agentleFS
theseus-runtheseus-run/theseus/.agents/skills/theseus-design/SKILL.md

theseus-design

Skill1 starsChanged 6 months agoMITView on GitHub

Use for Theseus product and domain architecture decisions, especially mission-dispatch semantics, primitive boundaries, runtime concepts, operator model, package ownership, and deciding whether a concept belongs in core or scaffolding.

The file

Show its rules and text

We split this skill into the rules it gives an agent

18 rules. About 0 are common, 17 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use this skill for Theseus domain design.

    line 8
  2. Distinctive:

    Before adding a primitive or package-level concept, ask: can this be removed while the mission system still works?

    line 53
  3. Distinctive:

    If yes, it is scaffolding.

    line 55
  4. Distinctive:

    Keep Mission, Tool, Capsule, Dispatch, and Satellite distinct.

    line 76
  5. Distinctive:

    Do not rename concepts casually.

    line 77
  6. Distinctive:

    Keep UI transport separate from dispatch semantics and runtime host semantics.

    line 78
  7. Distinctive:

    Treat the runtime host as live composition of primitives, not as a new primitive.

    line 79
  8. Distinctive:

    Treat fixed crews, named agents, and planning loops as harness scaffolding unless promoted by a current design note.

    line 80
  9. Distinctive:

    Prefer source-editable modules over plugin APIs, dynamic loading, manifests, or extension registries.

    line 81
  10. Distinctive:

    Treat systems as runtime behavior modules, not generic files and not full ECS machinery.

    line 86
  11. Distinctive:

    Keep Mission/Capsule composition replaceable where possible.

    line 88
  12. Distinctive:

    Treat Capsule as the mission black box.

    line 90
  13. Distinctive:

    Keep Mission primitive, but allow mission schemas and mission types to evolve.

    line 93
  14. Distinctive:

    Keep isolation provider-shaped.

    line 96
  15. Distinctive:

    Treat tool/model execution as Sandbox/Workspace-relative where world effects are involved.

    line 99
  16. Distinctive:

    Prefer clean-break evolution for WIP runtime contracts.

    line 101
  17. Distinctive:

    Keep design notes honest about status: current doctrine, draft idea, or superseded history.

    line 104
  18. Distinctive:

    Treat skills, crews, agents, retries, and planning loops as harness/scaffolding unless the primitive floor would break without them.

    line 105

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Theseus Design

Use this skill for Theseus domain design. Do not use it as an Effect API guide; load an Effect-specific skill for Effect mechanics.

What Theseus Is

Theseus is a mission dispatch system, not a chatbot.

Human intent: "I need X done. Go do it."

The system tracks the mission, dispatches work, records what happened, and returns a result.

Mission exists because chat is the wrong root object for serious work. Chat may be an interface mode, but work needs a structured envelope: objective, completion definition, scope, authority, evidence, and lifecycle.

Theseus is also a self-editable harness: agents should be able to evolve the source by adding typed systems, satellites, projections, sinks, tool/model catalog modules, and static assembly. It is not a plugin host.

Theseus should be isolation-native. Autonomy should be backed by explicit execution envelopes and promotion boundaries, not only by repeated permission prompts or command allowlists.

Primitive Floor

Everything else is built on these or is scaffolding:

PrimitiveWhy it stays
MissionHumans need a structured work envelope with objective, completion definition, scope, authority, evidence, and lifecycle
ToolModels need typed, controlled world access
CapsuleMissions need a durable black box for review, continuation, evidence, artifacts, and accountability
DispatchThe system needs to invoke a model with context and get a result
SatelliteDispatches need scoped observation and policy middleware

RuntimeBus is not a synonym for Satellite. RuntimeBus is the operator/client transport concept for runtime UIs and should adapt to the runtime command/control/query surface if implemented.

Sandbox and Workspace are runtime/harness concepts, not aliases for each other. Sandbox is execution isolation. Workspace is source-state isolation inside a Sandbox. A git worktree is a Workspace provider, not a security Sandbox.

Design Tests

Irreducibility

Before adding a primitive or package-level concept, ask: can this be removed while the mission system still works?

If yes, it is scaffolding. Keep scaffolding optional and outside primitive contracts.

Future-Proofing

Design as if a much better model arrives soon: larger context, stronger instruction following, fewer workflow crutches.

Likely scaffolding: verification loops, planning agents, cycle caps, retry rituals, and current Grunt/Agent distinctions.

Likely primitives: Mission, Tool, Capsule, Dispatch, Satellite.

Boundary Ownership

  • Core owns primitives and stable runtime contracts.
  • Tools own concrete world access.
  • Server owns wiring, hydration, persistence, and transport adaptation.
  • Icarus owns operator UI.
  • Docs own design memory, not runtime truth.

Concept Rules

  • Define the human/operator concept before defining runtime machinery.
  • Keep Mission, Tool, Capsule, Dispatch, and Satellite distinct.
  • Do not rename concepts casually. If a new term appears, decide whether it replaces or refines an existing term.
  • Keep UI transport separate from dispatch semantics and runtime host semantics.
  • Treat the runtime host as live composition of primitives, not as a new primitive.
  • Treat fixed crews, named agents, and planning loops as harness scaffolding unless promoted by a current design note.
  • Prefer source-editable modules over plugin APIs, dynamic loading, manifests, or extension registries.
  • Require explicit assembly for behavior that affects what agents see, can do, observe, decide, or auto-load. Conventions are allowed only when represented as named, typed, ordered, removable source modules.
  • Treat systems as runtime behavior modules, not generic files and not full ECS machinery. A system advances or reacts to live harness state.
  • Keep Mission/Capsule composition replaceable where possible; do not bake one current mission system into unrelated runtime concepts.
  • Treat Capsule as the mission black box. For now, one Mission owns exactly one primary Capsule. Do not create free-floating Capsules or arbitrary nested Capsules.
  • Keep Mission primitive, but allow mission schemas and mission types to evolve. Implementation, research, brainstorm, review, planning, incident, and quick task missions may need different ceremony.
  • Keep isolation provider-shaped. Docker Sandboxes, Sandcastle, Vercel Sandbox, E2B, Daytona, Modal, local Docker/Podman, host mode, and test fakes are candidate providers, not primitive doctrine.
  • Treat tool/model execution as Sandbox/Workspace-relative where world effects are involved. Promotion back to host/project truth should be explicit.
  • Prefer clean-break evolution for WIP runtime contracts. When a contract changes, migrate first-party adapters, tests, docs, and skills in the same pass instead of preserving compatibility shims.
  • Keep design notes honest about status: current doctrine, draft idea, or superseded history.
  • Treat skills, crews, agents, retries, and planning loops as harness/scaffolding unless the primitive floor would break without them.

Decision Checklist

  • Does the design pass the irreducibility test?
  • Does it survive a stronger model?
  • Does it define one concept clearly?
  • Is it in the right ownership layer?
  • Does it create duplicate terminology?
  • Does it accidentally turn scaffolding into a primitive?
  • Is it useful to an operator dispatching real work?
  • Does it need a plugin API, or can Theseus add a typed source module and static wiring?
  • Is behavior introduced through explicit assembly, or hidden behind ambient file/config discovery?
  • Could an alternate mission or audit model replace this part without rewriting unrelated packages?
  • Does this treat chat as the domain object instead of an interface mode?
  • Does this assume host execution or host checkout when it should carry Sandbox/Workspace identity?

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.