- name
>-
SKILL.md
The skill file stays in the publisher's repository.
Paste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add mvschwarz/openrig --skill openrig-operating-model -a cursor -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add mvschwarz/openrig --skill openrig-operating-model -a claude-code -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add mvschwarz/openrig --skill openrig-operating-model -a codex -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add mvschwarz/openrig --skill openrig-operating-model -a grok -ySend the copied message in a Grok Bot chat. The Bot fetches this SKILL.md from GitHub and adds it to the skills all your Bots share.
Install this skill: https://github.com/mvschwarz/openrig/tree/HEAD/packages/daemon/assets/plugins/openrig-core/skills/openrig-operating-modelFrom the publisher
Canonical term: the Operating Model. How an OpenRig topology organizes every kind of context so that any agent — cold, fresh, or five generations in — can find what it needs and knows where to write what it learns. Authored context stays in addressed Markdown and manifests. Supported writers record semantic judgments; their read models remain rebuildable from those artifacts.
This skill owns the structure: trees, chains, tracing, and placement of authored knowledge. Resolve installed context through its current library addresses and the project's selected loadout; a private corpus is not a required public dependency.
Companion implementation assets in this skill's folder:
templates/ and scripts/ — starters and working tools, described in §8.Everything lives on one of two trees, and every kind of context is a chain: the same filename at every level of a tree. To know anything, you trace from where you are toward the root, reading that one filename at each level — nearer levels refine or override farther ones. The trace (§5) is the act of doing that ascent deliberately, with your own file reads.
fleet → instance → rig → pod → seat.project → mission → slice → proof item.An instance is one OpenRig daemon and the rigs it manages; a fleet is all instances. Where a daemon runs is a deployment fact, not an altitude — say the instance's name when you mean a particular one.
The one-parent law: the trace follows the
directory path and NOTHING else — no pointer fields, no link-following, no
branching, ever. The trace is the instrument confused agents reach for, so it
must be simpler than anything it corrects: path-only ascent fails only in
obvious ways. A child that relates to a second parent gets an also-serves:
ANNOTATION (read by humans and agents, walked by nothing) — and a genuine
two-parent tie usually means the item belongs one level up, under the common
ancestor, with mentions pointing down. The operational many-to-many lives in
the queue's tags, which are maintained by use. Legacy serves: lines are
transition bridges for the flat layout only: never required, never walked,
deleted as work nests properly.
The filename law: one name per chain, identical at every level. The folder
tells you whose it is; the filename tells you what kind it is. A seat named
pm-lead in two different rigs has two different LEARNED.md files — the path
is the identity. Never invent per-level names (no SEAT.md, no RIG.md): that
breaks the trace, the self-description, and every tool at once.
This table is the model. If you internalize one thing, internalize this.
The chains carry what is true of a position. How a kind of thing works — the operating
model — ships in the mode-neutral openrig-core plugin; an operating-mode plugin
(openrig-lab, openrig-factory, or openrig-hq) may refine it.
Chain files sit on nodes, not on the shelves that hold them. rigs/, pods/, seats/,
missions/ and slices/ are shelves — the trace passes through them and expects nothing there.
ONE authored node file: SPEC.md. Intent lives in its FRONTMATTER; the specification lives in
its body. Alongside it sit three files with different jobs and different writers:
A scaffold may create NOTES.md and its starter instructions; its lived entries are never
generated or projected. It is the work tree's lived file, the way LEARNED.md is the topology
tree's — the field-notes rung, where raw observation goes before anything has earned a place in the
node itself. Keeping lived files out of the render path is what makes them safe to write freely in.
Legacy name: older workspaces may use MISSION_NOTES.md. Keep legacy
content addressable; use NOTES.md for the chain because its name works at every altitude.
Current acceptance: Scope relationships and policy are authored in manifests; an item
judgment is recorded once against its promise, evidence and policy revision. Slice, mission
and project readiness derive upward. A distinct outcome or publication decision remains its
own authority. Queue ownership and generic done are custody facts, not proof acceptance.
Historical checkbox marks stay readable without acquiring current item authority.
Use rig proof --help for the supported write/read path. The project owner selects
proofPolicy.judges in project.yaml, mission.yaml or slice.yaml (nearest wins),
using exact actor addresses. There is no implicit fixed-role review conveyor.
rig proof judge mission/slices/slice#1 --verdict accept --reason 'Observed the outcome' --evidence proof/result.md
resolves the item identity, revision, evidence digests and retry identity. Evidence may be
an existing legacy artifact or a non-code outcome. A patch-equivalent subject also names
its actual comparison/adoption receipt with --comparison; the agent owns that judgment.
rig proof show returns the current basis and retained receipt; it does not publish anything.
