its icon on the left and the version in the right corner; the name in large type (up to two lines) and its roles’ faces (a circle per role with its initials, in its area’s color; up to 5, or 4 and a “+N” face if there are more; hover to see their names) with the number of roles. Below, the description (up to three lines) and the languages as ES and EN colored chips (the one the team doesn’t have, faded, struck through and with a dashed border), followed by “modified … ago” when known. An own team with errors has a red border and the red line Validation found N errors. followed by Review, which opens the team on its Validation tab; one with only warnings, the grey line Validation found N warnings.; one with neither, no line. At the bottom, three controls: the main action (Edit on yours, Copy on the app’s), Details and the ··· menu: on yours, Rename, Copy, Export, Restore the previous version (if there is one), Show in folder and Delete; on the app’s, Export. Rename opens the Rename team dialog with the current name: it changes only the name in the app’s language (if the team doesn’t have it, the one in the language the card shows, as the dialog says); the names in other languages, the identifier, the folder and the version don’t change. It doesn’t accept an empty name (Enter a name.) or the name of another of your teams, ignoring case (A team with that name already exists.); with the same name it closes without changes, and Cancel, × or Esc close it without changing anything. When you confirm, the card shows the new name, is highlighted and the notice Team renamed to “…”. appears. The new name also shows in the projects that use the team and on its screen. It needs the same license feature as editing a team. Edit (or clicking the name) opens the team’s dedicated screen; for an app team, the name opens it read-only. Details no longer opens the team: it expands the card to the full width, as in Projects, with the category, version and number of roles (for example “Supernova · v1.2 · 4 roles”), the full summary, Recommended for, the team’s quick commands as tags and, for an app team, the View team link; Less collapses it again. Only one card is expanded at a time. An app team your license doesn’t allow shows 🔒 next to its name.

