Task Management & Human Approval Gates Guide

Septra combines autonomous AI agency with strict risk governance. By establishing a fundamental distinction between autonomous internal research and external production modifications, Septra guarantees that no AI agent can alter your live ad spend, publish public content, or mutate production data without explicit human approval.

---

1. Two-Tier Task Architecture: System vs. Inbox#

Every Septra workspace provisions two reserved, purpose-built project queues designed to separate background autonomous work from customer-visible approvals:

| Project Name | Audience & Visibility | Task Types & Workflows |

| :--- | :--- | :--- |

| System (Internal) | Agent-Internal Only (Hidden in client UI) | Autonomous background cadences, daily data syncs, recurring SEO audits, scheduled GEO scans, and weekly financial rollup calculations. |

| Inbox (Client-Facing) | User-Facing (Central Work Queue) | Actionable deliverables requiring human approval, strategic campaign proposals, generated blog drafts, ad copy changes, and budget reallocation proposals. |

Why the Separation Matters#

Cadence Isolation: Agents run dozens of daily diagnostic and evaluation tasks in System without cluttering your operational task board.
Action Escalation: When a background audit in System identifies an optimization opportunity (e.g. Gale detecting a high-intent keyword opportunity or Aria identifying $1,200 of wasted ad spend), the agent drafts an actionable solution and routes it directly into Inbox for human review.

---

2. Four-Stage Task Lifecycle#

Work in Septra transitions through four distinct lifecycle states:

┌─────────┐      ┌─────────────┐      ┌───────────────┐      ┌────────┐
│  Todo   │ ───► │ In Progress │ ───► │   In Review   │ ───► │  Done  │
└─────────┘      └─────────────┘      └───────────────┘      └────────┘
                                      ▲               ▲
                                      │ Human Action: │
                                      │ Approve / Edit│
                                      │ / Reject      │

Todo: Tasks queued by users, scheduled by platform cron triggers, or delegated by executive agents awaiting agent processing.
In Progress: Actively being analyzed, researched, or authored by assigned specialist agents.
In Review (The Human Approval Gate): The agent has produced a complete deliverable artifact (e.g. ad creative, blog article, budget adjustment proposal) and is paused awaiting human authorization.
Done: Approved and successfully dispatched to target production systems with comprehensive audit trails.

---

3. The Safety-First Approval Gate Principle#

Septra strictly enforces operational safety tiers:

Tier 1: Read-Only Actions (100% Autonomous)#

Ingesting advertising metrics from Google Ads, Meta Ads, LinkedIn Ads, and Reddit Ads.
Analyzing Google Analytics 4 sessions, conversion events, and user paths.
Scraping competitor websites and public marketing changes.
Crawling site pages for technical SEO defects and Schema markup.
Querying LLM search engines (Perplexity, ChatGPT, Gemini) for brand visibility audits.

Tier 2: Write & Mutation Actions (Human Approval Required)#

The following actions never execute autonomously; they are always halted at the In Review gate:

Advertising: Pausing campaigns, adjusting ad group budgets, adding negative keywords, or deploying new ad headlines.
Social Media: Publishing live posts to LinkedIn company pages, X (Twitter), or Instagram feeds.
Content Management: Creating new articles, publishing drafts, or updating pages on your live CMS.
Code Repositories: Submitting pull requests or committing code changes to connected GitHub repositories.

---

4. Reviewing & Executing Deliverables#

When reviewing a task card in the In Review column:

Deliverable Preview: Inspect rich deliverable previews including formatted markdown articles, comparison tables, and visual creative assets.
Agent Reasoning Trace: Review the background reasoning, performance data, and source citations that led the agent to make the recommendation.
Three Execution Pathways:
- Approve: Immediately triggers live execution. The platform dispatches API calls to the target platform (e.g. Google Ads API or CMS endpoint) and moves the task to Done.
- Request Revision: Add inline feedback or instructions. The agent reads your comments, updates the deliverable, and resubmits it for review.
- Reject: Dismisses the proposed change. The deliverable is archived without modifying external production systems.