seed-from-web

name	seed-from-web
description	Bootstrap-time skill that web-researches the latest popular security certifications per domain and drafts JSON stubs for any missing ones, as one PR per domain.

seed-from-web

When to use

• Right after the initial hand-curated seed PR has been merged, to expand coverage from ~30 to ~100+ certs.
• Periodically (e.g. quarterly), to catch certifications introduced since the last sweep.

Process

For each domain in data/domains.json:

1. WebSearch with queries shaped like:
   • "top security certifications 2026 <domain label> ranked"
   • "<domain label> certification ANSI accredited 2026"
   • "<domain label> certification DoD 8140 baseline 2026"
   • For Japan-specific domains, add: "<domain label> 認定資格 IPA OR JPCERT 2026"
2. Collect candidate cert names, deduplicating against the existing manifest. Skip courses, training programs, and non-proctored "badges".
3. For each surviving candidate, call the add-cert skill workflow (fetch official + 2 third-party, write JSON, leave evaluation null). Populate prerequisites.recommended_certs[] when the official page identifies prerequisites — these power the roadmap UI's arrow overlay and are easy to harvest at this step.
4. Group results by domain into one PR per domain. PR title: seed: <domain> (<n> new certs). PR body lists each cert with its official URL.
5. Run make validate and make manifest before opening the PR. Run make evaluate locally if the maintainer wants tier outputs in the same PR; otherwise leave evaluation.computed_tier = null and let the maintainer trigger evaluation on review.
6. Run python3 scripts/infer_relationships.py --dry-run so the per-domain PR also surfaces any prereq edges the new certs should acquire. Apply the accepted edges before merge so the roadmap arrows reflect the new certs immediately.

Refusal conditions

• A candidate cert lacks any third-party evaluation: skip it, list it in the PR body under "deferred candidates" with the reason.
• A candidate's official site is unreachable: skip it, add a TODO in the PR body.

Anti-goal

This skill must not simply scrape Paul Jerimy's or CyberDudeKZ's lists and convert them to JSON. The list of candidates must come from web search (or government / accreditation bodies). The two reference roadmaps may be linked in sources[] with "use": "inspiration_only" but never as the primary discovery mechanism.