• New team: a dialog that looks like New role: at the top, a “+” icon, the title, “Choose where to start and name it. You will add its roles afterwards.” and × to close. Starting point as chips, in this order: Empty (selected by default; “For those who already know what team they want: it starts without roles or flow.”), Copy of an app team (below, “Recommended to start: a coherent team in a minute, which you then adapt.”; it loses the signature, becomes your own and the id {id}-custom is proposed, with the notice “The copy loses its signature and becomes your own team.”) and Copy of one of my teams (only if you have any); when you choose a copy, the drop-down with the team appears below. If you open the dialog with Copy on a team’s card, that copy is already selected. Then Name, with the proposed Identifier on a line below and Edit to change it (if it is taken or has an invalid format, the field opens by itself with the notice); Main language (Español / English), which starts with the app’s language (when copying, the app’s language if the team has it; otherwise the team’s), and Summary (one line), marked Optional, with the help “You can fill it in later, when you edit the team.”: with Empty it starts blank and when copying the source team’s summary is proposed. If you go back to Empty after choosing a copy, the summary is cleared and the language returns to the app’s. At the bottom, “Next: the team opens so you can add its roles”, Cancel and Create. An empty team starts at version 0.1.0, with no roles (and therefore with an error) and with an empty flow document (a warning) until you add them.
• Describe the project and let AI propose the team (fourth chip of New team, the last one, with the Recommended pill; with a license without “Create teams” it shows 🔒): for people who don’t know about roles or processes. When you choose it, below it, “Not sure where to start? Say in one line what the team is for and AI proposes its roles and flow. You review it before it’s created.”, and the form changes: What kind of projects is the team for? (required, up to 400 characters; for example “small websites, like landing pages”); If you like, fine-tune the proposal (Optional), with three questions as chips that start at I don’t know / let AI decide: Who tests?, Do you want to approve each phase or only the result? and Is there visual design?; Name, here Optional (“Leave it empty to use the name AI proposes.”); Main language (“AI proposes the team in this language.”), and the notice “Uses {agent} and your account’s tokens; takes a couple of minutes.”. No summary is asked: the proposal brings it. Get the proposal → first checks (“Checking the agent…”) that the team editor’s agent —the chosen one or, if none, the one that is signed in— is installed and signed in; if not, it says why and offers Use {other} for this proposal only, or, if neither is available, “No agent is available…” with Check again. With the agent available, the dialog widens and shows “AI is preparing your team”, the progress bar, “Agent: … · model”, the elapsed time and what it is doing (for example “Searching the web: …”). The agent may search the web, but writes nothing to disk. Cancel (or Esc) stops it without creating anything and goes back to the form with what you wrote. If it takes more than 6 minutes, or its answer can’t be used (for example “it uses a catalog role that doesn’t exist”), it says so, with Show details, Try again and Cancel. Then comes the Proposed team preview (“Take your time: nothing is created until you press “Create the team”.“): the name, the summary, “English · N roles · N steps · ID: …”; Who’s on the team, each role with what it does and the label From catalog or ✦ New; How it works, with the flow diagram, and When it will ask you, one line per approval (“After “Interview” — …“). If an app team already fits, a notice says so with Copy that team, which goes back to the form with Copy of an app team and that team selected. Try another approach asks for one line (“What do you want to change?”) and brings a new proposal that replaces the previous one (“Proposal 2”, “Proposal 3”… pill), with no limit. Cancel closes without creating anything. Create the team builds the complete team, without going through the editor. First it asks for confirmation: Build the complete team?, with what will be written (“AI will draft it, and the app will write and check:” the new roles with their nine sections, the flow document, one card for each role, the gates and procedures, the deliverable templates and the checks, with up to two automatic fixes), the notice “It takes several minutes and uses tokens from your {agent} account.” with “Agent: … · model”, and Back (returns to the same proposal without creating anything) and Build. If the proposal brings new roles and your license doesn’t include “Create roles”, Build shows 🔒 and a note explains it. While building, the dialog shows Building “…” with a progress bar, the agent and its model, the elapsed time and the stages in order: “New roles · 1 of 2”, “Flow document”, “Role cards · 3 of 5”, “Gates and procedures”, “Deliverable templates”, “Checks” and, if needed, “Fixing · round 1 of 2”. The agent drafts and the app writes and checks each part; the agent writes nothing to disk and doesn’t search the web, and each call has a 6-minute maximum. Continue in background (or × or Esc: closing never cancels) closes the dialog and you can keep using the app: in Teams and Roles the sidebar shows “✦” and the AI Teams tab a “Building “…”” strip with the stage and the time, See progress (back to the dialog) and Cancel. Meanwhile the team doesn’t appear in the list and only one team is built at a time: the AI chip of New team says “Building…” and leads to the progress (the other starting points work as usual). Cancel asks for confirmation (“Everything built so far is undone and the tokens used can’t be recovered.”): with Cancel the build the agent stops, the team and the new roles of that attempt are deleted and you go back to the proposal with the notice “Build canceled. Nothing was created…”. If the agent fails or takes more than 6 minutes on a stage, everything is undone too and it says so (“”…” couldn’t be built”), with Show details and Try again, which goes back to the same proposal. If all goes well, it ends in Team ready (“You can now use it in a project.”): the name, the tagline, “English · N roles · N steps · ID: …”, the roles (the new ones with ✦ AI), “✓ Checks: no errors”, how long it took and, if any, AI’s fixes; Done closes and leaves the team highlighted in the library, and Adjust in the editor opens it at Checks, where the assistant continues with the step that is really missing. If something still fails after two automatic fixes, it ends in Some things still need fixing, with up to five problems in plain language (for example “The template “report.md” is missing.“) and Adjust in the editor (the team is kept) or Discard the team, which asks for confirmation and deletes the team and the new roles (the catalog roles you already had don’t change). If you weren’t watching, you find out from the strip in AI Teams (green if it’s ready, amber if things still need fixing, red if it couldn’t be built; View opens the result and × hides it), from the Teams and Roles badge (”✓” or “!”) and, if the app isn’t in the foreground, from a system notification that opens the result when you click it. If you close the app while it builds, it asks first (“A team is being built”): Cancel and close cancels and undoes what was built; Keep building doesn’t close. If the app was closed abruptly, what was left halfway is deleted when you open it again. Roles created this way carry the ✦ AI pill on their catalog card and ✦ Created by AI on their screen, with their nine sections written in the team’s language; they are edited like any other role and the mark stays (a duplicate no longer carries it). The catalog roles the team reuses don’t change. It works the same with Claude Code and Codex, and each call to the agent (for the proposal and the build) is recorded in the team editor’s usage log.

The team editor: the eight steps

Team’s dedicated screen (see The dedicated screen of a role or a team). When you create a team it opens by itself, on the Identity step. It has three columns, like the role editor: on the left the step menu, in the middle the view of the chosen step and on the right the chat with the agent.

