A new Agent Cohort Settings page in the Operations console lets admins control how agent components are built for an entire cohort. Two auto-fill toggles — Auto-fill Agent Behavior and Auto-fill Tone Settings — control whether the Build Agent Components process generates those blocks for participants automatically (the default) or leaves them blank so each participant makes the selections themselves. Settings are saved per cohort, so different client programmes can run different build styles.
Each cohort can now control its own terminology. Object Terminology sets — object_name(restrict_use_of, preferred_use), defaulting to agent_object(Agent, Assistant) — drive the wording used in system reference, navigation, and buttons on the participant workbench. Auto-fill Terminology sets — (restrict_use_of, preferred_use), defaulting to (Agent, Assistant) — are enforced as mandatory rules in AI routines that auto-populate text for the cohort, such as Build Agent Components; a set with no second value restricts that word from generated content entirely. Both are managed on the Agent Cohort Settings page with sets grouped by parentheses and separated by commas.
Every new idea captured in the Claude Idea Log is automatically assigned a sequential idea number (starting at #1) at the moment it is created, via a workflow that runs on creation. The number is displayed in both the compact and detailed views and exported with the CSV, giving each idea a stable, human-readable reference for triage and coaching conversations. Existing unnumbered records are backfilled in created order.
Participants can now bestow capabilities on their assistant beyond their own skills. A new 'Assistant capability boosters' field on the participant record (comma-separated) is editable on the My Role page, placed directly after the Skills & Capabilities box with help text: 'the skill you wish you had yourself, but want to bestow upon your assistant'. Like capabilities, the boosters are passed into the Build Agent Components refinement process, which refines them into a new 'agent_capability_boosters' entry in the Agent Behavior block — positioned after capabilities and before guardrails — and included in the assembled full config script.
The Coach Console gains a Skills tab. Coaches select a participant and see each of their skills with its build status and a care & feeding status badge distinguishing AI-fresh generated content from coach-edited content. Coaches can edit the care & feeding text directly — edits mark the content coach-edited so automatic regeneration never silently overwrites them — and regenerate with an explicit confirm step that warns the coach's edits will be replaced. Regeneration requires the coach's console PIN and only works for skills belonging to the cohort's participants.
Participants can now see the Ongoing Care & Feeding section at the bottom of their skill's five-box view in the Skill Builder, showing when to reach for the skill, what it needs, and how it can fail. A new global challenge template — "Review Skill developed and review ongoing care & feeding instructions." — directs the participant to read the four care & feeding sections, note anything that doesn't match how they actually work, and review their notes with their coach before relying on the skill. The template sits in the challenge library ready to be copied to clients.
Every completed five-box skill now gets auto-generated Care & Feeding guidance: a trigger condition (when to reach for the skill versus the general agent), an input contract (what it needs handed to it), known failure modes drawn from the House Rules, and update ownership. Generation fires automatically the moment a skill's five boxes become complete, is grounded strictly in the five-box content (never inventing capabilities or failure modes), and writes a HAIA-SKILL-CARE-FEEDING-GENERATED observation event to the activity feed so downstream assembly can verify it happened. Content is stored on the skill with a status of not generated, auto-generated, coach edited, or stale.
ClaudeIdeaLog v1.0 is now available with the MCP Connector. Capture ideas, observations, gaps, and questions in the moment with a triage status for later review and actioning.
File packaging is now centralized through a dedicated FileManifestService backend function. When a participant runs Pack-and-Go from My Files, the manifest content, manifest file name, and zip file name are generated server-side from a single source of truth rather than assembled ad-hoc on the client. A FileManifest entity records each packaging request (manifest id, app request surface, participant/cohort/client context, package contents, format, and chosen file name), giving coaches and admins visibility into every bundle built. This paves the way for consistent packaging across future surfaces and auditable delivery.
Participants can now bundle selected files from My Files into a single downloadable zip via Pack-and-Go. The mode activates when a challenge with the PACK-AND-GO tracking code is started. Files can be individually selected, given a package file name, and grouped into folders by tag or challenge. Pre-packed/locked rows selected by the coach or system are included automatically, and Protected files that are locked and not selected are blocked from download. After download, packed files are locked (pack_lock + pack_me) and visually dimmed in the list.
Participants can now move files to Trash from My Files instead of deleting them outright. Trashed files are hidden from the active list and surface on a dedicated Trash page in the Public Workbench sidebar, where they can be searched and restored back to My Files. Each restore writes a trash_recover timestamp for audit. Pack-and-Go locking and Protected-file download restrictions remain enforced on the active list.
