Literature Review Agent

Name: Literature Review Agent
Author: Ar9av

Step 3 of the PaperOrchestra pipeline: discover candidate papers via web search, verify them through Semantic Scholar, cross-corroborate…

You say

Install this skill Read the source first Free Written by Ar9av · unverified publisher

Context cost: 35.7k tokensestimated from the bundle, loaded when it triggers
Bundle: 24 files · 142.7 kB15 scripts among them — read before you run
Licence: MITfree to use
Last change: —no release on file
Servers it uses: Noneruns standalone

What it does

Step 3 of the PaperOrchestra pipeline (arXiv:2604.05018). Execute the literature search strategy from outline.json — discover candidate papers via web search, verify them through Semantic Scholar (Levenshtein > 70 fuzzy title match, temporal cutoff, dedup by paperId), cross-corroborate against Crossref + OpenAlex to flag hallucinated citations, build a BibTeX file, and draft Introduction + Related Work using ≥90% of the verified pool. Runs in parallel with the plotting-agent.

Installed, it changes the agent in these ways.

What this skill changes about the agent is not written down here yet. The listing was collected from its source, and the description is in its own SKILL.md.

Workflow

Runs a procedure end to end.

literature-reviewcitationssemantic-scholarbibliography

Filed under

Research

The skill itself

This is the whole product. A skill is instructions the model reads, so there is nothing behind the listing you cannot see first — the front matter loads with every session, and the body below it loads when the skill triggers.

SKILL.md19.9 kB · 454 lines

--- name: literature-review-agent description: Step 3 of the PaperOrchestra pipeline (arXiv:2604.05018). Execute the literature search strategy from outline.json — discover candidate papers via web search, verify them through Semantic Scholar (Levenshtein > 70 fuzzy title match, temporal cutoff, dedup by paperId), cross-corroborate against Crossref + OpenAlex to flag hallucinated citations, build a BibTeX file, and draft Introduction + Related Work using ≥90% of the verified pool. Runs in parallel with the plotting-agent. TRIGGER when the orchestrator delegates Step 3 or when the user asks to "find citations for my paper", "draft the related work", or "build the bibliography". ---

6# Literature Review Agent (Step 3)

8Faithful implementation of the Hybrid Literature Agent from PaperOrchestra

9(Song et al., 2026, arXiv:2604.05018, §4 Step 3, App. D.3, App. F.1 p.46).

11**Cost: ~20–30 LLM calls.** This is one of the two longest steps (the other is

12plotting). Wall-time floor is set by Semantic Scholar's 1 QPS verification

13limit.

15## Inputs

17- workspace/outline.json — specifically intro_related_work_plan with the

18 Introduction search directions and the 2-4 Related Work methodology

19 clusters

20- workspace/inputs/conference_guidelines.md — used to derive cutoff_date

21- workspace/inputs/idea.md, workspace/inputs/experimental_log.md — for

22 framing the Intro and grounding the Related Work positioning

24## Outputs

26- workspace/citation_pool.json — verified Semantic Scholar metadata for

27 every paper that survived verification

28- workspace/refs.bib — BibTeX file generated from the verified pool

29- workspace/drafts/intro_relwork.tex — drafted Introduction and Related

30 Work sections, written into the template, with the rest of the template

31 preserved verbatim

33## Two-phase pipeline (App. D.3)

35```

36PHASE 1 — Parallel Candidate Discovery

37 For each search direction in introduction_strategy.search_directions:

38 For each limitation_search_query in each related_work cluster:

39 - Use the host's web search tool to discover up to ~10 candidate papers.

40 - Run up to 10 discovery queries in parallel (host-permitting).

41 - Collect (title, snippet, url) tuples — no verification yet.

42 → PRE-DEDUP before Phase 2 (see Step 1.5 below)

44PHASE 2 — Sequential Citation Verification (1 QPS, with cache)

45 For each candidate (after pre-dedup), sequentially:

46 0. Check s2_cache.json first (scripts/s2_cache.py --check).