Correct with reject, or withdraw when prior acceptance no longer stands. History stays
in the addressed proof home. A repeated request returns its original receipt and current
readiness; --replace deliberately reaffirms an old identical judgment after a correction.
Changed promises, policy or evidence require a current judgment. Explicit optional
`` markers preserve identity across wording edits; the
product otherwise derives identity from the full promise and reports ambiguity.
Intent composes; the body does not. The trace reads the intent: FIELD at every altitude, so
four levels unfurl as four sentences rather than four documents. The body is read only when you
are standing on that node.
compose.py up <node> --name SPEC.md --field intent --root <project>
A level whose file exists but lacks intent: is reported as a gap — a mission with no intent is
real information, never silently skipped.
Legacy README.md nodes stay valid indefinitely — the resolver prefers SPEC.md and falls
back, so nothing is forced to migrate and dormant missions need no attention.
PROGRESS stays separate. Narrative and legacy marks are retained testimony. For a
project with selected proof policy, current readiness comes from rig proof show and the
shared existing views, not hand-maintained parent checkboxes or copied status prose.
scripts/compose.py progress remains a legacy checklist renderer; it does not certify
attributed acceptance. Intent composes downward; proof readiness aggregates upward.
The axis behind the columns: every context kind has a template half (what ships — SOP, the default culture) and a learned half (what living in it taught — LEARNED, culture amendments). Only the pace of change differs: values change rarely and deliberately, like a constitution; practice changes constantly and cheaply, like working notes; intent changes at real pivots. Keep each file's pace; don't constitutionalize your notes or scribble on the constitution.
Seat knowledge needs a durable home that successors and other runtimes can read.
A private runtime memory or an urgent handover packet alone can omit standing
duties and the reasons behind a practice. Keep general operating craft in the
shared skill and position-specific knowledge in the seat's LEARNED.md chain.
At a handover, read the current chain alongside the packet and record any missing
job context before claiming readiness. Each occupant maintains what the seat has
learned; the handover packet carries the current transition.
Sections, in order — see templates/LEARNED.md:
Size: soft guidance, not a rule. Keep it as small as honestly covers the job — attention is the budget, and every reader pays it. Some seats genuinely need more; write what the job needs. The failure mode to watch for is the rulebook that only ever grows. Distill lessons into practices with reasons at deposit boundaries (before clear or handover); use that habit, not a line count.
A trace is walking your chains with your own file reads and writing down
where you stand: what am I doing (queue/NOTES) → under what contract (SPEC) →
toward what intent (SPEC.md intent field, leaf to root) → by what practice (LEARNED +
SOP) → within what values (CULTURE). A few written lines at the end.
scripts/compose.py up assembles any chain for you.
Two rules give the trace its value:
When to trace — one principle: trace when enough has changed that your
picture of where you stand may be stale — after a large stretch of work, at a
boundary (boot, handover, new mission, confusion), or when someone asks you to
reorient. Why not simply "every N hours": identical scheduled prompts fade
from an agent's attention with repetition, and idle seats accumulate ritual
traces that crowd out real context.
Where a schedule fits your context anyway, use one — but prefer gating the
action on evidence of change: scripts/trace-due.sh decides "has enough
happened since my last trace?" deterministically and stays silent when the
answer is no, so a scheduler can fire it as often as it likes.
rig walk — one idea, two endsThese get confused because they share a word. They are not in conflict; they are the same thing seen from either end, and the relationship is worth holding.
A chain is an ordered sequence of context meant to be absorbed one piece at a time. Two kinds exist and both are chains in that sense:
LEARNED.md; SPEC.md with intent:). The sequence is position: leaf → root.A chain can be traversed two ways, and that is the only real difference:
Pacing is the mechanism in both directions, and it is the load-bearing part. Absorption
between pieces is what makes a chain land; a concatenated dump of the same bytes is a failed
delivery regardless of content. That is why rig walk elapses --pace between pieces, and why a
composed render is meant to be read as a sequence rather than skimmed as a wall.
Use distinct names for the two directions:
compose.py up renders a node's chain to the root.rig walk — the push verb for paced delivery into a seat's pane.Use refocusing for the current trace workflow. A delivery receipt and an
agent's read-depth report answer different questions; pacing alone does not
establish understanding.
openrig-core's operating-model
skill and the shipped culture change through their owners, never by an instance
editing in place. If it is wrong for everyone, propose the change to its owner;
if it is wrong for you, that's what LEARNED.md is for.Everywhere: date what you write (from the clock), and correct by adding a dated correction rather than silently rewriting history.
