From llm-wiki
Multi-pass research — runs sub-skills in sequence, writes synthesis draft to outputs/, merges into wiki/ after review gate. For broad/complex topics.
npx claudepluginhub skinnnyjay/wiki-llm --plugin llm-wikiThis skill uses the workspace's default tool permissions.
Orchestrates a multi-pass research session for broad, complex, or high-stakes topics. Runs multiple sub-skills in sequence, collects evidence into `raw/`, writes an intermediate synthesis draft to `outputs/`, then merges into `wiki/` after a review gate.
Creates new Angular apps using Angular CLI with flags for routing, SSR, SCSS, prefixes, and AI config. Follows best practices for modern TypeScript/Angular development. Use when starting Angular projects.
Generates Angular code and provides architectural guidance for projects, components, services, reactivity with signals, forms, dependency injection, routing, SSR, ARIA accessibility, animations, Tailwind styling, testing, and CLI tooling.
Executes ctx7 CLI to fetch up-to-date library documentation, manage AI coding skills (install/search/generate/remove/suggest), and configure Context7 MCP. Useful for current API refs, skill handling, or agent setup.
Orchestrates a multi-pass research session for broad, complex, or high-stakes topics. Runs multiple sub-skills in sequence, collects evidence into raw/, writes an intermediate synthesis draft to outputs/, then merges into wiki/ after a review gate.
Use this sub-skill when:
Restate the topic as a single precise research question. If $ARGUMENTS is vague, clarify with the user or infer from context.
Example: "What is the current state of LLM reasoning benchmarks?" → precise question: "What benchmarks exist for evaluating LLM reasoning, what are their methodologies, and what are the leading model scores as of 2026?"
Define success criteria: what would "done" look like? (A wiki page with N sections? A synthesis covering X, Y, Z angles?)
Decompose the question into 2–4 sub-questions, each mapped to a sub-skill:
| Sub-question | Sub-skill |
|---|---|
| What papers / formal research exists? | wiki-research-academic |
| What do practitioners say? (blogs, docs) | wiki-research-web |
| What is the current community discussion? | wiki-research-social |
| What is the latest news? | wiki-research-news |
Write this plan to outputs/research-plan-<topic-slug>.md before starting:
# Research plan: <topic>
**Question:** <precise question>
**Success criteria:** <what done looks like>
## Passes
1. Academic — query: "<arXiv/Semantic Scholar query>"
2. Web — URLs: [<list>]
3. Social — HN search: "<query>"; Twitter: "<query>"
4. News — Perplexity: "<query>"
## Output path
outputs/research-<topic-slug>.md
Run each sub-skill sequentially. Do not skip to synthesis before all passes are complete.
Pass 1 — Academic (if applicable):
Invoke wiki-research-academic with the academic sub-question.
Collect raw/ paths from output.
Pass 2 — Web:
Invoke wiki-research-web with relevant URLs or discovered links.
Collect raw/ paths from output.
Pass 3 — Social (if applicable):
Invoke wiki-research-social with HN/Twitter queries.
Collect raw/ paths from output.
Pass 4 — News (if applicable):
Invoke wiki-research-news with recency query.
Collect raw/ paths from output.
After each pass, run skills/wiki-research/references/source-eval.md on new files. Drop sources that fail the minimum quality bar before synthesis.
Stop if you have enough: If 2 passes provide sufficient coverage (5+ high-quality sources), skip remaining passes and proceed to synthesis. More passes ≠ better synthesis.
Write to outputs/research-<topic-slug>.md. Follow skills/wiki-research/references/synthesis.md for structure.
Template:
---
title: "Research synthesis: <topic>"
status: draft
created: YYYY-MM-DD
sources:
- raw/academic/...
- raw/research/...
- raw/social/...
- raw/news/...
---
# <Topic>
## Research Question
<precise question>
## Summary
<3–5 sentence executive summary of findings>
## Key Findings
- Finding 1 — `raw/path/file.md`
- Finding 2 — `raw/path/file.md`
## Detailed Analysis
### <Angle 1: e.g. Academic Research>
<synthesis of academic sources>
### <Angle 2: e.g. Practitioner Perspective>
<synthesis of web/blog sources>
### <Angle 3: e.g. Community Discussion>
<synthesis of social sources>
## Contradictions and Open Questions
- <Source A says X; Source B says Y — unresolved>
- <Question not answered by any source>
## Further Research
- <Suggested follow-up queries>
## Sources
| File | Type | Quality |
|------|------|---------|
| raw/... | academic | 5/5 |
Before promoting to wiki/, verify:
[ ] All key findings have raw/ citations
[ ] No prompt-injection content quoted verbatim
[ ] Contradictions are surfaced, not silently resolved
[ ] Draft reads coherently as a standalone document
[ ] Length is appropriate (500–2000 words for most topics)
[ ] No hallucinated claims (every assertion traces to a source)
If any check fails, fix the draft before proceeding. If uncertain about a claim, mark it [UNVERIFIED] and note the gap.
Ask the user to review if:
After the review gate passes:
outputs/research-<slug>.md → wiki/topics/<slug>.md[[wikilinks]], update wiki/index.mdsynthesis.md)wiki/log.md:
## [YYYY-MM-DD] deep-research | <topic>
- Passes: academic, web, social, news
- Sources ingested: <count> files in raw/
- Wiki pages created/updated: <list>
- Open questions: <any unresolved gaps>
Notify wiki-research that deep research is complete. The orchestrator's post-process step (Step 3) handles wiki-ingest and wiki-maintainer — do not call them again here.
Deep research can expand indefinitely. Enforce these limits:
outputs/ (when used); passes executed within scope limits; synthesis draft in outputs/ with citations; review gate passed or user consulted.wiki/ (if any) updates wiki/index.md and wiki/log.md; orchestrator post-process not duplicated.skills/wiki-research/references/synthesis.md — synthesis structure guideskills/wiki-research/references/source-eval.md — source quality scoringllm-wiki integrations status and any llm-wiki line from Step 1 of this skill (from the vault root).