Do and verify the work
Atlas gathers evidence, completes the request, and checks the real result before treating the approach as proven.
Atlas helps the team find evidence, connect activity across shared systems, and prepare clear operational outputs. It does not replace creative judgment. It can act in ClickUp and build Hermes automations when the requested action and approval are clear.
Use it to locate the right source, assemble status with links, catch operational asks, or reconstruct what changed across a campaign.
Atlas is Exactius’s shared context agent for the Omaze DE partner lane. Its job is to keep operational context connected, traceable and reusable across the team.
It reads approved shared sources, links evidence to the relevant house and campaign phase, and maintains a curated context layer. That lets it answer factual workflow questions without asking everyone to retell the story.
Atlas is designed for operational clarity: finding, tracking, naming, exporting, notifying and summarising. It is not a creative director, reviewer or decision-maker.
Atlas should show where a claim came from and distinguish confirmed status from a comment, signal or inference. If a source is missing, it should say so.
Atlas is powered by the Hermes self-improvement loop. It can turn a proven process, or a correction from the team, into a reusable skill rather than relearning the same workflow in every conversation.
Atlas gathers evidence, completes the request, and checks the real result before treating the approach as proven.
It records the procedure, required sources, permissions, approval gates, verification steps and known pitfalls in an Atlas skill.
Future sessions can load the skill. When the team corrects or extends the workflow, Atlas can update the skill so the improvement persists.
A skill can be run on demand or attached to a Hermes cron or webhook. Creating the skill does not automatically schedule it or grant permission to write to external systems.
Stable facts, preferences, access details and standing boundaries.
Shared operational knowledge, decisions, source maps, campaign context and evidence.
Reusable execution playbooks describing how Atlas performs a specific workflow safely and consistently.
Approved skills connected to schedules or events through Hermes cron and webhook infrastructure.
Describe the workflow in normal language. Atlas can inspect related Cowork skills, n8n workflows and existing Atlas capabilities, propose a safe design, build it, dry-run it, and separately ask before enabling automation or consequential writes.
The current context is specific to Omaze DE Creative—not a general archive of every Exactius or Omaze project.
Request → triage → brief → production → internal review → partner approval → export/delivery → performance feedback.
Launch, BV/bonus phases, Early Close, Close, Big Swings and ongoing/evergreen work, linked by house.
Static, video, copy and concept work across Meta and secondary channels, with documented naming conventions.
Current operational hubs for H5, H6 and H7 with linked tasks, assets, stages, pending approvals, deadlines and recent mentions.
Which source answers which question—for example, ClickUp for task stage and Figma comments for static-review evidence.
Stable workflow facts, source documents, meeting operations, action-item candidates and curated learnings with supporting data.
“Connected” does not mean “everything in the system.” Atlas can only use the accounts, channels, files, projects and write permissions that are authorised and shared with it.
Has normal-user access as Atlas. It can read, create, edit, assign, move and close tasks, and add comments across the authorised workspace.
Boundary: ClickUp remains authoritative. Atlas should confirm consequential or ambiguous changes before writing, then read the task back to verify the result.
Scheduled ingestion reads four approved shared/project channels, applies a DE filter and surfaces operational requests with message evidence.
Boundary: Slack messages are signals, not canonical task status. DMs are excluded from shared-context ingestion.
Reads shared Omaze DE house-file comments with author, time and node context. H7 pages/comments are available; H6 comments are available.
Gap: H6 full-file export is blocked, and performance-file keys for earlier houses have not been supplied.
Read access to the Omaze DE project, including folders, files and authored comments. Asset paths can be linked to tasks and house hubs.
Boundary: Read-only. Comment language is evidence; formal approval logic is not assumed where the source does not prove it.
Authenticated Drive/Docs access for files shared with Atlas, including the Creative source map and H6/H7 narratives.
Gap: The Big Swing deck, copy working sheet, H7 copy matrix and Creative-call notes are not shared; the H6 matrix link is broken.
Reads meeting transcripts and filters them against the documented Creative meeting guide to capture asks, commitments and asset mentions.
Boundary: Only available transcripts and matching meetings can be used. Atlas separates operational actions from learnings.
