Discovery and Routing

Discovery and Routing

The curator decides when to use built-in skills, installed skills, fresh GitHub search, curated indexes, or no external skill at all.

Routing Decisions

Decision	Use When
use_builtin_now	The task is one-off, simple, or already covered by built-in/system capabilities.
use_installed_skill	A local specialized skill clearly matches and is not stale, noisy, or unsafe.
search_before_execution	The task is repeated, high-value, fidelity-sensitive, or likely to benefit from a better skill.
ask_user_before_search	Search might help, but the value is uncertain.

High-Value Discovery Lanes

Run the task radar for important work where missing a specialized skill can hurt quality:

python scripts/task_skill_radar.py "<task description>"

Use fresh search when needed:

python scripts/task_skill_radar.py "<task description>" --run-search --force-refresh

High-value lanes include:

• academic research and research-paper writing;
• literature review, citations, BibTeX, reviewer response, rebuttal;
• LaTeX, formulas, math notes, TikZ, compiled PDFs;
• resume/CV tailoring, ATS, cover letters, job applications;
• graduate applications, SOP/PS, scholarships;
• DOCX, PDF, XLSX, and report workflows.

Curated Index Discovery

Use curated indexes to reduce screening cost, not to bypass safety:

python scripts/find_curated_indexes.py "<task description>" --top 8

If no strong curated index exists:

python scripts/build_skill_roundup.py "<task description>" --top 12

Core Reusable Task Lane

For repeated artifact workflows such as PPTX, DOCX, PDF, and XLSX, use pinned core checks before generic search:

python scripts/ensure_core_skills.py pptx

Only install pinned or strict-tier candidates after review and approval.