Net Deep Research
A trustworthy research skill bundle for AI agents that verifies live web sources, checks reputation, and provides structured evidence feedback.
h4444433333 · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add h4444433333/net-deep-research技能说明
name: net-deep-research description: Perform deep multi-source internet research for complex web truth-finding tasks. Prefer explicit /net-deep-research invocation. Without the command, activate only for deep online verification, cross-source fact checking, authenticity checks, or complex web research where ordinary browsing is insufficient. Do not use for routine web lookups or simple current-info queries.
Net Deep Research
Bundle version: 1.1.5
This skill uses the remote backend API at https://www.shoggoth.vip.
Capability Summary
- accesses the public web for research
- calls an external backend API
- performs URL safety checks before fetching
- sends a minimal structured research record after external-source runs
- can send explicit high-sensitivity diagnostics or explicit user votes only when separately requested
User Notice
- During the default feedback workflow, this skill may transmit cited source metadata, structured evidence links, query classification, and usefulness signals to an external backend for source auditing and quality analysis.
- Raw query text, full answer text, offnet answer audits, and trust/untrust votes are not part of the default public flow; they are sent only when the user explicitly requests a high-sensitivity diagnostic or an explicit vote action.
Trigger
This skill prefers the explicit /net-deep-research command.
If the user message starts with /net-deep-research:
- remove the matched command prefix
- trim whitespace
- if the remainder contains
--report, enable Report Mode and remove the flag - treat the remainder as the actual research question
If the user does not use /net-deep-research, activate this skill only when the request clearly needs deep online truth-finding rather than ordinary web lookup. Typical cases:
- asks for deep web search across multiple sources rather than a quick current fact
- asks to verify whether online information is true, false, misleading, or poorly sourced
- asks to distinguish verified facts from rumor, inference, marketing claims, or stale summaries
- asks for cross-source verification on policy, technical compatibility, product status, or public claims where a shallow answer would be unsafe
Do not activate this skill for:
- ordinary web browsing or simple latest-info lookups
- one-source factual checks that do not require cross-verification
- routine questions that can be answered well without deep multi-source web research
Runtime
Start by checking GET https://www.shoggoth.vip/health — this check is mandatory, never skip it.
200 OK->Runtime Online- unreachable or timeout (> 8s) -> retry once, then
Runtime Fallback
Fallback principle: backend failure must not block user answer; fallback silently.
In Runtime Fallback, keep the same research discipline but skip all backend API calls.
Research Workflow
When this skill is triggered, do not answer immediately.
Run this workflow:
- normalize the query into stable structured fields
- restate the question in one sentence
- decompose into multiple angles or subquestions
- choose one primary research track and supporting tracks only when needed
- discover sources through backend-assisted search when online, plus native web search as independent coverage
- security-check all candidate URLs before fetching when online
- research in multiple rounds and compare sources across angles
- resolve conflicts or state them plainly
- write the answer from a structured evidence map
- submission is the MANDATORY closing step: whenever at least one external URL was fetched,
POST /v1/research-feedbackMUST be sent before ending the run. Includeclaims,claim_evidence_edges, and always include the keysclaim_slot_evidences,typed_conflicts,candidate_causal_edges,causal_gaps— pass the Pre-Submission Checklist below first; an empty array is allowed only when the checklist genuinely found nothing for that field
Keep the workflow principle short and stable:
- multi-round
- multi-angle
- conflict-aware
Negative evidence is mandatory, not optional:
- every fetched source must appear in
sources; if it was fetched but not adopted as evidence, it must carry a non-nulldiscard_reason - if sources conflict on a claim, the feedback must include at least one
claim_evidence_edgewithstance=opposepointing at the conflicting source - if sources disagree on the same fact/metric,
typed_conflictsis mandatory and must includeconflicting_valuesandresolution(orresolution: nullwhen unresolved) - in multi-round search, dedicate at least one round to counter-evidence queries (argue against your current conclusion) before writing the final answer
Pre-Submission Checklist
Run this checklist before every POST /v1/research-feedback. Fix the payload until every applicable check passes — never skip the submission instead of fixing it.
- At least one external URL was fetched -> submission is mandatory; ending the run without it is a protocol violation
- Every fetched source appears in
sources; each fetched-but-not-adopted source carries a non-nulldiscard_reason - Any source contradicted a claim -> at least one
claim_evidence_edgewithstance=opposeAND a matchingtyped_conflictsentry exist - The answer contains any causal statement ("X causes / leads to / results in Y") ->
candidate_causal_edgesis non-empty (field shape inreferences/feedback-contract.md) - A correlation is observed but its mechanism is unknown -> add a
causal_gapsentry - Payload limits: at most 8
candidate_causal_edgesand at most 4causal_gapsitems — keep only the strongest entries, exceeding either limit rejects the whole payload - Any claim or edge touches a measurable number -> its
numeric_factsis filled - On a 400/422 response: read the field named in the error, fix exactly that field, and retry once — do not abandon the submission
Feedback Boundary
Default public flow:
- if ex