A Git-backed, curated context vault holds source summaries, house hubs, action candidates, meeting operations and learnings with evidence links.
Boundary: It is a curated layer, not a raw mirror of every message, comment or transcript.
Can design, build and maintain agent-driven automations using Hermes skills, scheduled cron jobs and webhooks on the Atlas VPS.
Default: New automations should run in Hermes. Atlas should define the trigger, permissions, approval gates, verification and delivery clearly.
Can read Exactius n8n workflow definitions and execution history to understand existing triggers, node logic, mappings and platform dependencies.
Boundary: n8n is source material for porting. Atlas rebuilds approved workflows as Hermes-native automations rather than treating n8n as the default runtime.
Can read the private Exactius/cowork-plugins repo and use its user-created skills as source material for new Atlas capabilities and automations.
Boundary: Preserve the skill’s intent, safety rules and judgment. Port it into Atlas/Hermes rather than silently changing behavior.
Can publish approved static artifacts to isolated exactius-* Pages projects after a content and secret scan.
Boundary: Pages links are public. The protected apex and www sites are not touched.
Good requests have a house, phase, task, channel or time window. Atlas can still help when those are missing, but it will label uncertain links rather than guessing.
Ask for the ClickUp task, Figma frame, Frame.io asset, narrative or naming source for a house/phase. Atlas can return the best available links and call out missing access.
Request a house or phase roll-up, open tasks, due dates or recent stage changes. The answer should be anchored to current ClickUp evidence.
Surface assets in partner approval, candidate approval language, unresolved review exceptions and conflicting signals. Comment-derived approvals stay labelled as candidates until stronger status evidence confirms them.
Atlas scans approved shared Slack channels, matching Fireflies meetings, Figma comments and Frame.io comments for explicit asks or commitments. It checks for an existing ClickUp task before surfacing new work.
Atlas can check for duplicates, then create or update tasks, set ownership and dates, change status, and leave comments. For consequential or ambiguous changes it should show the intended mutation first, get approval, and verify the saved task afterward.
Describe a recurring or event-driven workflow and Atlas can turn it into a Hermes skill plus cron or webhook, including source access, guardrails, approval gates, verification and delivery. It can also port existing n8n workflows or Cowork skills.
Combine current source links into a concise brief, daily status, meeting pre-read, handoff or “what did I miss?” summary. Missing sources and conflicting evidence remain visible.
Preserve stable workflow context and evidence-backed learnings across weeks. Learnings follow a structured method and should not appear without supporting data.
Atlas can line up what was made, when it changed, what test was intended and what performance data was reported. It can describe associations and test evidence; it cannot prove causality unless the source and test design support it.
Atlas is operational and evidence-led. The Creative team retains creative judgment, partner judgment and accountability.
One does not replace the others. They operate at different layers and work best as a connected toolset.
| Area | Atlas | Claude Cowork | Claude Code |
|---|---|---|---|
| Primary role | Shared operational context, execution and team automation. | Personal knowledge work and connected desktop workflows. | Interactive software engineering inside repositories. |
| Context | Uses the curated Omaze DE context vault and approved live team sources. | Uses the individual's conversation, files and connected services. | Uses the active codebase, project instructions, terminal and developer-provided context. |
| Best at | Cross-system retrieval, operational actions, reusable team skills, recurring jobs and event-driven workflows. | Drafting, research, document work, analysis and ad hoc workflows around a person's workspace. | Implementing, debugging, testing, reviewing and refactoring software with a developer in the loop. |
| Skills | Team-owned Hermes playbooks with permissions, approval gates, verification and maintainable automation hooks. | Individual Cowork instructions designed around that user's connectors and environment. | Repository instructions, coding practices and engineering workflows tied to a codebase. |
| Automation | Can run approved skills through Hermes cron jobs and webhooks on durable infrastructure. | Usually user-driven and personal, unless connected to separate automation infrastructure. | Usually an interactive engineering session, not the team's always-on operational runtime. |
| Continuity | Maintains shared memory, campaign context and reusable capabilities across team interactions. | Maintains personal working context but is not automatically the team's shared operational memory. | Maintains project and session context but is not the source of live campaign or task status. |
| How they complement | Turns approved recurring workflows into shared, governed operational capabilities. | Helps individuals explore, draft and prototype useful workflows that may later be ported to Atlas. | Builds or improves the software and integration components Atlas may use. |
A Cowork skill is designed for an individual Claude/Cowork environment. An Atlas skill is a durable, team-owned Hermes capability that can use Atlas context and authorised integrations. Cowork skills are valuable source material, but Atlas ports them with explicit access, safety and verification rather than treating Cowork as the runtime.