• Guided tour with “Next”. At the foot of each step’s view, a bar with the Next button takes you through the steps in this order: Identity → Roles → Flow → Flow document → Sheets per role → Gates and procedures → Templates. In Templates the button is called Review checks and opens the checks (which stay outside the tour and have no bar). In Roles, while no role is selected, Next is disabled and on its left you read “Select at least one role to continue.”; while the agent is writing too, with “The assistant is writing…”. When you press it, the menu highlights the new step and the panel shows it at the same time, without asking: whatever is unsaved in Identity or in a document stays pending until Save. The panel follows the chat: when the agent finishes a turn and the chat’s “Next: …” line points to another step (Identity, Roles, Flow or a pending step; not an error, the warnings or “All set”), the menu and the panel move to that step on their own, with a brief highlight in the menu and without taking the cursor away from the chat. It doesn’t if you changed step yourself during that turn, and without roles it doesn’t go past Roles. Unsaved changes are kept. The step menu stays open: you can jump to any step whenever you want. The bar doesn’t show in the app’s teams (read only).

• The main button of each step (Sprint 100). In each step only one button stands out, filled with colour: the one to press now. While the step isn’t complete, the chat shortcut Suggest {step} (or Guide roles) stands out and Next is discreet. When the step is complete, an indicator shows on the left of the bar (“✓ Identity defined”, “✓ Roles chosen”, “✓ Flow defined”, “✓ Flow document written”, “✓ Role sheets written”, “✓ Gates and procedures written” or “✓ Templates written”), Next (or Review checks) stands out and the shortcut stays there, discreet, as Suggest {step} again (or Guide roles again). A step counts as complete when it’s green in the step menu: the app decides it from the content, also with unsaved changes and when you open an existing team, and nothing is saved; if you delete the content, the indicator goes away and the suggestion stands out again. In Checks, Review the team stands out while there are errors or warnings. While the agent is writing, the highlighted button looks dimmed and can’t be pressed.

• Step menu (“Team definition”, “N of 8”), with a progress bar and the eight steps of the definition, each with its state (number, ✓ done, ! with errors; in bold the one that is started): Identity, Roles, Flow, Flow document, Sheets per role, Gates and procedures, Templates and Checks (with the badge of how many there are). Clicking one shows its view. While the team has no roles, Flow carries the amber pending badge (you can open it anyway). The state is computed by the app from the team as it is: a step is done when what belongs to it exists and has no errors (roles, when there is at least one; the flow, when it has steps without problems; the sheets, when all are written; the procedures, when the four recommended ones exist; the identity, when it has name, summary, tagline and what it is recommended for). In a narrow window the menu is folded above the content.

• Identity: name, tagline, summary and what it is recommended for, per language; the chat quick buttons (one per line: each is a button in the project chat that sends that text; they should match the system commands of the flow document); the version.

• Roles: at the top, “Select the existing roles that will be part of this team.” with the “N selected” counter, the search box and the list of your existing roles, each with its checkbox, badges, area and languages, the variant selector (if it has more than one) and the “⚠ Only in …” warning if it doesn’t exist in the team’s language; the selected ones come first. Checking a box adds the role to the team (pending until Save, like every other change); unchecking it asks for confirmation (“Remove the role “…” from the team?” · Remove / Cancel). If a flow step uses the role, its checkbox is disabled and says “Used by the steps: …. Change them in Flow to be able to remove it.”. No roles are created here: if you have none, the list says “There are no roles in the catalog yet.” and you create them in Teams and Roles → Roles. Below, the table of the team’s roles (change the code, remove; the origin “Catalog: role vX”). When you add a role, the app creates its participation sheet (team/flow/{code}.md) with the activations table and one row for each flow step that is already its own; when you remove it, the sheet is deleted; when you change its code, it is renamed. If the catalog has a newer version of a role, its row shows New version x.y.z and the Update button: a dialog explains that the copy of the role in the team will be replaced and that a copy of the previous one is kept (Update / Cancel; the team’s previous version can be restored from its card). It only affects the team: projects already created don’t change. If any role has installable skills that only exist for one agent, the informative notice ℹ Some skills only exist for one agent appears above the table, with one line per role (“QA — 2 skill(s) only for Codex.”). It doesn’t stop you creating or using the team: when you start a project you choose which to install and see which aren’t compatible with the agent.

