Templates

Compare N sides of a subject in parallel and surface inefficiencies, gaps, and divergence. Each side gets its own subagent applying the standard research methodology to its own context. The lead then produces a structured diff that highlights what's worth acting on. Use this for: vendor / product / market / region comparisons; price arbitrage scans; feature gap analyses; regulatory delta scans.

General-purpose deep research, the canonical Anthropic-style multi-agent pattern. The LeadResearcher decomposes the query into distinct angles with minimal overlap, fans out parallel subagents that each run a broad-then-narrow search-evaluate-refine loop, and synthesises a structured report with cited sources, a confidence score, and open questions. Cost scales with the optional ``depth`` input.

Triggered automatically when PageDrifter detects a meaningful change on a watched page. Runs a focused, shallow research pass: what changed, why it matters, and what (if anything) the engine should remember. The output is intentionally short and structured so it composes cleanly with the wiki compiler's append flow.

Batch enrichment of Croatian company profiles. Each item in the ``inputs`` array carries a company name and optional OIB / country hint; the workflow runs the enrichment pipeline over each in turn and aggregates the structured fields ready to merge back into tvrtko's database. v1 wires this template to the built-in BatchEnrichment stub workflow. The migration path (story 015) is to graduate to a custom ``tvrtko.CompanyEnrichmentWorkflow`` Python class under ``config/workflows/tvrtko/`` once the deterministic FINA → Bisnode → news → merge pipeline is needed. The template syntax stays the same; only the ``workflow`` field changes.

Run the wiki compiler for a project. Walks the project's ``kb/{project}/raw/jobs/`` archive and upserts compiled wiki entities under ``kb/{project}/wiki/{type}/{slug}/``. Idempotent — already-processed jobs are skipped via the compile state file, so this template is safe to chain after every research job. This template is the canonical follow-up step in research chains. Templates that produce wiki-relevant content reference it via the ``follow_ups`` field so the engine queues a wiki maintenance job on the maintenance lane after every research job completes.

Run the wiki linter for a project. Pure structural checks — no LLM calls, no inference. Walks ``kb/{project}/wiki/`` and reports any missing files, malformed JSON/YAML, empty articles, stale entries outside the project's freshness windows, and entity types outside the project's allowed namespaces. This template is the canonical second follow-up step in research chains, after wiki-compile. Findings are returned in the job output for the dashboard to surface; the lint job itself always succeeds even when findings exist.