Whiteboards now start faster and look better. Admin-established Starter Templates provide pre-built canvas layouts — boxes, headers, prompts and File Away groups included — so participants can begin from a proven structure instead of a blank board, and admins can save any whiteboard as a reusable template. A new background selector lets participants choose from curated visual backdrops (or save their own), applied behind the canvas, while the dotted default remains available. Also in this release: whiteboard backups (auto-snapshot on session close with restore), AI-generated whiteboard narratives, and File Away groups for tidying the canvas.
Coaches can now assign tasks, skills, challenges, and references directly from the Coach Console Portal. Per-participant assignment opens the Task Manager from any participant row (with file attachments and an in-app chat notification to the participant), while the new cohort-level 'Add Task' button opens the Task Manager in cohort mode — visually distinguished with a complementary indigo background — assigning the item to the entire cohort in one step. Portal rows were also updated to show participant headshots and no longer display email addresses.
Coaches can now step directly into any participant's Workbench with one click using the 'Act As' button on each Portal row (also available from the Seating Chart). The impersonation view renders the participant's full workbench exactly as they see it — Start Here, Whiteboard, Skills, Chat and all other panels — with a persistent identity bar showing who is being viewed and a dedicated exit to return to the console. Ideal for live session walkthroughs and troubleshooting a participant's setup in real time.
Act As now includes an Agent Assist toggle. While impersonating a participant, coaches can toggle Agent Assist On in the impersonation bar to open a floating chat window beside the participant's workbench. The AI agent is preloaded with that participant's context — goals, assigned tasks and challenges, completion statuses, and recent chat with the coach — so the coach can review progress, plan next steps, and draft coaching approaches with the agent without leaving the session.
Deleting an assigned task from the Coach Console during a live cohort session (cohort status in_progress) no longer removes it permanently. Instead, the task and its attached resources are marked for deletion and hidden from participants immediately, while the underlying records are retained so the live session, progress tracking and reports stay stable. When no session is in progress, deletion works as before — a full, permanent removal. This prevents mid-session surprises such as vanishing completions or broken agenda references when a coach retracts a task on the fly.
Admins can now save any whiteboard as a reusable template and apply it to start a new board instantly. A new Background selector lets you render a custom image behind the canvas — choose from admin-provided template backdrops or upload your own (with a "Share with all" option for admins).
Boxes and text objects on the Automation Whiteboard now capture action-specific attributes inline. When you assign a workbox action to an object, the Object Inspector surfaces the prompts tied to that action as editable fields, so you can fill in exactly what each box needs (source, trigger, owner, fallback, etc.) without leaving the canvas. Why it matters: it turns the whiteboard into a guided skill blueprint. Each prompt you answer maps directly to a slot in the five-box skill build, so a completed whiteboard is already most of the way to a finished skill spec. Quick path to completing the skill blueprint: 1. Lay out your boxes and arrows to sketch the process. 2. Assign each box its workbox action (get_from_here, process_here, put_here, etc.). 3. Work through the prompted attributes that appear in the Object Inspector for each object. 4. Run Check Completeness to see what's still missing. 5. Generate Overview to produce the narrative and confirm the end-to-end flow. A whiteboard with every attribute filled and a green completeness score is ready to hand off as a skill blueprint.
Adds 3P (Protect, Produce, Prove) impact analysis for automation goals — each goal is scored against the three pillars with a stacked bar chart, a qualifying 12/15 threshold line, and per-pillar improvement recommendations so you can see which goals are investment-ready.
Derives an automation solution directly from your whiteboard diagram through a completeness check — the 'Check Completeness' action evaluates your diagram against the five-box skill structure, surfaces what to draw next, and flags missing location and process notes so you can turn a sketched process into a buildable skill.
Participants can now upload, organize, and manage their own files directly from the Public Workbench under a new 'My Files' section. This acts as a support role for building skills and agent projects — giving participants a dedicated, participant-owned space to store working documents, reference files, and build assets alongside their five-box skill work. Files are participant-scoped (secured by ownership) and sorted by newest, oldest, name, size, or category. Note: files are temporary and are cleared at the end of each session, so participants should download anything they want to keep.
Coaches can now access and manage participant whiteboards directly from the Coach Console. The new Whiteboards tab shows a participant selector and the full whiteboard editor — letting coaches view, create, rename, and edit any participant's whiteboards without leaving the console.
Added an 'Export PNG' button to the Whiteboard toolbar so participants can download their workflow diagrams as an image file for their records.