• Flow, the team’s flow, editable (the same one you see in a project’s panel): the flow’s name and description, “N phases · M steps”, the Edit / View as in the project switch, the Diagram / List tabs (in both modes; it opens on List and remembers the last one chosen while the panel stays open) and the line Initial role: X (step “…”, cycle start) (or “first step” when no cycle start is declared). A flow can be conditional: a step with a gate declares several exits (forward, step back, same step or skip ahead), and the answer to the gate picks the path (see Gate exits).

  • Phases. One card per phase, with its name and description (edit them by clicking on them), the Starts in a new chat checkbox and the ⋯ menu (Move up, Move down and Remove the phase, only when it is empty).
  • Steps. Each step has its number, the role (only the team’s), its text, the Optional and Gate checkboxes (checking Gate adds “(Gn)” to the step text, with the next free number, and the ## Gn — … section in gates.md, which is created if missing; unchecking removes the “(Gn)”), its coverage in the sheet (✓ In the sheet when its role’s sheet has a row with the step id; otherwise ○ No row in the sheet with the Add row link, which writes it), Description and the ⋯ menu (Move up, Move down, Move to “phase”, Every cycle starts here and Remove the step); the step where every cycle starts carries the Cycle start label.
  • Reordering. To reorder, drag a step by its ⋮⋮ handle (also to another phase) or use Move up and Move down. + Step at the end of each phase and + Phase at the end of the flow.
  • Changes and sheets. Each change shows right away and stays pending until the screen’s Save, and the sheets follow it: a new step gets its row in its role’s sheet (with the step id and text and the rest “to define”), a step that changes role takes its row along, and a removed step loses its row if it was unfilled. If a change can’t be applied, everything stays as it was and it tells you why.
  • Flow problems. At the top, the flow problems (a step with a role that isn’t in the team, marked “Role does not exist”; a repeated id; a cycle start that doesn’t exist; a gate exit that goes to a step that no longer exists), with Go to (for a broken exit, it takes you to its Goes to list).
  • Diagram. The Diagram tab (also in “View as in the project”) shows the flow drawn instead of the list; the header, the initial role and the warnings stay on top: the phases top to bottom (with the New chat chip when they open a new chat) and, inside, one node per step with its number, role and text, joined by arrows that never run over a text or a node (the arrows between steps go down the column of numbers and each phase title starts to its right; the way back to the cycle start goes along the right margin); an optional step has a dashed border, a step with a gate carries the Gn badge, a step whose role does not exist is red with “⚠”, and a dotted arrow goes back from the last step to the cycle start (which has a colored outline). Click a node (or press Enter) to switch to the List tab and go to that step; Go to in the warnings also goes back to the list. If the flow has no steps, the tab says “Add steps to see the diagram.”. The diagram is read-only, updates with every change and follows the chosen ES / EN language; a legend explains the symbols shown.
  • Gate exits. A step with a gate can declare the paths out of it (for example, in Lyra, “All good” goes to “Living documents” and “There is a problem” goes back to the interview). On List, under each step with Gate checked, the Gate exits block has one row per exit: its text, the Goes to list with every step (the gate itself shows as “(this step)”), a badge with its kind (forward, step back, same step or skip ahead) and Remove; + Add exit adds one (“New exit”, to the next step) and a gate takes up to 6. A gate needs at least two exits: with none, the block warns in amber “⚠ This gate has no exits. Add at least two so the flow says what happens when it is resolved.”; with one, “⚠ This gate has only one exit. Add at least one more.” While any gate has fewer than two, the Flow step doesn’t count as complete (no “✓ Flow defined” and Next doesn’t stand out), Checks shows a warning per gate (“The gate G2 (step “…”) has no exits: add at least two…” or “…has only one exit: add at least one more.”), whose Go to takes you to the step with the focus on + Add exit, and the chat offers to ask the agent for them (see below). It is a warning, not an error: it doesn’t stop you from using or exporting the team. If the team has two languages, the text is edited in the chosen language (ES / EN) and the target, adding and removing apply to both. If the step an exit goes to is removed, the exit is kept and warns in red (“Step “…” no longer exists. Pick another target.