anchors flow
Operate the work flows (the puzzle of actions and results).
A FLOW is work driven by shape rather than by memory.
An ACTION (flows/actions/*.action.md) declares what it does and which RESULTS itoffers. A FLOW (flows/*.flow.md) fits actions together and says where each resultgoes. What used to be a rule to remember ("NEVER close with a blocking gate red")becomes the absence of a fitting: the BARRED result has no link to `done`.anchors flowSubcommands
Section titled “Subcommands”anchors flow build
Section titled “anchors flow build”Scan flows/ and write the flow graph into the map.
anchors flow build [flags]| Flag | Default | What it does |
|---|---|---|
--root |
. |
project root |
anchors flow next
Section titled “anchors flow next”The VALID exits of a step — the command that drives.
Answers "from here, where can I go", instead of listing everything andasking someone to choose. It is what separates having a flow from having one moredocument.anchors flow next <step> [flags]| Flag | Default | What it does |
|---|---|---|
--root |
. |
project root |
anchors flow show
Section titled “anchors flow show”Draw the flow — assembled at RUNTIME from the graph.
Draws the flow from the graph, never from a stored diagram.
A versioned diagram ages against the flow it describes, and the repository alreadymeasured what that costs (see the docs-fresh gate). Here the drawing is derived,so it cannot diverge.anchors flow show [flow] [flags]| Flag | Default | What it does |
|---|---|---|
--direction |
TD |
diagram direction: TD (top-down, tree-like) or LR (left to right) |
--mermaid |
emit the diagram as mermaid (boxes and arrows, for rendering) | |
--root |
. |
project root |
The work queue — family
Section titled “The work queue — family”Commands of the same family — what tells each apart:
| Command | What it does |
|---|---|
anchors next |
Pull and claim the next item from the queue (the worker calls this) |
anchors queue |
List live tasks (the work the watcher queued) |
anchors work |
Emit the work prompt of a stage (spec|code|feature|test|review) for a target |
anchors done |
Close claimed task(s) (moves to the .anchors/done/ history) |
anchors drop |
Discard a task from the queue without completing it (removes; does not archive) |
anchors reclaim |
Return to the queue the tasks stuck in claimed (dead worker) |
anchors deliver |
Record the delivery of a stage — the review trigger |
anchors decided |
Release the card escalate --for-user stopped, once the decision is made |
anchors discard |
Take off the board the card that no longer makes sense, without deleting it |
anchors unblock |
Open the work card that unblocks a card stuck in needs-user |
anchors escalate |
Open the issue for a change needed in the plan or the spec |
anchors report-bug |
Report a bug in Anchors itself at github.com/co2-lab/anchors |
anchors synthesize |
Two PRs conflicting in CONTENT become a synthesis card |
anchors watch |
The background watcher: sees changes and QUEUES work |
anchors monitor |
Watch the project’s long processes: one line when one starts, stalls, dies or ends |
anchors flow |
Operate the work flows (the puzzle of actions and results) |
anchors board |
The project board — published by the pipeline, or served live |
Anchors enables continuous, autonomous AI programming through a reactive task queue.
1. The Watcher & Queue Cycle
Section titled “1. The Watcher & Queue Cycle”anchors watch: Runs in the background. Whenever you or an AI agent saves a file (e.g.order.spec.md), the watcher analyzes what’s missing and enqueues the next action (create-feature,implement-code,run-tests).anchors queue: Lists the current tasks waiting in line.anchors next: The AI agent calls this to claim the next card without needing human prompts.anchors work <target>: Emits the structured prompt instructing the model what to do for that specific card.anchors deliver: The model marks completion; Anchors validates the delivery and advances the task.anchors done <id>: Closes the card and moves it to history.
2. Handling Obstacles & Escalations
Section titled “2. Handling Obstacles & Escalations”anchors escalate: When an AI finds an ambiguous requirement, it parks the card inneeds-userand opens an issue.anchors unblock: Once the human answers the question in the issue, this unblocks the card.anchors decided <card>: Resumes execution with the human’s decision registered.