47 If HIT: use cached response, skip live S2 call. No throttle needed.

48 If MISS: proceed with live request below.

49 1. Query Semantic Scholar by title:

50 GET https://api.semanticscholar.org/graph/v1/paper/search?query=<title>

51 &fields=title,abstract,year,authors,venue,externalIds&limit=5

52 (Public endpoint, no key. Throttle to 1 QPS for live requests only.)

53 2. Store the S2 response in cache: s2_cache.py --store.

54 3. Pick the top hit. Check Levenshtein title ratio against the original

55 candidate title. If ratio < 70: discard.

56 4. Bonus: if year and venue exactly align with hints, add a +5 point

57 match-quality bonus.

58 5. Require: abstract is non-empty.

59 6. Require: paper.year (or month if known) strictly predates cutoff_date.

60 Months default to day-1: e.g., "October 2024" → 2024-10-01.

61 7. If all checks pass, add to verified pool.

62 After all candidates are verified, dedup by Semantic Scholar paperId.

63```

65The host agent does the LLM/web work; the deterministic helpers in scripts/

66do the math.

68## Step-by-step

70### 0. Derive cutoff_date

72Parse conference_guidelines.md for the submission deadline. The paper aligns

73research cutoff with venue submission deadline (App. D.1):

75| Venue | Cutoff |

76|---|---|

77| CVPR 2025 | Nov 2024 |

78| ICLR 2025 | Oct 2024 |

79| Other | One month before the stated submission deadline |

81Encode as YYYY-MM-DD. Months default to day-1 (e.g., 2024-10-01).

83### 1. Phase 1: Parallel Candidate Discovery

85From outline.json:

87- All introduction_strategy.search_directions (3-5 queries)

88- For each cluster in related_work_strategy.subsections:

89 - The cluster's sota_investigation_mission becomes a search query

90 - All limitation_search_queries (1-3 each)

92For each query, **use your host's web search tool** (e.g., WebSearch in

93Claude Code, @web in Cursor, the search tool in Antigravity). Collect the

94top ~10 candidates per query: title, abstract snippet, source URL.

96If your host supports parallel sub-tasks, fire up to 10 concurrent search

97queries. If not, run sequentially — slower but functionally equivalent.

99#### Optional: Exa as a Phase 1 backend

100

101If your host has no native web search, OR you want a research-paper-focused

102backend with better signal-to-noise, you can use [Exa](https://exa.ai) via

103the bundled scripts/exa_search.py helper. It is **opt-in** and reads

104EXA_API_KEY from the environment — the repo never commits a key.

105

106```bash

107export EXA_API_KEY="your-key-here" # get one at https://dashboard.exa.ai/

108python skills/literature-review-agent/scripts/exa_search.py \

109 --query "Sparse attention long context transformers" \

110 --num-results 15 \

111 --discovered-for "related_work[2.1]"

112```

113

114Output is a normalized candidate list ready to merge into

115raw_candidates.json. Phase 2 verification (Semantic Scholar fuzzy match,

116cutoff, dedup) is unchanged. See references/exa-search-cookbook.md for

117the full recipe, query patterns, cost estimates, and security notes.

118

119#### Optional: Tavily as a Phase 1 backend

120

121If your host has no native web search, OR you want an LLM-optimized search

122backend with high relevance scoring, you can use [Tavily](https://tavily.com)

123via the bundled scripts/tavily_search.py helper. It is **opt-in** and reads

124TAVILY_API_KEY from the environment — the repo never commits a key.