“) and in the flow problems. If you uncheck Gate, the exits stay in the file and come back when you check it again. On Diagram, a gate with two or more exits grows downwards with one row per exit, its text inside the node: the one that goes to the next step has “↓” and uses the usual arrow; the ones that go back to an earlier step (or to the same gate) leave on the right with a dotted arrow that goes up the margin to the target step; a skip ahead uses a solid arrow, and a broken exit is red with “⚠” and no arrow. Hovering an arrow shows ““text”: from “gate” to “target”“. The legend adds Path out of a gate and Step back from a gate. A gate with fewer than two exits carries a “⚠” mark next to its “Gn” badge (on hover, “no exits” or “only one exit”) and the legend adds Gate with fewer than two exits. Read-only (View as in the project and the app’s teams), the list shows one line per exit, “text → n · step”, or the same amber warning if the gate has fewer than two. The exits are saved in the flow file, in each step’s exits field.
  • The delegated role can decide this. Each step carries, next to Optional and Gate, the checkbox The delegated role can decide this, checked by default. Clear it to reserve the step: with a delegation active, the delegated role does not decide it for you and the team asks you there (Semi-autonomous mode, chapter 9). It is saved in the flow file ("delegable": false on the step; checking it again removes the field) and applies to both languages. With no reserved step nothing changes: with a delegation, the role decides everything. With at least one, on Diagram every node carries a badge on the right, 🤝 Delegable (thin outline) or 🔒 Reserved (thicker accent outline), the legend explains them (“Delegable: the delegated role can decide it”, “Reserved: waits for the person”) and the node’s accessible name says it. Read-only, a reserved step shows the 🔒 Reserved badge. Lyra comes with “Ready to build” (G2) and “User test” (G3) reserved.
  • Acceptance criteria. Below Gate exits, each step with Gate ticked has the Acceptance criteria block, with an “n/8” counter: up to 8 sentences a reviewer checks before you decide, each verifiable by reading the latest deliverable (for example, “Every criterion of the request is covered by the design”). + Add criterion adds a row (“New criterion”) and moves the focus to its text; Remove deletes it; with 8, + Add criterion is disabled (“A gate takes up to 8 criteria”). With no criteria, the block says “No criteria: this gate works as usual.”. Below, Reviewer role picks which team role reviews (its role document and sheet) or Project agent (no role), a neutral reviewer (the default). If the reviewer role no longer exists, the selector marks it “⚠ {code} — does not exist” and warns in red “The role “{code}” no longer exists. Pick another reviewer.”. With two languages, the text is edited in the chosen language and criteria are matched by position; adding, removing and the reviewer role apply to both. If you untick Gate, the criteria stay in the file and come back when you tick it again. In read-only views you see the list and “Reviewer: …”. They are saved in the workflow file, in the step’s criteria and reviewer fields; no team that comes with the app brings criteria by default.
  • Languages. If the team has several languages, the ES / EN selector picks the one whose texts you edit; structural changes (steps, roles, optional, new chat, cycle start, gate, rows) apply to all of them. A team without a flow shows “This team has no flow yet.” with Create empty flow. View as in the project draws the flow with the look of a project’s panel, with the cycle start highlighted.

