using-agent-skills

name	using-agent-skills
description	Select and apply the repository skills that match the current task. This is mandatory for every agent/model execution.

Using Agent Skills

Purpose

Use this skill to route work to the skills that actually exist in this repository. It is a selector and workflow contract, not a catalog of uninstalled provider skills.

Required selection process

1. Identify the task type, affected subsystem and requested outcome.
2. Inspect the local skill catalog under .agents/skills/.
3. Select the smallest set of relevant skills; load each selected SKILL.md completely before acting on that skill.
4. Load a root skill first when it routes to sub-skills, then load only the applicable sub-skills on demand.
5. Record the selected skills and why they apply in the task context or active plan.
6. Run the selected skill's verification and the project validation required by .agents/rules/testing.md.

Do not reference or invoke a skill name that is not present in this repository. If a missing skill would materially change the implementation, stop and report the gap instead of pretending it exists.

Mandatory project skills

Situation	Skill	Requirement
Every agent/model execution	using-agent-skills	Select applicable skills before consequential work.
User explicitly asks to create, formalize or replan a plan	implementation-plan-authoring	Create or amend the persistent plan; stop at READY unless execution is separately authorized.
User explicitly asks to execute, implement, apply or continue a provided/identified plan	plan-conformance-execution	Execute with checkpoints, evidence and conformance.
Requirements, formulas or normative specification changes	req-traceability	Update the spec and generate traceability evidence.
Formal audit, technical findings or structured report	report-writer	Produce an evidence-based Markdown report.

Project skill catalog

These are the available top-level skills and their routing areas:

Skill	Use when
build-full	Building, packaging or deploying Echo Visualizer.
ci-cd	Reviewing or changing CI/CD automation; descend to its applicable sub-skill.
csharp	C# coding or MSTest-related work; descend to the applicable sub-skill.
decision-writer	Recording a durable decision or ADR.
dependency-upgrade	Upgrading framework, package or toolchain dependencies.
dotnet	.NET architecture/testing routing; descend to the applicable sub-skill.
graphics	Graphics work; descend to Direct3D, Win2D or RenderDoc sub-skills.
incident-writer	Writing an objective operational incident report.
issues-writer	Writing an issue, defect or technical-debt record.
localization	Localization work; descend to the relevant translation sub-skill.
multitask-coordinator	Only when the user explicitly invokes multi-workstream coordination.
report-writer	Formal reports and technical findings.
req-traceability	Requirements/specification/formula maintenance.
rust	Rust work; descend to the applicable ownership, errors, concurrency, performance, review or testing sub-skill.
scripting	PowerShell/automation work; descend to the applicable scripting sub-skill.
sub-agents-execution	Delegating work to an external CLI agent when useful and authorized.
windows	Windows API/platform work; descend to the applicable sub-skill.
winui	WinUI 3 work; descend to design, workflow, packaging, review or UI-testing sub-skills.
writing-style	Voice analysis or voice-matched editing.

Role selection for delegation

Role selection is separate from skill selection. Before delegation, the model may choose zero, one or multiple compatible descriptors from .agents/roles/. If no descriptor is a close match, declare a task-local role in the prompt or execution context. A role adds scoped context; it does not replace this skill, the general rules or project policies.

Verification contract

Every selected skill must have a concrete verification result. For project changes, also follow .agents/rules/testing.md. For documentation or agent contract changes, validate paths, references, scripts and links; do not claim runtime behavior that was not exercised.

Plan activation boundary

Do not select either plan skill merely because the task is non-trivial. An ordinary request without a provided execution plan uses the relevant direct skills. If the task is too broad or risky to execute safely without a plan, ask the user to provide one or authorize plan authoring before making the change.