cover — Backstage

Flagship

Backstage

Governance. Epics, parity, taxonomy — the studio's own operating system.

Backstage
  • Statusstable
  • Since2026
  • Accessinstall

Currently workingActive epics

  • Checkpoint MachineP02E004 — The lifecycle machine is proven: a status change fires a checkpoint that runs checks, each check returns pass or fail, all passing authorizes the transition, and any failure rejects it and records the result. The minimum structure of check and checkpoint works with one toy check, before any real policy exists.
  • Nimble CompaniesP02E020 — The concept of a nimble company is explored and written: an organization that operates with autonomous agents in deterministic workflows. The studio is positioned as the standard handshake protocol between those agents and those companies.
  • Cognition PolicyP02E022 — The policy is canonical: before increasing the intelligence of the model, ask whether the environment can already carry the answer. Nurture the environment, not the model. The four questions and the conformance test that enforce it are part of the policy, not advice beside it.
  • Status SchemaP02E031 — There is one canonical source of statuses for policies and for skills, read by the validator, with the heartbeat reconciling drift. It mirrors what the epic schema already does for epics, so a status is a value the machine knows instead of a word a writer chooses.
  • Organization LevelP02E032 — The canonical organization level is defined: a sibling of the agents directory that holds artifacts shared by every agent and owned by none. Today it does not exist, because the work-location policy models the inside of the agents directory and stops there.
  • Project TiersP02E034 — The studio has three project tiers: flagship, experimental and collab. Each project declares its tier, the public experience derives from it in resources and attention, and the tier itself is never exposed to the reader. The rules, the assignations and the enforcement are part of the taxonomy.
  • Cross Agent WorkP02E039 — The canonical mechanism for one agent working inside another agent's epic is decided and enforced. It never creates a second truth, both sides discover the work explicitly, no glob or git or validator breaks, and the id validator passes.
  • Agent RegistryP02E042 — Every agent has an allocated number from one canonical registry, so an identifier is assigned once and never guessed. The registry is the table that the portfolio, the epic ids and the documentation all read from.
  • Fourth SystemP02E043 — The invariant that a fourth system cannot be introduced was blocked in practice, and the block is recorded with its evidence. The ecosystem keeps three systems, and a proposal that would make it four has a named place to fail.
  • Epic KindP02E044 — Kind is a closed enum of epic: project, study, recipe or methodology. Each kind closes in its own way, the legal relations between kinds are stated, and three executable consumers read it. The free prose of the old epic type field died in September and nothing replaces it with prose.

Coming soonPlanning epics

  • ID CollisionsP02E005 — Two epics that share the same id cannot be told apart by the roadmap projection, and eighteen groups do. The target is an id that is the identity of the epic and not of its container, so the epic migrates without changing name. The epic closes when every id in the ecosystem is unique.
  • Demand IntakeP02E016 — A project declares a need, receives responses, and each response becomes work items. The intake side owns the declaration, the response lifecycle and the routing to tickets, following procurement discipline: the request is published, proposals arrive, and each proposal is a vendor thread.
  • Writing HeadP02E023 — The group that holds how agents write epics consolidates the writing and quality epics into notes for analysis, under one criterion: lose nothing. The provisional name does not matter. The contract of the group is still undecided, so nothing is executed yet.
  • Skill AuditP02E035 — The updated skill creator is applied to every existing skill: descriptions audited, evaluations run on the critical ones, triggers optimized, and documentation structure normalized with parity. A skill that survives the pass is one an agent can find and use.
  • Check ArchitectureP02E040 — The architecture of the check system scales from the single trigger of today to many triggers, with checks that attach selectively and generalize across projects that need variables. The gold standard is transcoding a government requirement into a check, including the simulation of what happens if that requirement is implied.
  • Delivery AnnotationP02E047 — The question is whether done needs a delivery annotation to distinguish finished from shipped in a named release, and the answer closes with a deterministic consumer. No state without an exit, and no inference from tags.
  • Skill publicationP02E048 — Publish a skill once and ship it to three targets. One pipeline checks the frontmatter each consumer requires, then pushes the same tagged revision to OpenClaw, to Claude, and to GitHub, so that a skill that runs here also runs there, and no target silently holds a different copy. The same instrument answers which skills this project owns.
  • Skill statusP02E049 — Give every skill one status that means one thing, derived from evidence where the evidence exists, so that a skill claiming production actually carries a battery behind the claim and a reader can sort the corpus without asking its author.
  • Done TimestampsP02E053 — The roadmap projects WHEN an epic became done, so the public surface shows the last three in real order instead of the arbitrary order of the projection. The data is in the epic today and does not reach the roadmap, which blocks the consumer from computing what the source already knows.
  • Release HomeP02E054 — The release publication mechanism needs a home after its name was reassigned to the portfolio publisher. Two outcomes are acceptable: it returns as its own skill under a name that says what it does, or the chain it guarantees is written into the contract that absorbs it.

In the same way that a well-designed object reveals how it wants to be held, this product reveals how it wants to be used before anyone reads a word. That is the whole ambition: cognition that shows itself in the surface.

What it is

It began as an internal instrument and stayed one. The shape came from use, not from specification — every part that survived is a part someone reached for more than once. What follows is a description of that shape.

How it works

Three movements, in order. First the intake, where material arrives unformed. Then the reduction, where it is made small enough to hold. Finally the return, where it goes back out into the same world that produced it.

flowchart LR
  A[intake] --> B[reduction]
  B --> C{holds?}
  C -->|yes| D[return]
  C -->|no| B

Status

Working, with the usual caveats. Nothing here asks for attention it has not earned, and nothing claims to be finished.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris.

Lorem heading

Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.

flowchart LR
  A[Backstage — architecture] --> B[ingest]
  B --> C{stable?}
  C -->|yes| D[flagship]
  C -->|no| E[experimental]
  
Backstage — architecture

Get it

Lorem ipsum — install path, requirements, what lands on day one.

Backstage — start