coding-skill-selector

Coding skill selector

Load coding skills for the languages/domains THIS TASK will touch.

1. Always: Skill("preferences") + Skill("architecture")

2. Language menu (preferred path): the session's routing prompt included a "Project language menu" block — an engine-run scan with per-extension file counts and tool probes. Pick from it ONLY the languages this task will actually touch: counts are a hint (1 stray file ≠ project language), the task text decides. Do NOT re-scan the project yourself. No menu in this session (spawned agent, scanner unavailable, empty block) → fallback: determine extensions from the task's files yourself. Tool-bound skills (a skill that mandates a CLI tool) load ONLY when the language's menu line ends in an available-probe marker — (<tool> OK) vs (<tool> missing) — or, when there is no menu or the language's line is absent from it, after one command -v <tool> probe succeeds. Tool unavailable → skip the tool skill and work with regular tools; a tool discipline only applies where the tool exists.

3. By file extension (menu-picked or fallback):

   • .py → Skill("lang-python")
   • .hx → Skill("lang-haxe")
   • .as → Skill("lang-as3")

4. Top-up rule: if mid-task you touch a file of a language whose skills are not loaded (task turned out wider than it looked) — load that language's skills per rule 3, including its tool conditions, BEFORE editing the file.

5. By domain (detect from imports, paths, or task description):

   • YOLO / object detection → Skill("domain-yolo")
   • Pixi.js (imports from pixi.js, PIXI globals) → Skill("domain-pixi")
   • Person re-identification (ReID) → Skill("domain-reid")
   • Game dev (physics, precision, camera, sprites) → Skill("domain-gamedev")
   • UI components / display-tree apps (ownership, state, events) → Skill("domain-ui")
   • Diagnosing a bug / fixes that don't stick after 2+ rounds → Skill("debugging")
   • OpenFL / hxcpp native target → Skill("target-openfl-native")
   • Math overflow / numeric boundaries → Skill("math")

6. By tooling/platform:

   • CMakeLists.txt / CMake build → Skill("build-cmake")
   • .github/workflows / CI / GitHub Actions → Skill("ci-github-actions")
   • AWS Lambda / .NET deploy → Skill("aws-lambda")
   • MCP server setup / .mcp.json / MCP troubleshooting → Skill("mcp-setup")
   • Claude Code settings.json / hooks / permissions / plugin config → Skill("claude-code-config")
   • Workflow YAML in templates/ or .claude/workflows/ → Skill("workflow-authoring")
   • Setting up / tuning a project linter, or a new project with no lint config → Skill("lint-setup") (skip if not installed; it loads the user's preferences-lint lists itself — neither loads for normal coding/review)
   • Fetching web docs / external content → Skill("web-reading")
   • Verifying a web/canvas app via Playwright or a browser MCP (screenshots, captures, perf) → Skill("browser-verify")

7. If 3+ independent files or parallel work → Skill("task-delegation")

Call Skill(name) for EVERY matched skill.

Gap check: If files involve a language/domain/tool with no matching skill above, ask the user: "No skill for [X] — proceed without, or create one first?"