Any node can be rendered: its chains assembled into one document
(scripts/compose.py; up = your effective view from a leaf, down = every
chain file under a root — run down at a tree root and you get the whole
operating picture in one document). Two rules:
tree shows the shape of a directory — where you sit, and where the
screams are, before you read a word of content.The root render + diff is how a high-altitude seat keeps a current mental
model of a changing fleet without reading everything: render the tree root,
diff against your previous render, read only the diff
(scripts/trunk-diff.sh).
Context lives at the narrowest scope that needs it; skills are only for what has no scope. Seat-specific knowledge → that seat's LEARNED.md. Rig practice → the rig's files. Only truly scope-free craft (useful to any agent anywhere) belongs in the skill layer.
A plugin cannot hold LEARNED. Plugins are shared; LEARNED is per-instance. Two rigs installing the same orchestrator plugin must not share what one seat learned about its own merge desk. That is why chains exist alongside plugins rather than being replaced by them.
And plugins are the distribution mechanism, which imposes a hard test. A plugin ships skills;
a skill may ship a script; that script must run unmodified on a stranger's machine. So it
resolves paths from configuration (rig config get workspace.root), never from a literal, and it
never asks the agent to work out which of several candidate directories is the real one. If a user
points their workspace at a git repo or a projects/ folder, everything keeps working with no
further setup. A script that only works for its author is not shippable, however correct it is.
A node holds what is true of THIS position and stays true.
Then check which tree. Traps, practice and how-we-work are position knowledge on the topology
tree (LEARNED.md). What is being built is the work tree (SPEC.md). One rig owns both; they
still do not mix.
Keep specification depth proportional. A project carries stable intent; a
mission carries its intent and mission-level specification in the same authored
SPEC.md. Slices provide their concrete implementation scope. Follow the selected
mission-slice-sop and any explicit mode overlay; mission specification is not a
requirement to repeat every slice detail.
Before you write down something you learned about a project, decide which kind it is.
LEARNED.md, or as the command that derives it. It never goes into shipped content
or a shared world.Agents find a project's declared context with rig context work-install, which lists what the project declares:
its intent, the context files in its project.yaml install block, and its skills. When project knowledge should
reach every agent working on the project, add it to that declaration rather than starting a second index.
These are independent, and merging them is seductive because the merged version is prettier.
stage:
enum (wip | provisional | established | canonical | superseded | retired).A pod-level item can be raw observation; a seat-level one can be canon. A skill can contain something immature and still be the right home, because audience picked the file.
So promotion to canon is NOT a move up the tree. It is a maturity event and can happen at any altitude — a seat-level observation that is universally true graduates straight to a skill. What earns maturity is evidence: recurrence, independent corroboration, a measured cost, surviving change, surviving an attempt to falsify it. Facts about mechanisms can skip the ladder (backticks substitute in double-quoted shell strings is one command away); inferences about practice must accrue (never broadcast to a large rig took an incident).
LEARNED.md is not a staged item — it is the bed everything lies in. Its gradient is positional: the dated append-log at the bottom is raw observation, the concise sections at the top are what survived. Attach distillation to a trigger: distil at deposit boundaries (pre-clear, pre-handover) where a write is already required and the author still remembers why each line exists; refocus merely notices when the log has outgrown the distilled part.
scripts/scaffold.sh creates any missing chain files from templates/ and
never overwrites or deletes anything — so a brand-new workspace and a
living system are the same command with different starting states. Created
files are marked UNSEEDED until their real owner writes the first true
version. Never mass-produce LEARNED.md content for other instances — each
instance writing its own first version is both how the knowledge becomes real
and how you discover which seats cannot yet describe their own job.
scripts/ are macOS-flavored in places (stat -f); verify
platform compatibility before trusting them on another OS.templates/{SPEC,LEARNED,SOP}.md ·
scripts/{compose.py, trace-due.sh, trace-stamp.sh, trunk-diff.sh, scaffold.sh}
© 2026 Skills Rack.
---name: openrig-operating-modeldescription: >----# packages/daemon/assets/plugins/openrig-core/skills/openrig-operating-model/SKILL.mdFAQ
Download packages/daemon/assets/plugins/openrig-core/skills/openrig-operating-model/SKILL.md from the publisher, or copy the path and open the repository. PlugMCP lists the skill and does not host the file.
The publisher keeps it at https://github.com/mvschwarz/openrig.
Claude, Cursor, Codex, and Grok Build can read a SKILL.md you add from the publisher's repository. Grok Bot installs it when you send the skill's GitHub folder link in a chat.