Name a Skill and describe when to use it
Describe a Skill’s task and activation conditions, then distinguish metadata parsing from host selection checks.
Skill authoring
$ARGUMENTS Variable
Define invocation arguments and map accepted flags to behavior.
Activation Scope
Describe both activation cases and exclusions for a skill.
Skill Composition
Compose existing capabilities through named inputs and outputs.
Eval-Driven Skill Improvement Loop
Improve a skill against preserved tasks, baselines and observable assertions.
Compositional Prompt Assembly
Assemble single-purpose prompt fragments under explicit inclusion conditions.
Pressure-Scenario TDD for Skill Authoring
Test skill rules under realistic pressure against a no-skill baseline.
Capability Retirement with Migration Paths
Retire a distributed capability with replacement and migration information.
Action-Verb Tool Abstraction
Separate portable capability instructions from host-specific tool bindings.
Invocation Economics (Context vs Cognitive Load)
Choose automatic or user invocation according to discovery needs and human control.
Classify text by its control function
Evaluate instruction text according to whether it defines a tool contract, routes a task or directs loaded behavior.
Calibrate instruction specificity to fragility
Choose broad guidance, parameterized recipes or exact commands according to the task’s justified variation and failure cost.
Environment as instruction source of truth
Avoid maintaining a stale prose cache of cheap executable environment facts.
Cross-Skill Evidence Distillation
Promote only recurring actionable principles with a stated failure consequence and a comparison against existing rules.
Decision-Enabling Audit Reasons
Make keep, improve, retire and merge verdicts state the evidence, action and replacement rather than a vague label.
Evidence-Gated Rule Promotion
Promote a local learned behavior only after checking recurrence, scope and counter-evidence.
Preserve Behavioral Control Semantics Across Localization
Preserve operational controls in agent instructions when translating or adapting prose.
Preserve Literal Examples Across Invocation Templates
Keep example-language syntax separate from the host's invocation-template syntax.
Preserve Executable Identifiers Across Localization
Separate translated explanation from literal symbols used by an instruction loader or reference resolver.
Codify a proven run with replay
Turn an accepted successful interaction trace into a deterministic reusable procedure, then verify replay against its accepted evidence.