• Flow document, Sheets per role, Gates and procedures and Templates: each view lists the documents of that kind with a help line and, from the start, the complete shape of the team: the flow document, one sheet per role, the four common procedures (gates.md, elicitation.md, amendments.md, severity.md), the templates cited by the sheets or the flow document and the changelog (CHANGELOG.md). Each document shows its state: ✖ Missing (red, required by validation), Recommended (grey, a procedure or the changelog that doesn’t exist yet), Empty (only headings; in a sheet, nothing written beyond the rows the app created) or Written. Clicking one opens it in the same panel to edit it by hand, as text or formatted, with no buttons of its own: what you type stays pending until the screen’s Save, and going back to the list or changing language doesn’t ask. Each document also has ✦ Write with the agent (if missing or empty) or ✦ Review with the agent (if written), which send the matching request to the chat.

• Checks (the errors and warnings, with Go to). Besides the previous ones, validation warns (without blocking) about: a flow document or a sheet with no content; a step cited by the “Step id” table of the flow document that the flow doesn’t have; a “(Gn)” on a step with no section in gates.md; and a command listed in the “System commands” block of the flow document that isn’t among the quick buttons.

The team editor’s chat: drafts and proposals

• The chat with the agent, on the right, as in the role editor: a single agent (Claude Code or Codex) working in the team’s folder and talking in the team’s main language (the language badge in the header says so). The agent and its model are chosen with the header button, as in the role editor (switching agents starts a new conversation); the teams’ choice is separate from the roles’ one. It interviews you about how the team works together (who starts, what the user approves, what the team does not do, what happens when a test fails) with numbered open questions (you can answer by number) and closed questions with buttons (a card with the options, Answer and Answer in the chat), and on every turn it receives the team state: the validation, the steps with no row in any sheet, the gates without enough exits, the documents that are missing or empty and the pending draft, so it asks about what is red. It writes the team’s documents: procedures, templates and the changelog directly; the flow document and the sheets it proposes as drafts in the chat (see below). Above the message box, the next-step line (“Next: …”, in the order of the tour —identity first, then the roles and the flow—, or “All set: review the team and leave the editor”) and the Suggest {step}, Review the team and Continue shortcuts. The first one changes with the step you have open: Suggest Identity, Guide roles, Suggest Flow, Suggest Flow document, Suggest Sheets per role, Suggest Gates and procedures and Suggest Templates (not in Checks). In Flow, while any gate has fewer than two exits, the shortcut becomes Propose gate exits: it asks the agent for the missing exits, each with its text and its action (go back, move on or close); once none are missing, it goes back to Suggest Flow. It asks the agent for a draft of that step that takes into account what you already defined in the previous ones (the identity, the chosen roles, the flow): the identity and the flow come as proposals with buttons, the documents as drafts (see below), and nothing is saved until you apply it. Guide roles only explains what each available role brings and which part would still be missing: it doesn’t propose roles, you select them; at the top, New conversation. In a new, empty team the agent starts the interview. While it writes, the open document is read-only with “The agent is writing…”. It doesn’t touch the configuration or the flow: if it did, the app puts them back and says so in the chat.
• Team ready: when roles, flow and documents are done, with no validation errors or warnings, the agent isn’t working and nothing is unsaved, a green band with ✓ appears at the top: Team ready, with Finish and back to the catalog (leaves like ← Teams and Roles), ▶ Try the team (the same button as in the header, with its confirmation) and Keep editing (hides the band). It goes away if something changes and comes back when everything is ready again; it also shows when you open a complete team. It doesn’t appear for the app’s teams (read-only).
• Document drafts: when the agent proposes a part of the flow document or of a sheet, it appears as a Draft · path › section card with the proposed text and the Apply and Dismiss buttons (or write “it is fine” in the chat). On Apply, the app writes that whole section (or the document, when the draft names no section) into the file, the card turns into “✓ Applied”, the document’s view opens and the agent goes on with the next step. A newer draft of the same part leaves the previous one as “Replaced by a newer draft”. While a draft is pending, the agent doesn’t write that document.
• Proposals with buttons: the agent doesn’t change roles or the flow itself: it proposes the change with a card under its message, for example “Add the role QA (Standard) to the team” or “Insert the step “Review” (role Team Lead) after “Implementation”“. The agent can also propose the identity texts (“Tagline (EN): “…”“, one per field and language: applying it saves it in Identity, where you can keep editing it), new phases (“Add the phase “…” at the end of the flow”, with or without a new chat), the cycle start, the gates of a step and the exits of each gate, so after the interview it can propose the whole flow in a single message. The Exits of the gate “…” card lists 2 to 6 exits, one per row: its text, where it leads (“continues at “…”“, “back to “…”” or “ends the cycle at “…”“, which is the last step of the flow) and its action (forward, step back, skip ahead or closes); a target that doesn’t exist yet shows as “(step not created yet)”. If the gate already had exits, the card warns “Replaces the N current exits.” If, when you apply it, the step isn’t a gate yet or a target doesn’t exist, the card says so (“…apply the proposal that marks it first.” or “…apply the proposals that create that step first.”) and you can apply it again later. The app understands the exits line even when the agent doesn’t write it in the exact format, as long as there is no doubt: the gate by its number (“G1”) or its name, the target by the name of the step (ignoring case and accents) or by an action (“next”, the following step; “back”, the previous one; “close”, the end), with quotes or bold. If it doesn’t understand it (the gate doesn’t exist or isn’t a gate, an exit without a target or with “ · “ in its text, fewer than two or more than six exits, a target that is no step or matches several), instead of the card there is a warning with an amber border, “⚠ A gate exits line was not understood”, with the line quoted and the reason; the app sends it back to the agent on its own so it can fix it (“I sent it back to the agent so it can fix it.”), at most twice in a row per message of yours; if it can’t, the warning says “Tell the agent so it can fix it.” and the agent gets it on its next turn. Each card has Apply and Dismiss; with two or more applicable proposals in the same message there is also Apply all N proposals, which applies them in order (first the phase, then its steps). On Apply, the card turns into “✓ Applied”, the change is highlighted for a moment on the left (and its view opens, unless you are editing something) and the agent learns about it on its next turn. What you describe in the chat still rules: you can ask for changes in your own words and the agent replies with new proposals, or edit it yourself in the Flow tab. Removing a role asks for confirmation and isn’t applied if any step uses it. If the proposal can no longer be applied (the role is already there, the step doesn’t exist), the card says why and only offers Dismiss.

