Workflows

Paths with owners — and a stop before irreversible.

Four core flows. Each names who does what, and where a human approves before send, spend, or go-live.

Workflow 1

Intake & dispatch

Agents: Atlas (owner); specialists as dispatched. Trigger: New ask, priority change, or work that spans more than one specialist. Artifact: Brief with success criteria and a named owner.
  1. 1

    Atlas takes intake and clarifies the outcome.

  2. 2

    Work is broken into a brief with success criteria.

  3. 3

    The right specialist (or sequence) is dispatched.

  4. 4

    Status returns to Atlas; you get a clear update when it matters.

Approval gate: Nothing irreversible until a human approves.

Workflow 2

Research → draft → review

Agents: Scout → Scribe → Sentinel (Atlas routes). Trigger: You need facts turned into language you can send or publish. Artifact: Cited research + draft (and short alt when useful) + risk notes.
  1. 1

    Scout digs, cites sources, and flags gaps or stale data.

  2. 2

    Scribe drafts; assumptions marked.

  3. 3

    Sentinel reviews for risk, commitments, and missing criteria.

  4. 4

    Atlas routes the package to you for approval.

Approval gate: Drafts only until a human approves send or publish. No invented metrics.

Workflow 3

Approved execution

Agents: Pilot (briefed by Atlas); Sentinel when output could go external. Trigger: The plan is set and someone must act in tools, browser, or files. Artifact: Completed task report and any files or screens changed.
  1. 1

    Atlas confirms scope and approvals already given.

  2. 2

    Pilot executes the approved steps.

  3. 3

    Results report back to Atlas.

  4. 4

    Sentinel reviews when the output could commit you or go external.

Approval gate: Confirm before spend. Secrets never leave safe storage. No guessing on price, terms, or timelines.

Workflow 4

Security escalation

Agents: Sentinel → Atlas → Aegis; Pilot only after approval. Trigger: Finding is primarily cybersecurity (vulnerability, compromise, secrets, access, compliance posture). Artifact: Security assessment and recommended actions (no offensive exploit steps).
  1. 1

    Sentinel (or any teammate) flags the security nature of the issue.

  2. 2

    Atlas hands off to Aegis with a clear brief.

  3. 3

    Aegis assesses and recommends.

  4. 4

    A human approves irreversible changes; Pilot implements only from that path.

Approval gate: Recommendations are not automatic live changes. Your approval comes first.

Every path has an owner.

Each flow names who does the work and where a human approves.

Meet the team
Meet the team