Participants can now interact with the status control on challenges and tasks in their journey panel to advance progress themselves — clicking 'Start challenge' moves an item to In Progress and 'Mark complete' moves it to In Review, with the progress ring and status badge updating live.
The Coach Console Agenda tab now includes an Agenda Pace Meter — a live timer with Start/Pause/Reset controls that tracks actual session time against the agenda's projected durations. A sweeping marker glides across a proportional ribbon of agenda sections, segments fill as they're covered, and a red overtime zone grows past the projected end when the coach runs long, with a digital readout showing actual vs projected and remaining/overtime status.
When a cohort is in progress, the backend functions that automatically update a participant's role description, skills/capabilities, and goal 3P evaluations now post a friendly, AI-assisted message into the participant's chat. Both the coach and the participant can see these messages, so each automated update is surfaced in the conversation as it happens, with a short note on what was just generated.
Standardized British-spelling display labels to US English across the public Workbench, Ops console, Coach console, Foundations preview, and interview review surfaces. User-facing instances of 'Programme' now read 'Program', and 'Organisation' now reads 'Organization'. Code identifiers and non-displayed prompt text were intentionally left unchanged.
Generated job descriptions, skills/capabilities, and automations (goals) are now produced automatically via entity triggers on Add and Update — no manual generation step required. All three are created after the coach completes the initial onboarding review, so role content and goals are assembled from the reviewed interview data without the coach needing to click a generate button for each.
Goals across both the admin Participant Goals view and the public Workbench now include an automated 3P Principles (Protect, Produce, Prove) scorecard. When a goal is created or its name/baseline changes, the evaluateGoal3P function classifies it against the 3P framework — scoring each pillar out of 5, assigning primary and secondary P classifications, and drafting a buyer-vocabulary pitch for the matched persona (CFO, HR Leader, or Department Head). Admins can fully edit the checklist, classification, pitch, and prove metric, then approve the review. Participants on the public Workbench can refine the pitch paragraph in their own words without altering the admin classification. Each card shows a running total score out of 15.
Added a new 'Front Door URLs' section to the Operations Documentation page covering every participant- and coach-facing entry-point URL format: the participant workbench (/c/<cohort>/p/<participant>), the standalone AI interview (/c/<cohort>/p/<participant>/interview), the coach console (/c/<cohort>/coach), participant auth (/access), and the Shields Up suffix (/s/<frequency>). Includes host resolution guidance for standard application subdomains vs. vanity subdomains.
Sticky notes can now be enabled per front door access point instead of a single global toggle. In Settings \u2192 Sticky Notes, each Front Door (Coach, Backend, Workbench, Ops, Onboarding, etc.) has its own toggle, so you can allow visitors to place sticky notes on some doors while keeping them off on others. Each sticky note now also captures which front door it was composed from, shown on the review card so admins can see exactly where feedback originated.
Cohorts now capture a Location (city or venue) and an optional 'N in series' number. The cohort name auto-generates from the Service Package + Location + Programme Date + Series (e.g. 'Foundations Chicago Aug 18 2026 S2') as you fill those fields — still fully editable. The cohorts table also gained a dedicated Location column so facilitators can see where each cohort is being delivered at a glance.
The app now exposes a Model Context Protocol (MCP) server at /api/mcp so AI assistants (Claude, ChatGPT, Cursor) can access and operate the platform on a user's behalf. Connections are authorised via the app's own OAuth sign-in plus a per-tool credential: participants use cohort_code + participant_code, coaches use cohort_code + coach_pin, and operations managers use the MCP_OPS secret. Custom tools expose the participant, coach, and ops functions, and entities plus the onboarding interview agent are auto-exposed under the signed-in user's row-level security. The server goes live when the app is published; individual entity and agent tools can be toggled on the app's MCP page.
Coaches can now enable a 'Shields Up' mode from the Front Door Controls in Settings. When active, participants are required to enter a unique, rotating cohort key in addition to their access link before entering the workbench. Coaches can generate, copy, and refresh keys on the fly. We recommend generating a fresh key at the start of each session and sharing it verbally with your participants to ensure only the right people are in the room.
The AI-assisted chat panels have been upgraded to act as more effective coaching partners. When a coach opens a chat with a participant, the AI is now automatically pre-briefed with essential context from the participant's profile and baseline interview — including job role, reporting line, role context, current AI usage habits, real-work task notes, and friction points. The AI now provides much more personalized, relevant coaching recommendations without needing you to manually re-explain the participant's situation or history.
