Concept boundary
Working agreement
Standing rules for how work is done in this project.
It is
- Project-wide rules
- Validation standards
It is not
- One task’s checklist
- A transcript dump
Your first real task
Choose one real, low-risk outcome. You do not need a perfect prompt. Axis should help supply the project context, plan, tools, and proof around an ordinary request.
Before you start: open Axis. With Axis Cloud that means opening your server’s address in any browser and signing in. If you run Axis yourself, see Self-host Axis. Make sure at least one model provider is connected.
Proof sequence
The Brightside mission
A real Axis session improved a client update without sending or publishing anything.
1 · Durable home
The project carries the standing rules and related work. 2 · Ordinary request
The user describes the outcome and safety boundary in normal language. 3 · Inspectable result
The session keeps the result, validation, and exact change evidence together. 4 · Real browser proof
Technical work follows the same lifecycle and ends in visible product proof.
Mission outcome
Section titled “Mission outcome”By the end, you will have:
- one project with a continuing session,
- one completed piece of work,
- evidence you inspected yourself,
- one useful correction or open item preserved for later.
1. Choose the durable home
Section titled “1. Choose the durable home”Open the project the task belongs to, and start the chat in the right Space, or create one (“Website”, “Hiring”, “Client: Brightside”). The project scopes the working agreement, memory, services and saved results; the Space gives the agent its goal, current status and Up next list before it starts.
For a first mission, use reversible work: improve a small page, investigate an error, organize a report, or review a set of changes.
2. Ask normally
Section titled “2. Ask normally”Describe what you want as you would to a capable colleague:
The settings page feels slow when it first opens. Find the cause, improve it,and show me evidence that the real page is faster and still looks right.You supplied the outcome and the observed problem. Axis and the agent should discover available project rules, inspect the code or source system, form a plan when needed, choose tools, and identify appropriate proof.
If a fact is essential and only you know it, the agent asks. It should not turn every discoverable detail into a questionnaire.
3. See contextualized execution
Section titled “3. See contextualized execution”The session becomes the continuing thread for this outcome. Open the Working fold to inspect the chronological work: reasoning, reads, commands, edits, browser actions, and other tool calls.
A multi-step job may show a compact plan above the composer. You can let it run, add a follow-up to the steering queue, or stop it if the direction is wrong.
Concept boundary
Section titled “Concept boundary”Concept atlas
Keep each kind of context in its proper place
Concept boundary
Memory
Lessons or decisions worth recalling later.
It is
- Corrections
- Preferences
- Reasons
It is not
- Current source truth
- Every past message
Concept boundary
Up next
The outcomes a Space still needs to deliver.
It is
- Open outcomes
- Agent checklists
- Parked items
It is not
- Permanent policy
- Long-term knowledge
Concept boundary
Session history
The supporting record of what happened here.
It is
- Requests
- Actions
- Evidence
It is not
- The project’s rulebook
- A replacement for source files
4. Inspect evidence, not the completion sentence
Section titled “4. Inspect evidence, not the completion sentence”Open the proof that matches the work:
- a real screenshot for a visual change,
- before/after timings for performance,
- destination state for an external update,
- tests and changed lines for code behavior,
- a saved report or Collection item for durable output.
The agent’s summary helps you navigate. It is not itself proof.
5. Exercise human judgment
Section titled “5. Exercise human judgment”Decide whether the outcome is right, whether the trade-off is acceptable, and whether any consequential next action is authorized. If the work needs a login, message send, deployment, purchase, deletion, or similarly scoped authority, Axis should stop at a clear human checkpoint.
A denial leaves that proposed action unexecuted. An approval authorizes only the action and scope described by the gate.
6. Make the next session better
Section titled “6. Make the next session better”Preserve only what should survive:
- Save a durable correction or lesson to Memory.
- Add unfinished work to the Space’s Up next list.
- Put a standing rule in the project’s working agreement.
- Keep a useful output in Collections.
That closes the loop: the next session starts ahead instead of from zero.
Mission complete when
Section titled “Mission complete when”You can point to the exact session, the exact evidence, the judgment you made, and the knowledge or open work that will carry forward.