Skip to content

// Methodology - October 30, 2026

Definition of ready for AI agent tasks.

A definition of ready is a short checklist a work item must pass before a team starts it, such as clear acceptance criteria, a size the team can finish and no open dependencies. It is the entry check, where a definition of done is the exit check. For AI agents, it decides which cards an agent may pick up.

What is a definition of ready?

A definition of ready (DoR) is the list of conditions a team agrees a work item must meet before anyone starts it. The Agile Alliance glossary puts it this way: as with the definition of done, "the team makes explicit and visible the criteria (generally based on the INVEST matrix) that a user story must meet prior to being accepted into the upcoming iteration." The expected benefit, in the same entry, is to avoid "beginning work on features that do not have clearly defined completion criteria."

INVEST is Bill Wake's checklist for good user stories, from his 2003 post INVEST in Good Stories, and SMART Tasks: Independent, Negotiable, Valuable, Estimable, Small and Testable. Many teams build their DoR from it. The idea itself is not new. The Agile Alliance entry says the first formal description of a definition of ready "seems to be from October" 2008, and notes that the ETVX (entry-task-validation-exit) framework from 1985 anticipates Scrum's definitions of "ready" and "done".

Is the definition of ready part of Scrum?

No. It is a practice some Scrum teams add, not part of the Scrum Guide. The 2020 Scrum Guide gives each artifact one commitment: "For the Product Backlog it is the Product Goal. For the Sprint Backlog it is the Sprint Goal. For the Increment it is the Definition of Done." There is no definition of ready on that list.

The guide uses the word "ready" once: "Product Backlog items that can be Done by the Scrum Team within one Sprint are deemed ready for selection in a Sprint Planning event. They usually acquire this degree of transparency after refining activities." So Scrum's own test for ready is size, reached through refinement. Anything more is a team's choice.

Definition of ready vs definition of done: what is the difference?

One guards the start of the work, the other guards the end. Both are team-wide checklists, unlike acceptance criteria, which belong to a single card.

Point Definition of ready Definition of done
Checked Before work starts When work is finished
Question it answers Do we have what we need to begin? Does the result meet our quality bar?
In the 2020 Scrum Guide No (only "ready for selection") Yes, the commitment for the Increment
On a kanban board Guards the move from Backlog into Ready Guards the move into Done
Typical item "Acceptance criteria are written, each with a check" "Tests pass and the change is reviewed"

Our glossary entry on the definition of done covers the exit side, and acceptance criteria for AI agents covers how to write the per-card criteria a DoR asks for.

What goes in a definition of ready?

A few conditions that prevent the problems your team actually has, and nothing more. In his article on the definition of ready (last updated June 2023), Mike Cohn gives these example rules a team might use:

  • The acceptance criteria (conditions of satisfaction) have been fully identified.
  • The story has been estimated and is under a certain size, often around half of the team's velocity.
  • The team's designer has mocked up, or fully designed, any screens the story affects.
  • All external dependencies, on another team or an outside vendor, have been resolved.

There is no standard list. Cohn's own words: "There is no universal definition of ready that is suggested for all teams."

Is a definition of ready a good idea?

For most human teams, Cohn says no, and his reasoning holds up. A rule that says something must be 100 percent finished before work starts turns the DoR into a gate, and "the definition of ready becomes a huge step towards a sequential, stage-gate approach." He concludes: "for most development teams, I do not recommend using a definition of ready. It's often unnecessary process overhead." Where a team does use one, he suggests two things: avoid 100-percent-done rules (with a possible exception for dependencies on other teams or vendors), and favor guidelines over rules.

That argument rests on people overlapping their work and talking it through as they go. Wake's INVEST makes the same point: a good story is negotiable, and "details will be co-created by the customer and programmer during development."

An AI agent working alone changes that. It cannot walk over and ask. Anthropic's Claude Code best practices describe what happens instead: "In a run you started before going to bed, nobody tells it, and you find out the next morning that it spent the whole night building on the wrong assumption." So for cards an agent will work unattended, a short ready check pays for itself. Still follow Cohn's advice: include only the conditions the agent needs, and use guidelines wherever a rule would just add a gate.

What does "ready" mean for an AI agent?

Ready means the card holds everything the agent can't work out from the code: the goal, how to check it, what it may touch, and who signs off. Two vendors spell this out. Anthropic's best practices list five questions to answer "before you leave a task running on its own", starting with "Does it know what done looks like?" and ending with "Where does the result go, and who signs off?" GitHub's guide to getting the best results from Copilot says an ideal task includes "a clear description of the problem", "complete acceptance criteria on what a good solution looks like" and "directions about which files need to be changed."

GitHub also lists tasks you may want to keep for yourself, including ambiguous tasks, production-critical issues, incident response, and tasks involving security, personally identifiable information or authentication. That makes a useful last item: some cards are never ready for an agent.

Here is the definition of ready we suggest for agent cards, with Cohn's rule-or-guideline split applied:

Condition Rule or guideline Why an agent needs it
The goal states the outcome in one or two sentences Rule GitHub: "a clear description of the problem to be solved"
Each acceptance criterion names a check the agent can run Rule Anthropic: if the agent has nothing to run that tells it the work is done, "looks done" is all it has to go on
The card says where to look and what to leave alone Guideline GitHub notes the agent can often find the right code itself, so paths help but need not be complete
Cards it depends on are in Done Rule The one 100-percent rule Cohn allows for: an agent can't finish work that waits on someone else
Small enough for one person to review in one sitting Guideline Smaller cards mean smaller diffs to check; see breaking work down for agents
Not security, auth, personal data, an incident or an open question Rule GitHub lists these among the tasks to keep for yourself
The hand-off is named: where the result goes and who reviews it Rule Anthropic: "Where does the result go, and who signs off?"

