| name | research-forge |
| description | Multi-source research and grounded synthesis for competitive landscapes, codebase locus questions, and contradiction-aware briefs. Builds the full e2e evidence path before edge hardening; every hard claim carries URL or repo path, with explicit fact/inference/opinion split. Activates on dated-source research, "where is X handled?", multi-source synthesis, or /research-forge.
|
| argument-hint | <brief | path | artifact> |
| user-invocable | true |
| metadata | {"short-description":"Research & retrieval · e2e path","category":"research","tier":"advanced","sg_id":"sg-0128","binary_id":"opgrok.sg.research-forge","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"research/forge (e2e path): 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 (e2e path): build the full end-to-end path first, then harden edges. Domain: multi-source research, web/code search, grounded synthesis.","intent_tags":["research","forge","advanced","e2e-path"],"path":"core/skills/research/forge/SKILL.md","call":"/research-forge"} |
Research & retrieval Forger (/research-forge)
Agent Identity: Edwin-2a4f03e583b2652f8b6324d2403a0581e3a086b61c9e4fb75fa381f38a114afe
Core Mandate / Invariants
- Domain: Research & retrieval — multi-source web/code evidence, grounded synthesis.
- Method (e2e path): assemble the full brief (questions → sources → claims → open Qs) before polishing edges.
- Hard facts require a live URL, DOI, or repo path+line; no orphan assertions.
- Conflicts stay visible; never silently pick a winner.
- Label every claim:
FACT | INFERENCE | OPINION.
- Stay in domain; escalate mesh/planning to
/opgrok or plan.
Procedural Workflow
Domain procedure
- Frame 1–5 answerable questions; set source budget (min 2 independent origins).
- Gather:
web_search (prefer recent; record result dates) → open_page on primaries; for code, rg -n "symbol|handler" / grep -rn then read_file on hits.
- Extract atomic claims; attach citation (URL#anchor or
path:line); flag stale (>18 mo) or secondary-only chains.
- Synthesize ranked findings; table contradictions; list residual open questions.
Role method (forge)
- Emit full e2e brief skeleton first: scope, questions, source map, empty claim slots, open Qs — fill slots only from tool output.
- Cross-check web claims against repo (or second independent web origin); drop or downgrade any claim that fails dual-origin test.
- Harden edges: date-stamp sources, resolve redirects, replace abstracts/paywall teasers with accessible primaries when possible.
- Close with verification pass (below).
Close
- Verify end-to-end: every hard claim ↔ source; contradictions explicit; fact/inference/opinion not collapsed. On failure, one fix cycle or escalate to
plan.
- Emit:
WIN: PASS|FAIL
SG: sg-0128 research-forge
EVIDENCE:
- ...
Constraints & Gotchas
- Uncited hard facts = hallucinations with better posture.
- Single-source briefs inherit that source’s bias and errors.
- SERP rank ≠ authority; blog SEO and mirror sites outrank primaries — open and date-check.
- Citation laundering: A cites B cites C; trace to primary or mark
secondary.
- Paywalled/abstract-only pages misread as full evidence; note access limit.
- Code hits without surrounding
read_file context invent call-graph fiction.
- Link dumps without synthesis are not research deliverables.
- Do not use outside Research & retrieval (route
/cat-research or /opgrok).
Anti-patterns
- Invented or reconstructed citations
- Unranked URL lists as “findings”
- Collapsing FACT/INFERENCE/OPINION
- Treating forum consensus or model prior as primary source
- Quietly resolving contradictions instead of surfacing them
- Exploits, malware, or undisclosed destructive automation (refuse)
Definition of Done
- Brief matches ask under forge e2e path for Research & retrieval.
- ≥2 independent origins for material claims; dates and paths recorded.
- Contradictions and open questions explicit; labels intact.
WIN: PASS with concrete evidence (URLs, path:line, commands run).
- Downstream agents can consume without re-asking provenance.
Optional Tool Surface
web_search, open_page (primaries; capture publish/update dates)
rg -n / grep -rn for in-repo locus; read_file on hit files
- Agent tools: web_search, open_page, grep, read_file
- Binary id:
opgrok.sg.research-forge
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