| name | skill-authoring |
| description | Create, update, and verify schema-compliant project skills with strong context engineering, resource depth, and evidence boundaries. |
| type | workflow |
| enforcement | suggest |
| priority | high |
Purpose
Use this skill when creating or changing .agents/skills/* so the result is a concise router plus a durable resource library. It protects context quality, schema compliance, evidence boundaries, and future-agent usability.
When to Load
- Keywords: create a skill, write a skill, skill resources, agentic engineering, context engineering, agent-context skill.
- Intent IDs:
create-agent-context-skill.
- File patterns:
.agents/skills/**, .Codex/contract/intents.yaml, Event.Architecture.Tests/AgentContext*Tests.cs, dev/active/** workstreams that plan a skill.
- A plan or handoff says implementation knowledge should become reusable agent guidance.
- A skill is becoming long, vague, under-resourced, or hard to verify.
When NOT to Load
- Not for ordinary application-code work unless the task changes agent instructions or skill files.
- Not for generic documentation edits that do not affect
.agents/skills, .Codex/contract, or agent-context tests.
- Not for creating external opencode configuration outside this repository.
- Not for adding a runtime feature where a domain, CQRS, auth, Blazor, or persistence skill is the primary rule source.
- Not for bypassing schema failures by adding skip-list exceptions.
Must-Read Docs
Top 5 Invariants
SKILL.md is a compact routing and invariant file, while deep explanations, templates, checklists, and domain material live in resources/*.md.- Every new skill folder name must match frontmatter
name, satisfy the required section order, and avoid AgentContextSchemaTests.SkipSchemaMigration exceptions.
- The skill must preserve evidence boundaries by separating verified repository facts, source-derived framework claims, assumptions, and future validation needs.
- Must-read links, resource indexes, and verification hooks must let a future agent reproduce the workflow without relying on prior chat context.
- Context updates belong in
dev/active/* when the skill work is substantial, especially before handoff, compaction, or switching tasks.
Top 5 Anti-Patterns
- Monolithic skill file, which exceeds the line cap and hides reusable depth outside resource files.
- Vague activation boundary, which loads the skill too often and wastes context on unrelated work.
- Unverified authority claim, which presents advice, plans, or framework interpretation as proof, certification, or implemented behavior.
- Broken resource graph, which leaves future agents with dead links, missing prerequisites, or no clear reading order.
- Schema-test bypass, which adds skip exceptions instead of making the skill conform to the project contract.
Minimal Examples
Skill authoring loop:
1. Classify the change with create-agent-context-skill.
2. Read AGENTS.md, _SKILL_SCHEMA.md, active plan/context/tasks, and source material.
3. Draft SKILL.md as a short router with exactly five invariants and five anti-patterns.
4. Move depth into resources/index.md and focused resource files.
5. Run architecture schema/link tests and update dev docs before handoff.
Minimum output boundary:
This skill provides workflow guidance grounded in the listed sources. It does not prove the product is correct, replace source review, or certify claims that require external validation.
Verification Hooks
dotnet test --project Event.Architecture.Tests/Event.Architecture.Tests.csproj --configuration Release --verbosity quietgit diff --check -- .Codex/contract/intents.yaml .agents/skills Event.Architecture.Tests/AgentContextLinkTests.cs
Related Skills