125

126```bash

127export TAVILY_API_KEY="tvly-your-key-here" # get one at https://app.tavily.com

128python skills/literature-review-agent/scripts/tavily_search.py \

129 --query "Sparse attention long context transformers" \

130 --num-results 15 \

131 --academic \

132 --discovered-for "related_work[2.1]"

133```

134

135Output is a normalized candidate list ready to merge into

136raw_candidates.json. Phase 2 verification (Semantic Scholar fuzzy match,

137cutoff, dedup) is unchanged. See references/tavily-search-cookbook.md for

138the full recipe, query patterns, cost estimates, and security notes.

139

140Combine all discovered candidates into a single working list. Tag each with

141the originating query ID so you can later attribute it to "intro" vs

142"related_work[i]".

143

144### 1.5. Pre-dedup before Phase 2

145

146**Always run this before starting Phase 2.** Multiple search queries routinely

147return the same papers (e.g., "Attention is All You Need" appears in almost

148every NLP discovery query). Verifying duplicates wastes 30-40% of S2 quota

149at 1 QPS.

150

151```bash

152python skills/literature-review-agent/scripts/pre_dedup_candidates.py \

153 --in workspace/raw_candidates.json \

154 --out workspace/deduped_candidates.json

155# Prints: "150 candidates → 97 unique (53 duplicates removed)"

156```

157

158Use workspace/deduped_candidates.json as input to Phase 2.

159

160### 2. Phase 2: Sequential Verification via Semantic Scholar (with cache)

161

162For each candidate in deduped_candidates.json, in **sequential** order:

163

164**Step A — check cache first** (no S2 call, no throttle needed):

165```bash

166python skills/literature-review-agent/scripts/s2_cache.py \

167 --cache workspace/cache/s2_cache.json \

168 --check "<candidate title>"

169# exit 0 + prints JSON → use cached response, skip Step B

170# exit 1 → proceed to Step B

171```

172

173**Step B — live S2 request** (cache MISS only, throttle to 1 QPS):

174

175**Preferred:** use the bundled scripts/s2_search.py helper — it handles

176auth, retries, and 429 back-off automatically:

177

178```bash

179python skills/literature-review-agent/scripts/s2_search.py \

180 --query "<URL-decoded candidate title>" --limit 5

181# If SEMANTIC_SCHOLAR_API_KEY is set the key is forwarded automatically.

182# If not, the public unauthenticated endpoint is used (≤1 QPS, still works).

183```

184

185Check whether the key is configured before starting Phase 2:

186

187```bash

188python skills/literature-review-agent/scripts/s2_search.py --check-key

189```

190

191**Fallback:** if you prefer your host's URL fetch tool, GET:

192```

193https://api.semanticscholar.org/graph/v1/paper/search?query=<URL-encoded title>&limit=5&fields=title,abstract,year,authors,venue,externalIds

194```

195Add header x-api-key: <SEMANTIC_SCHOLAR_API_KEY> if the env var is set.

196Be polite: ≤1 request per second for live requests. Cache hits are free.

197

198**Step C — store in cache** (after every successful live request):

199```bash

200python skills/literature-review-agent/scripts/s2_cache.py \

201 --cache workspace/cache/s2_cache.json \

202 --store "<candidate title>" \

203 --response '<full S2 JSON response>'

204```

205

206For the top hit:

207

208```bash

209python skills/literature-review-agent/scripts/levenshtein_match.py \

210 --candidate "Original candidate title" \

211 --found "S2 returned title"

212# prints integer 0-100. Discard if < 70.

213```

214

215Then check the temporal cutoff:

216

217```bash

218python skills/literature-review-agent/scripts/check_cutoff.py \

219 --paper-year 2024 \

220 --paper-month 9 \

221 --cutoff 2024-10-01

222# exit 0 if strictly predates, exit 1 if not

223```

224

225If both checks pass AND the abstract is non-empty, append the paper's full

226S2 metadata to the verified pool.

227

228### 3. Dedup and assemble the pool

229

230After all candidates are verified:

231

232```bash

233python skills/literature-review-agent/scripts/dedupe_by_id.py \

234 --in raw_pool.json \

235 --out workspace/citation_pool.json

236```

237

238The dedupe script keys on paperId (Semantic Scholar's internal unique ID),

239falling back to externalIds.DOI, then externalIds.ArXiv, then a

240normalized title.