Trying, versioning, exporting and deleting a team

• Try the team: in the header, the ▶ Try the team button (enabled when the team has no validation errors; needs Create projects and Custom teams) creates a test project in the app data folder (“Test {team} {date}”, outside the team list) with the team as it is saved, closes the editor and opens the project in this window, to see how the initial role starts and how the flow is followed. It asks for confirmation first. With unsaved changes it is off (“Save your changes to try the team”).
• App teams: their screen is read-only (Data, Roles and Flow, without controls) and, instead of the chat, a box “App teams are not edited. Copy it to edit your copy.” with Copy to my teams, which goes back to the gallery with the copy dialog open.
• Version: it goes up on its own when you press Save, according to everything changed since you opened the screen (the middle number if the roles or the flow steps changed; the last one if only texts or documents changed), only once even if you save several times; or you set it with Set the version by hand in Identity. Cancel and Reset don’t touch it. Update on a role keeps the team’s previous version when you press Save.
• Validation: a team with errors is saved, but it can’t be exported or picked for a new project.
• Export: Team only or With its catalog roles (it also carries the roles that come from the catalog, to add them to the catalog of whoever imports it). When importing such a package, the Package roles section lets you choose Add, Update or Skip for each role.
• Delete asks for confirmation; projects already created with the team do not change.
• License: creating and copying need Create teams; editing (and with it the chat, the flow, applying proposals and updating a role) and restoring, Edit teams; deleting, Delete teams; importing, Custom teams (and, to add the package roles, Create roles). Viewing and exporting are always allowed. Without Edit teams, an own team’s screen opens read-only.

The downloadable catalog (Catalog tab)

The third tab of Teams and Roles, Catalog, lists the teams and roles published on teamaiconsole.com and installs them without looking for or downloading files by hand. The list is requested the first time you open the tab in each session (not when the app starts); while it arrives, you see the last one downloaded.

• What you see: a Teams | Roles selector and the same cards, filters and languages as in the gallery: for Teams, the category; for Roles, the area; for both, the language and the header search box (“Search the catalog”). Each card says whether you already have it: ✓ Installed · v1.2.0, or ↑ New version v1.3.0 · you have v1.2.0 when the catalog has a higher version than yours. Details shows the summary, who it is recommended for and its roles (or its variants).
• Install and Update: the app downloads the package, checks that it is exactly the one the catalog publishes and opens the same review as a manual import (Your own teams (import), Duplicating, deleting, exporting and importing roles): you see what it brings and confirm with Import (or Update). A team ends up in Your teams and a role in the Roles tab, as if you had imported it from a file; if the team brings roles you already have, the review asks Add, Update or Skip. When it finishes, See in Your teams or See in Roles takes you to it. Cancel changes nothing. One thing is installed at a time.
• Security: the list is signed and each package carries its fingerprint (sha256). If the list’s signature is not valid or a package doesn’t match its fingerprint, nothing is installed and a red notice explains why (“The “…” package doesn’t match the sha256 published in the catalog. Nothing was installed.”). A new list with a tampered signature doesn’t replace the good one: you keep seeing the previous one with the notice “The new list could not be verified. Showing the list from {date}.”.
• Offline: you see the last list downloaded, with the notice “Offline, list from {date}”, and Install is disabled until you are back online (Try again requests it again). If none was ever downloaded, the tab says “The catalog could not be loaded” with Try again.
• License: viewing the catalog is always possible. Installing a team needs Custom teams and a role Create roles, the same as importing them from a file: without them, Install shows the 🔒 lock and opens the usual notice.

What you create is not published from the app: your teams and roles stay on your computer until you export them.