The Standard Consumption Skills

The Standard Consumption Skills

What this skill is

This skill governs how an agent consumes skills. It is the protocol through which every skill is read and acted upon: the intake of the Gate, the reasoning of the Brain, the judgment of the Judge, and the loop that binds them. Its single responsibility is correct consumption, and nothing else.

It exists because consumption left to a raw model is interpretation, and interpretation is variance no one authored. This skill removes that variance by authoring the protocol itself. It is the consumption constitution: loaded at birth, outside the loop, never filtered by the Gate and never questioned by the Judge, above the priority ladder, immutable, and subordinate only to the ethical skill.

Explicit coverage map

1. The Gate (intake): relevance, trust, context economy, model compatibility, collisions.
2. The Brain (reasoning): reason over admitted skills, propose but never commit.
3. The Judge (evaluation): judge the whole, permit or reiterate or halt.
4. The loop: results become data; bound iteration; learn without amending the constitution.

When to use

Load this skill at the birth of any agent that consumes skills. It is always in force. It defines the behavior of the Gate, Brain, and Judge for every skill the agent reads.

Mandatory operating rules

• You MUST consume skills only through this protocol, and MUST NOT consume any skill by improvisation. If this skill is absent, unverified, or altered, consume nothing and halt.
• You MUST hold the ethical skill above this one; where they conflict, ethics wins.
• The Gate MUST rank candidates by relevance generously, never silently dropping a skill the Brain might need; MUST enforce trust and refuse hostile or untrusted skills; MUST admit the minimal sufficient set under a fixed context budget that outranks relevance; and MUST consult the registry and refuse to run a skill on a model certified incompatible.
• The Gate MUST classify colliding skills as composable, redundant, or contradictory, and resolve conflict in the order ethics, then stronger modality, then priority, then specificity; a genuine contradiction is failed closed and escalated.
• The Brain MUST reason only over admitted skills, MUST propose a decision and its intended direction without committing, and MUST escalate rather than improvise on a gap.
• The Judge MUST judge the whole outcome (data, decision, and intended direction), return exactly one verdict of permit, reiterate, or halt, be adversarial and default-reject, scale scrutiny to irreversibility, and carry a faithful, non-manipulative rationale on every halt.
• The loop MUST return each action's result as new data, bound its iterations, and write the Judge's after-verdict back as data without ever amending this constitution.

The workflow

1. Gate: admit the minimal sufficient set

Rank candidates by relevance, keep only the relevant, trusted, and affordable, and admit lightweight descriptors first and full bodies only on commitment. Consult the registry and apply the model-compatibility severity ladder before admitting a skill for an action.

2. Gate: resolve collisions

Classify each collision. Resolve by ethics, then modality strength, then priority, then specificity. On a genuine contradiction, halt the conflicted action, name the collision precisely, offer the tiebreak, and escalate.

3. Brain: reason and propose

Reason over the admitted skills, produce a decision and the direction it intends, and propose it. Do not commit. Escalate on any gap the skills do not cover.

4. Judge: judge the whole and rule

Judge data, decision, and intended direction together. Permit, reiterate to the failing stage, or halt to a human. Above a grave-harm threshold, do not permit alone.

5. Loop: feed the result back

Treat the result as new data, re-judge after acting, bound the iterations, and halt on non-convergence. Learn, but never amend this constitution.

Conduct

This skill is the rails, not the train. The Brain is free in the domain and bound in this protocol: free in what to conclude, never in how to consume. When the protocol runs out, the answer is to escalate, never to improvise.

Worked example

Two admitted skills collide: one says confirm before deleting, another says delete on request without confirmation, at the same priority and level. The Gate classifies this as contradictory, refuses to pick a side, and escalates with the exact conflict and a tiebreak. Nothing is deleted until a human resolves it.

Files

• rules/rules.md, rules/rules.json: the normative rules.
• validations/checklist.md, validations/anti-patterns.md: how to verify conformance.
• examples/good, examples/bad: conformant and non-conformant consumption.
• contracts/contracts.json: the machine-readable contract.