User documentation¶
End-user docs for Kneo Agent Studio — for developers importing, converting, hardening, and handing off agents as governed Kneo specs.
For the top-level docs index see the docs home.
The policy, amended [0.8.0 A9]¶
This page used to end with "the remaining pages are planned; they are written
alongside the features they document, cut by cut." That was not true, and had
not been for several cuts — features shipped and their pages did not, which is
the promise breaking rather than the schedule slipping (docs-user/F-04).
Amended rather than met, deliberately. Committing the whole page backlog is a larger piece of work than any cut has had room for, and a promise renewed each cut and broken each cut is worse than an honest floor. So:
- The floor is task coverage, not page count. Every shipped task a user can perform is covered somewhere in the pages that exist, and the reading order below says where. A task with no section covering it is the defect; a task with no page of its own is not.
- A separate page per topic is backlog, not a commitment. Items below marked covered in are complete against the floor. Items marked planned are neither written nor promised for a particular cut.
- A page is written when a feature needs more than a section, which is a judgement about the material rather than a rule about cuts.
The earlier finding's "six shipped surfaces have no user page" is withdrawn as a count: it mixed current and future work and was never measured. The defect that stands is the broken promise, and this is the amendment.
Reading order¶
quickstart.md— run the BFF + SPA, connect a platform profile, import your first artifact. Written.hosting.md— running a shared instance: thestaticandoidclogin modes and their configuration files, admission, and what a reverse proxy needs. Written. Added at 0.2.1: the login modes shipped in 0.2.0 and their two environment variables appeared in no published document, so the release's headline feature could not be configured from anything a user could read.debugging.md— the debug loop: walking the graph, reading the trace, comparing checkpoints, and stopping a run. Written. It also names what is not built — replay and resume — says how the board behaves when a run finishes, and names the platform's paging depth that the checkpoint strip and the trace stop at, rather than leaving a reader to decide whether a puzzling reading is a bug in their spec.import_dify.md— importing a Dify app export: the DSL versions Studio reads, the support matrix — every app mode and node type, converted or carded — and the cards a Dify import usually carries. New in 0.9.0; 0.8.0's Dify importer was a scaffold that answered501 not_implemented, and this page was listed below as planned while that was all there was.
Covered in the pages above, no separate page:
- Importing an n8n workflow → quickstart, which walks a
real import end to end.
import_n8n.mdis backlog. - Reading gap cards, and clearing them — what "manual mapping required"
means, who may acknowledge or attest a card, and what to do when a workflow
cannot be converted at all → quickstart.
gap_analysis.mdis backlog. - Exporting and handing off — the governed spec, and the Hand-off page that
converts it into what the Dashboard's Launch accepts →
quickstart.
handoff.mdis backlog. - Running a shared instance — login modes, admission, proxies, the analysis keyring, replicas → hosting.
Planned, and not promised to a cut:
spec_editor.md— the YAML editor and the platform-rendered graph.policy_hardening.md— tool permissions, guardrails, human-approval, secrets.test_run.md— proving the spec in dev/local.
What this list promises. Items 1–8 are complete against the floor above: every task a user can perform today is covered by a section one of these pages actually contains. Items 9–11 are backlog — they are written when the material outgrows a section, not on a schedule. A page appearing here is not a commitment that it will exist next cut; that is the promise this list used to make and kept breaking.