lost-rob0t StarIntel OSINT
Goal
Run an extensive, evidence-first OSINT pass that produces traceable source records and canonical-document proposals.
Requires web research and repository access plus a current StarIntel Auto-Dig checkout.
Input
- a bounded research question, target, and time/jurisdiction scope;
- the current Auto-Dig corpus and executable v0.9 schema;
- web, archive, filing, repository, and structured-data research tools.
Output
A source ledger, identity-resolution notes, supported and conflicting observations, coverage gaps, and candidate StarIntel records with exact provenance.
Workflow
- Search the local StarIntel corpus before external research. Preserve matching IDs, paths, relations, and contrary records.
- Build a query matrix from exact names, aliases, identifiers, domains, addresses, organizations, dates, products, predicates, and jurisdiction-specific terms.
- Cover the relevant source families in references/source-playbook.md; do not stop after ordinary web results.
- Prefer primary records, then use independent secondary reporting to discover, contextualize, or challenge them.
- Record each source URL, publisher, title, publication/event date, retrieval time, locator, archive URL when useful, and content hash when bytes are retained.
- Resolve identity with stable identifiers and discriminators. Keep ambiguous people or organizations separate until evidence joins them.
- Extract atomic observations and attributed claims. Record the exact source location and which record or relation each item supports.
- Seek counterevidence and log meaningful negative searches, unavailable records, date ranges, and source families checked.
- Reconcile conflicts by source authority, directness, date, scope, and version; preserve unresolved conflicts instead of averaging them away.
- Hand accepted material to the canonical document and relation skills, then record follow-up targets for the next pass.
Rules
- A search result, source link, co-occurrence, graph edge, or actor agreement is a lead, not proof.
- Keep observations, attributed claims, analysis, and unresolved hypotheses distinguishable.
- Preserve exact names, identifiers, dates, amounts, quotations, URLs, and uncertainty.
- Never manufacture evidence, citations, retrieval results, negative-search coverage, or corroboration.
- Use only public, lawfully accessible sources and the user's authorized data.