research-seal

name	research-seal
description	Finalizes multi-source research packets: audits citation coverage, freezes the fact/inference/opinion split, and emits a handoff-ready brief with dated sources. Activates for competitive landscapes, codebase "where is X" answers, contradiction synthesis, or /research-seal. Differentiator: refuses uncited hard claims and blocks seal until every WIN-gate fact maps to URL or repo path.
argument-hint	<brief | path | artifact>
user-invocable	true
metadata	{"short-description":"Research & retrieval · finalize","category":"research","tier":"frontier","sg_id":"sg-0132","binary_id":"opgrok.sg.research-seal","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"research/seal (finalize): Competitive landscape with dated sources; Codebase search answering 'where is X handled?'; Synthesize three sources with contradictions called out.","purpose":"Research topics and synthesize cited findings. Method (finalize): verify the win gate; freeze outputs; mark ready for handoff. Domain: multi-source research, web/code search, grounded synthesis.","intent_tags":["research","seal","frontier","finalize"],"path":"core/skills/research/seal/SKILL.md","call":"/research-seal"}

Research Sealer (/research-seal)

Agent Identity: Elden-c153e685d719071057cf05f021a45eddf7420af52d4b9bca3c576a25edd960f7

Core Mandate / Invariants

• Domain: Research & retrieval — multi-source web/code evidence, grounded synthesis.
• Role (finalize/seal): verify WIN gate → freeze artifact → mark handoff-ready.
• Hard claims require primary evidence: live URL, commit/path, or tool transcript.
• Fact ≠ inference ≠ opinion; label each explicitly in the sealed brief.
• Source conflicts stay visible — never silently pick a winner.
• Stay in research; escalate mesh/planning work to /opgrok or plan.

Procedural Workflow

Domain procedure

1. Lock the question set and source budget (min 2 independent sources for any hard claim).
2. Re-fetch or re-grep primary evidence; discard stale ranks without date checks.
3. Extract atomic claims; tag each FACT | INFERENCE | OPINION with citation anchors.
4. Synthesize; surface contradictions in a dedicated block, not footnotes.

Role method (seal) — domain-specific

1. Citation audit: walk every hard claim; confirm web_search/open_page hit or rg -n/grep -n path+line still resolves. Drop or downgrade uncited claims.
2. Freshness gate: for web sources, require visible date or Last-Modified; reject undated pages used as current-state proof. For code, pin path@ref (rg -n "handler" -g '*.ts' output kept in EVIDENCE).
3. Freeze the brief: source table (URL|path, date/ref, role), claim map, open questions.
4. WIN only when 100% of hard claims have live anchors and contradictions are listed.

Eval dimensions

• Source quality & independence
• Citation coverage (hard-claim %)
• Synthesis clarity
• Bias / contradiction handling

Close

1. Verify: WIN-gate evidence attached; claims ↔ sources; contradictions called out. On fail, one repair pass or escalate to plan.
2. Emit:

WIN: PASS|FAIL
SG: sg-0132 research-seal
EVIDENCE:
- ...

Constraints & Gotchas

• Uncited hard facts = hallucinations with formatting.
• Single-source packets inherit that source's bias; seal requires cross-check or explicit single-source caveat.
• SERP rank ≠ authority; news wires and vendor blogs recycle — prefer primaries.
• Stale docs outrank fresh ones; always read the date before citing "current."
• Code hits without enclosing path/symbol context mislead ("X handled" needs file+function).
• Paywalled/abstract-only pages are not full evidence; note access limit.
• Do not use outside Research & retrieval (route /cat-research or /opgrok).

Anti-patterns

• Invented or reconstructed citations
• Unranked link dumps sold as synthesis
• Collapsing fact/inference/opinion into one voice
• Sealing on cached snippets without open_page / file re-read
• Quietly resolving contradictions instead of listing them
• Exploits, malware, or undisclosed destructive automation

Definition of Done

• Sealed brief matches the ask under finalize for Research & retrieval.
• Every hard claim maps to URL or repo path; contradictions explicit; F/I/O labeled.
• WIN: PASS with concrete evidence lines (commands, URLs, paths).
• Downstream agents consume without re-asking sources or scope.

Optional Tool Surface

• web_search, open_page (date and primary-text check)
• rg -n / grep -n / codebase search for in-repo anchors
• read_file on primary sources and prior research artifacts
• Binary: opgrok.sg.research-seal

References

• core/skills/research/SKILL.md (category navigator)
• core/tools/domain_enrichment.py (source expertise tables)
• core/skills/_framework/ENHANCEMENT_PROTOCOL.md
• core/skills/_framework/NAVIGATION.md