An agent can check five of these itself by reading the card. The size and the "keep for yourself" calls are judgment, so a person makes them when moving the card into Ready.

How do you run a definition of ready on a Flux board?

Make Ready a column, let only people move cards into it, and have agents check each card before they start. A layout that works: Backlog, Ready, In Progress, Needs review, Done. Cards sit in Backlog until a person has checked them against the list above and moved them to Ready. Agents take work only from Ready.

Put every ready condition in the card description. As of October 2026, the Flux MCP server tool get_card returns the card's fields and its comments, but not its checklist items, labels or assignees, so a condition kept in a label or the checklist section is invisible to an MCP agent. The list_cards tool returns every card in a column in board order, each with its title, description, priority and due date, so an agent can scan the whole Ready column in one call. get_board lists each column's cards but without their descriptions.

Then give the agent the check itself, in CLAUDE.md, AGENTS.md or a skill:

## Ready check (run it before you start any card)

Only take cards from the Ready column. Never start a card in Backlog.
Read the card with get_card. It is ready only if its description has:
1. Goal: the outcome, in one or two sentences.
2. Acceptance criteria, each with a check you can run
   (a test, a command and its expected result, or a URL).
3. Depends on: card IDs, or "none". Every card listed must be in Done
   (check with get_board) or archived (get_card shows archivedAt set).
   A card you cannot find is not ready.
4. Hand-off: where the result goes and who reviews it.
Scope (where to look, what to leave alone) helps but is not required.

If any of the four is missing or unclear, do not start the card.
Post one comment that begins "[your agent name] Not ready:" and lists
what is missing. Then move the card back to Backlog and stop.
If it passes, move it to In Progress with move_card before you change anything.
Never move a card into Ready yourself. A person does that.

When a card fails, the agent's comment tells the person who readied it exactly what to fix. For example:

[claude-web-1] Not ready:
- Acceptance criteria: "Works on mobile" has no check I can run.
  Suggest: "At 375 px wide the form fits without sideways scrolling,
  screenshot in the hand-off comment".
- Depends on: WEB-41 is listed but is still In Progress.
Moved back to Backlog. Nothing changed in the code.

Three things about Flux to plan around, checked against the server code in October 2026:

  • The gate is a rule, not a lock. Flux does not enforce column order. A key with the Write scope, whose owner is at least a member of the workspace, can move a card to any column on its board. The rule "Never move a card into Ready yourself" holds only because the instructions say so.
  • Sign every comment. Comments made with an API key appear under the name of the person who created the key, and get_card returns comments without author names. The bracketed agent name shows people and later runs who wrote it.
  • The comment is the record. Changes made over MCP are not written to the Flux activity log, so the "Not ready" comment is the trace of why a card went back. move_card always puts the card at the bottom of the column you move it to, so a card sent back to Backlog lands at the bottom of Backlog. Drag it up if it should be readied next.

With several agents on one board, the Ready column becomes the single queue they all pull from. Flux has no card locking, so give each agent its own Ready column or a claim convention. The kanban board for AI agents guide covers the Needs review column where a person checks the work before Done.

// FAQ
06 questions
01

What is a definition of ready in agile?

+

A definition of ready is a set of conditions a team agrees a work item must meet before they start it, for example acceptance criteria written, size estimated and dependencies resolved. The Agile Alliance glossary describes it as criteria, generally based on the INVEST checklist, that a user story must meet before it is accepted into the upcoming iteration.

02

Is the definition of ready part of the Scrum Guide?

+

No. The 2020 Scrum Guide names three commitments: the Product Goal, the Sprint Goal and the Definition of Done. It does not define a definition of ready. It only says that items the team can finish within one Sprint are "deemed ready for selection" in Sprint Planning, usually after refinement. A definition of ready is a practice some teams add on top.

03

What is the difference between a definition of ready and a definition of done?

+

A definition of ready is checked before work starts and asks whether the team has what it needs to begin. A definition of done is checked when work ends and asks whether the result meets the quality bar. On a kanban board the first guards the move into Ready or In Progress, and the second guards the move into Done.

04

What is a good example of a definition of ready?

+

A short list such as: acceptance criteria are written and each has a check, the item is small enough to finish in one sprint or one review, external dependencies are resolved, and anyone needed to answer questions is available. For AI agent cards, add two items: the card names where to look and what to leave alone, and it says where the result goes and who reviews it.

05

Is a definition of ready bad for agile teams?

+

It can be. Mike Cohn warns that rules requiring something to be 100 percent finished before work starts turn a definition of ready into a stage gate, and he does not recommend one for most teams. His advice is to keep it to guidelines rather than strict rules. The case is stronger for an agent working unattended, because it cannot ask questions mid-task the way a teammate can.

06

Can Flux stop an agent from starting a card that is not ready?

+

No. Flux does not enforce column order or check card contents before a move. Any API key with the Write scope, whose owner is at least a member of the workspace, can move a card to any column on its board. The ready check works because it is written in the agent's instructions, and because a person, not the agent, moves cards into Ready.

// Ready in, reviewed out

Ready the card once. Let agents take it from there.

Keep Backlog, Ready and Needs review on one Flux board, connect your agents over MCP or the REST API, and review their work before Done. Free to start, and agents that connect with an API key never take a seat.