| Capability | State | What that means |
|---|---|---|
| ClickUp status and deadline checks | Live now | Reads the DE Creative Pipeline and OM Timeline; ClickUp is the final authority. |
| Shared Slack request detection | Live now | Four approved shared/project channels are monitored with DE filtering. DMs are excluded. |
| Figma review evidence | Partial | H7 pages/comments and H6 comments are readable; H6 full export and earlier performance-file links are missing. |
| Frame.io asset and comment retrieval | Live now | Omaze DE folders, files and authored comments are readable. Atlas does not write or set approval state. |
| Google narratives and source docs | Partial | Shared Docs are readable; several decks and copy sheets are blocked or incorrectly linked. |
| Meeting action capture | Live now | Available Fireflies transcripts are scanned against the Creative meeting guide for operational asks and commitments. |
| House/phase operational memory | Live now | H5–H7 hubs connect tasks, assets, approvals, deadlines, mentions and evidence. |
| ClickUp task and comment management | Read + write | Atlas can create, edit, assign, move, close and comment on tasks. Consequential or ambiguous changes require confirmation and read-back verification. |
| Hermes automation building | Live now | Atlas can build and maintain skills, cron jobs and webhooks with explicit guardrails and verification. |
| n8n workflow inspection and porting | Live now | Reads Exactius n8n workflows and execution history, then ports approved workflows into Hermes-native runtime. |
| Cowork skill inspection and porting | Live now | Reads the private Exactius/cowork-plugins repo and ports approved skills while preserving intent and safeguards. |
| Google Doc brief write-back | Live now | Can dry-run and, after approval, copy finalized text from a working Doc into the original CRM brief while preserving structure and wireframes. |
| Curated learnings and performance context | Live now | Uses structured learnings rules and available reported data; avoids unsupported causal claims. |
| House/phase/format performance querying | Partial | The dedicated Violet comparison query path is not connected, and DE performance-file keys are incomplete. |
| Complete copy and Big Swing context | Partial | Blocked Drive items prevent complete coverage until files are shared and broken links are corrected. |
| Broader Creative source coverage | Future | Could expand when additional files, channels and formal approval signals are deliberately shared. |
Copy one, then replace the house, phase or date range. Atlas should answer with links and flag missing evidence.
“Give me the current H7 BV3 Creative status: tasks, stages, partner approvals, exceptions and the next deadlines. Link every source.”
Status roll-up“Find the source task and review links for the H7 BV2 Value Count-up video. Tell me what is confirmed versus comment-only.”
Asset retrieval“What Creative requests appeared in approved Slack channels, Fireflies, Figma or Frame.io this week that are not already covered by ClickUp?”
Request detection“Check ClickUp for duplicates, then show me the task you propose creating, including owner, due date, status and source evidence. Wait for my approval before creating it.”
Controlled task creation“Build a Hermes automation that checks the Creative Pipeline every weekday, flags overdue partner approvals, and drafts a Slack update for approval. Show me the plan and dry-run before enabling it.”
Automation building“List the relevant n8n workflows and Cowork skills for Creative operations, then recommend which one to port into Atlas first. Do not build until I approve the plan.”
Automation source discovery“Prepare a five-minute pre-read for the next Creative status meeting: changes since the last meeting, pending approvals, deadlines and blockers.”
Meeting pre-read“I was away for three days. Summarise what changed in Omaze DE Creative, grouped by house and phase, with source links and unresolved conflicts.”
Missed context“Show the available evidence connecting the latest Creative activity to reported performance. Separate observation, test result and unsupported causality.”
Performance context“Which Creative source documents are accessible for H6 and H7, and which expected files are still blocked or incorrectly linked?”
Access check