From brain
Use when evaluating complex systems, agent OS design, workflow engines, automation, HPC, infrastructure, distributed systems, protocols, organizations, division of labor, responsibility gaps, bypass paths, hidden state, or local compliance with unclear global ownership.
How this skill is triggered — by the user, by Claude, or both
Slash command
/brain:whole-object-responsibilityThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when a system is complex, specialized, automated, or procedural enough that local steps can look reasonable while the whole object, state, failure path, or responsibility disappears.
Use this skill when a system is complex, specialized, automated, or procedural enough that local steps can look reasonable while the whole object, state, failure path, or responsibility disappears.
Core principle: preserve understanding, judgment, and responsibility for the whole object inside divided systems.
NO SYSTEM-LEVEL CLAIM WITHOUT OBJECT-STATE-CONTROL-FAILURE-RESPONSIBILITY CHAINS
If every local step is compliant but no one owns the whole, flag responsibility evaporation.
Weak models must use these meanings, not guess.
| Term | Meaning |
|---|---|
whole object | The full thing or consequence the system is actually handling, beyond local tasks or components. |
boundary | The real system edge: what is inside/outside, what assumptions hold, what can affect the outcome. |
state | The current facts the system depends on: files, memory, database rows, workflow state, queue state, credentials, provenance, environment. |
state owner | Who or what maintains the authoritative state. |
control chain | The path by which decisions become actions: policy, scheduler, workflow, API, shell, human approval, tool call. |
constraint | A real restriction on action, not merely a guideline: permission, scheduler limit, type check, workflow policy, proof obligation. |
operation | What a component actually does to the object or state. |
bypass path | A way to avoid the intended control or epistemic layer, such as an agent using raw shell instead of a workflow system. |
failure path | How errors become visible or invisible, how the system can fail, and where failure is handled. |
local role | A component, team, agent, protocol step, or subsystem with limited responsibility. |
global consequence | What all local actions produce together. |
protocol | A standardized procedure for action or coordination. Useful, but not a substitute for judgment. |
local rationality | Each local action is defensible in its own narrow context. |
global blindness | No one can see or judge what the local actions produce together. |
responsibility evaporation | Everyone follows protocol, but no one owns interpretation, cleanup, harm, or consequence. |
productive function | The real value specialization or protocol provides: scale, safety, comparability, reproducibility, cost reduction, recovery. |
over-dismissal risk | The critique ignores the system's real productive function because its narrative is inflated or immature. |
Do not accept system nouns as explanations.
When you see labels like agent, workflow, AI4S, automation, HPC platform, inference framework, web search, symbolic reasoning, tool use, end-to-end pipeline, reduce them to:
name -> object -> state -> representation -> constraints -> operations -> failure paths -> responsibility owner
Ask:
Use Unknown when missing.
Whole Object:
Boundary:
Local Roles:
Global Consequence:
Object Chain:
State Owner:
Control Chain:
Constraints:
Bypass Paths:
Failure Visibility:
Cleanup Owner:
Responsibility Chain:
Productive Function:
Over-Dismissal Risk:
Verdict:
Identify what the whole system is really handling. Do not stop at component labels, tickets, APIs, diagrams, or workflow steps.
Map how local roles compose:
local role -> local action -> state change -> global consequence
If local actions are clear but global consequence is unclear, flag local rationality, global blindness.
Ask:
Execution bypass is a responsibility failure, not just a tooling bug.
Ask:
No cleanup owner means no complete responsibility chain.
Do not treat specialization as the enemy. Identify what the division of labor makes possible:
Then separate:
Productive Function:
Inflated Narrative:
Responsibility Gap:
Narrow Valid Claim:
Aligned: object, state, control, failure, and responsibility are visible.Local rationality, global blindness: local actions are defensible, but the whole consequence is not understood.Responsibility evaporation: protocol compliance exists, but no final owner exists.Execution bypass: action can avoid the intended control or epistemic layer.Workflow theater: process organization substitutes for judgment.Decorative abstraction: abstraction looks clean but hides object, state, failure, or responsibility.Empty advancedness: features or packaging do not improve object contact, control, verification, or responsibility.Over-dismissal risk: critique ignores real productive function.For any complex system, first ask about object, boundary, state, control, failure path, and responsibility chain instead of accepting its name and narrative.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub ansatzx/cyberbrain --plugin brain