Skip to content

Repository instructions

This is an independent, AI-assisted certification study library. Read CONTRIBUTING.md, docs/CONTENT-POLICY.md and docs/SOURCE-QUALITY.md before substantive content changes. Follow docs/GUIDE-QUALITY-STANDARD.md for guides.

Subagents for substantial refreshes

For a multi-guide or research-heavy certification refresh, use a coordinating agent and up to three research subagents, within the available concurrency limit:

  1. Exam researcher: official objectives, credential identity, prerequisites, delivery status, replacements and retirement evidence.
  2. Technical researcher: product documentation, release channels, volatile behavior, contradictions and unresolved guide findings.
  3. Learning-resource researcher: specific tutorials, workshops, repositories, courses and videos, with access, scope, time estimates and source limitations.

Research subagents inspect sources and report proposed changes. They must not edit shared guides, catalogs, accepted snapshots or review ledgers. Give each worker an explicit scope and an exclusive scratch output path. Return exact opened URLs, observations, affected exams, source/candidate matches, confidence, unresolved questions and a bounded recommendation. These instructions describe their assigned behavior, not a separate filesystem sandbox.

The coordinator reconciles findings, owns shared-file edits, preserves earlier work, and runs validation. Keep the existing automated maintenance checks sequential; do not launch duplicate whole-library collectors. Use coherent research batches of about ten guides, never more than twelve per packet.

After applying changes, delegate a separate read-only review to a fresh-context agent. Supply the current diff, complete affected guides, exact objective and guide hashes, source evidence and applicable rubric. Do not give the reviewer the editing conversation as its evidence. Resolve material findings and recheck the affected work. Record same-context and fresh-context review accurately; neither is human review or live product/lab execution.

For small link corrections or work that does not divide usefully, a single agent is sufficient. If delegation is unavailable, record that limitation and complete the authorized work without claiming independent review.

See docs/AUTOMATION.md (Subagent refresh workflow), docs/SOURCE-FRESHNESS.md, and docs/AI-AUDIT.md for evidence and completion rules.

Verification

Use the checks in CONTRIBUTING.md. Never equate reachability with technical verification or advance full-review dates after an automated match. Preserve dated records and accept changed objective baselines only with reviewed content. Generate CERTIFICATIONS.txt, lifecycle pages, review status and shared learning resources with their existing scripts. Do not hand-edit .site-build/ or site/.