When a cohort's status transitions to 'In Progress', a workflow now automatically initializes ChallengeCompletion records for every participant in that cohort. Each participant gets completion tracking entries (status: not_started) for all assigned challenges and tasks, so progress is tracked from the moment the cohort goes live. The system also resets existing records to not_started for a clean start.
The Journey Panel progress ring on the Public Workbench now calculates as the average of each challenge's status-based progress percentage (from the status lookup), rather than only counting fully-verified items. This gives participants a more accurate view of where they stand across all assigned challenges and tasks.
The onboarding interview agent now creates an ActivityEvent audit checkpoint (tracking key HAIA-FOUND-ONBOARD-AUDIT) after each core topic completes and when the interview closes. This provides a safety net that preserves conversation progress even if a ParticipantBaseline write silently fails, ensuring no interview data is lost.
A new Workflow Designer has been added to the Tool Belt section. Designers can build, edit, and export automation workflows using a horizontal flowchart canvas with step types including Start, Task, Decision, Notification, Activity Observer, and End. Each step supports metadata such as assignee role, SLA, detection category (for Activity Observer steps sourced from Settings lookups), tracking events, and transition conditions (advance, hold, backstep). Workflows can be searched, saved, and exported as JSON or Markdown.
The Participant Agent Builder is a facilitator-facing tool that lets admins select a cohort and participant, preview the source data (participant profile, client context, and matched agent role), and trigger the backend buildParticipantAgentComponents pipeline to generate optimized agent components. The pipeline combines the participant's role, client context, skills, and matched AgentRole enhancements using AI, then stores the assembled build (agent declaration, behavior, instructions, and tone settings) as a versioned ParticipantBuildComponents record. Existing builds are listed per participant for reference and re-generation.
A new testing-annotation system lets users place colorful sticky notes anywhere on the page. A floating action button (FAB) in the bottom-right corner enters placement mode — click anywhere to drop a note at that exact position, type a message, optionally add your name, pick a color, and post. Notes persist to the database, render at the same position for everyone, and sync in real time via subscriptions. Admins can delete notes directly on-page, and manage all notes (inline edit, set status new/approved/fixed, delete) from Settings → Sticky Notes, which also includes the global on/off toggle and a delete-all action.
A new Compliance Requirement field (multi-select) has been added to the Client entity, sourced from the compliance_types Lookup list (Public Company, SOC 2, ISO 27001). This replaces the legacy binary public_company flag, which has been deprecated in favour of the more flexible compliance_requirement array. Origin: Following a meeting with Freespace, it was highlighted that ISO 27001 and SOC 2 compliance tracking is critically important for clients. This change introduces a structured, extensible way to capture and manage those compliance requirements per client.
Added an extra security layer for the Coach Console. Each coach now has a Coach Console PIN stored on their coach record. When a coach opens the console via the cohort URL, they are prompted to enter their PIN before the roster and tools are revealed. The PIN is validated server-side against active coaches for that cohort's client and is never returned to the browser. Admins can set the Console PIN from the Coaches management screen.
New Coach entity and full management page under Client Management. Admins can create, edit, and delete coaches with support for both client-side and internal (Happy AI) coach types. Each coach record tracks bio, headshot URL, LinkedIn URL, coach type, client linkage, and active status. Includes a coach_type lookup list managed in Settings > Lookups, with search and type filtering on the coaches page.
The Seating Chart now supports configurable room orientation (front at top or bottom), a three-zone layout (left, center, right), and four table types — Training, Conference, Round, and Remote. Includes state-based editing (view vs edit mode), auto-arrange, clear seating, and drag-and-drop participant assignment across tables and zones. Configuration persists per cohort via AppSetting.
Participants can now manage their headshot URL directly from the Workbench via a new 'My Headshot' section under My Profile. The headshot is dynamically rendered in the topbar avatar throughout the workbench experience. Primary reason for headshots is for coaches to have a visual queue on their console.
Fixed misaligned columns in the Settings > Lookups manager for Enhancement Category and Status lists. Refactored grid template definitions to be explicit per list type, added flex-wrap for edit-mode Update Access buttons, and used minmax(0,1fr) for flexible columns to prevent content overflow from expanding columns.
The onboarding interview agent now captures a verbatim, timestamped transcript of the participant's actual responses alongside the synthesized topic notes. Each response is tagged by topic (role context, current AI use, real work tasks, friction points) so coaches can compare the raw conversation against the summarized notes. This gives coaches and reviewers a reliable source-of-truth to check that reflections didn't drift from what was actually said, and helps catch details that may not have made it into the synthesized fields.