Tasks
Navin task board: plan, enrich and steer as human, agent or hybrid
August 5, 2026 · 4 min read · Navin team
How the agent manages tasks like an expert: depends_on, critical path, acceptance, evidence, ledger, human/agent/subagent assignment and autonomy.
A real expert does not manage tasks like a todo list: they decompose, order, enrich, assign, verify and replan when reality moves. That is exactly what the Navin board enforces - for you, for the agent, or for a hybrid duo.
Live kanban, dependency graph, critical path, acceptance criteria, mandatory evidence, mission ledger, optional autonomy. Not sticky-note candy: a delivery post.
What the agent can really do with tasks
| Capability | How | Why it matters |
|---|---|---|
| Plan | /blueprint, milestones, atomic tasks | A goal becomes an executable queue |
| Order | depends_on (not priority alone) | The board computes ready / blocked / critical path |
| Enrich | description, acceptance, validation, labels, agent hint | Every card is a brief, not a vague title |
| Assign | human:…, agent:…, subagent:… | Clear hybrid: who decides, who executes |
| Execute | board next → claim → work → comment → move | One discipline for human and agent |
| Verify | validation=test|lint|verify + evidence | No done without proof |
| Replan | ledger + ledger_replan / pause | Stalls and loops detected, no blind racing |
| Ship | autonomy: branch, PR, issue sync | The task ends as reviewable git, not "ok chat" |
Three collaboration modes
1. Dev-first (human pilots)
You create and refine cards in Tasks (Code or Project Home). The agent only takes what you hand it (/board task <id> or a manual claim). Ideal when business judgment stays with you.
2. Agent-first (agent as expert)
/blueprint or /mission: the agent reads the board (board plan), creates milestones and tasks, wires depends_on, sets acceptance + validation, then executes via /forge or /cruise. You approve the plan, not every micro-move.
3. Hybrid (the real expert)
The human sets the goal, constraints and irreversible decisions (blocked + human:…). The agent enriches, splits, executes ready work, spawns subagents on parallel items, and stops as soon as information is missing. Magentic-One / Progress Ledger pattern - native to Navin, no external framework.
Golden rule: depends_on, not priority alone
Priority says how much it matters.
depends_on says what is physically impossible before something else.
The board derives three useful facts:
- ready - every dependency is
done; - blocked - open dependency, or a human parked it;
- critical path - the longest open chain (shorten it, shorten the project).
board next returns the queue already sorted: critical path first, then what unblocks the most, then priority. The agent does not "feel" the next card: it follows the contract.
Enrich a task like a tech lead
A good Navin card carries:
- a clear title and a single deliverable;
- description with context and out-of-scope;
- binary acceptance (done / not done);
- validation (
test/lint/verify/manual/none); - depends_on toward real prerequisites;
- an explicit assignee (human, agent, subagent);
- ongoing comments (findings, file paths, decisions).
Audits (/inspect, /fortify, /debug) must create status=fix tasks for every confirmed finding - otherwise knowledge dies in the chat.
Statuses of a real pipeline
backlog → planned → in_progress → review → audit → fix → done (+ blocked / cancelled).
review: a human should look;audit/fix: findings to process;done: evidence attached when validation requires it;blocked: decision, credentials, or irreversible work - with a comment on what is missing.
Autonomy: when the expert chains
With board autonomy consent:
board next→ claim;- isolated
navin/task-<id>branch; - implementation + tests;
donewith evidence → push + PR;- GitHub issue closed when sync is on;
- next task, until blocked or the queue is empty.
You keep the merge. The agent stays the disciplined executor.
Comparison: SaaS todo vs Navin board
| Need | Todo / Jira alone | Navin agent board |
|---|---|---|
| Who writes tasks | Human | Human, agent, or both |
| Execution order | Subjective priority | Graph + critical path |
| Proof of done | Checkbox | Mandatory evidence |
| Agent context | Copy-paste | Card = runtime brief |
| Parallelism | Manual | Ready queue + spawn |
| Git delivery | Elsewhere | Branch + PR linked to the card |
FAQ
Can the agent create tasks on its own?
Yes: create / update / move / comment / milestone, like a human in the Tasks panel. Under autonomy, it chains without asking again between cards.
What if information is missing?
It sets blocked, assigns human:…, comments, notifies - and does not invent facts.
Does /blueprint execute code?
No. Plan + ledger + tasks. Build starts with /forge, /cruise or the Build button.
Where do I see the board?
Code workbench and the Tasks tab in Project Home: same component, same live events.
Download Navin · Features · Documentation
Recommended next reads
- Git / PR / issues autonomy → autonomous board
- Project continuity → Project Home
- Health radar → Vision 360 audit
- Agent / Plan / Review cycle → Code module
- Multi-agent team → Navin agents
Conclusion
Managing tasks like an expert is not checking boxes: it is planning a graph, enriching every card, choosing human / agent / hybrid, demanding evidence, and replanning without losing the thread. That is what Navin industrializes - so the agent is no longer a brilliant chat, but a delivery collaborator.
Try Navin on your machine
Local agent, cross-platform. Code, debug, scrape, leads, security and review - without leaving Navin.