241

242The script also computes and writes min_cite_paper_count =

243floor(0.9 * len(papers)) — the minimum number of papers the writing step

244must cite (the paper's ≥90% integration rule, App. D.3).

245

246**Immediately after dedupe_by_id.py**, validate and auto-fix the pool schema:

247

248```bash

249python skills/literature-review-agent/scripts/validate_pool.py \

250 --pool workspace/citation_pool.json --fix

251# Catches and fixes authors-as-strings, reports missing required fields.

252# Must pass before proceeding to Step 4.

253```

254

255### 3.5. Cross-index verification (Crossref + OpenAlex)

256

257Semantic Scholar is one index and can return a plausible record for a paper

258that does not exist, or attach wrong metadata. Re-check every S2-verified

259paper against two **independent** indices before building the bibliography —

260this is the practical defense against hallucinated citations leaking in.

261

262```bash

263# Optional but recommended: a polite-pool email gives faster, more reliable

264# service. The repo never commits an address.

265export PAPER_ORCHESTRA_MAILTO="you@example.com"

266

267python skills/literature-review-agent/scripts/cross_verify.py \

268 --pool workspace/citation_pool.json --inplace

269# Annotates each paper with a cross_verification field and writes

270# workspace/cross_verification_report.json.

271# exit 0 = all corroborated; exit 1 = WARN (something flagged or an index

272# was unreachable); exit 2 = usage error.

273```

274

275This is a **WARN gate, not a hard gate** (like validate_consistency.py): it

276flags suspicious citations but does not block the pipeline or delete anything.

277Review the low and conflict tiers in the report:

278

279- high — corroborated by ≥1 external index → keep.

280- medium — corroborated but year disagrees → keep, spot-check the year.

281- low — not found in Crossref or OpenAlex → **review by hand**. Note that

282 arXiv-only preprints (no DOI) are a common benign cause; low means

283 "could not corroborate," not "fabricated." S2 already confirmed it exists.

284- conflict — pool DOI disagrees with the external DOI → likely wrong record.

285

286Drop only the entries you genuinely cannot corroborate, then re-run

287dedupe_by_id.py onward. If both indices are unreachable (offline), the script

288degrades gracefully and the pipeline continues on S2 verification alone.

289

290See references/cross-index-verification.md for the full rationale, confidence

291tiers, and the arXiv false-positive note.

292

293### 4. Build the BibTeX file

294

295```bash

296python skills/literature-review-agent/scripts/bibtex_format.py \

297 --pool workspace/citation_pool.json \

298 --out workspace/refs.bib

299```

300

301The script generates citation keys deterministically from `firstauthor + year

302+ first significant word of title (e.g., vaswani2017attention`). It writes

303out only @article / @inproceedings / @misc entries — never invents

304fields. It also writes the canonical bibtex_key back into each paper record

305in citation_pool.json.

306

307**Immediately after bibtex_format.py**, sync keys in intro_relwork.tex:

308

309```bash

310python skills/literature-review-agent/scripts/sync_keys.py \

311 --pool workspace/citation_pool.json \

312 --tex workspace/drafts/intro_relwork.tex \

313 --inplace

314# Replaces every \cite{agent_key} with \cite{canonical_bibtex_key}.

315# Eliminates citation_coverage gate failures caused by key mismatch.

316```

317

318These two steps replace the manual Python snippets that were previously

319required. The pipeline is now:

320

321```

322dedupe_by_id → validate_pool --fix → cross_verify --inplace → bibtex_format → sync_keys

323```

324

325### 5. Draft Introduction + Related Work

326

327This is where you (the host agent) actually write text. Load the

328**verbatim Literature Review Agent prompt** at references/prompt.md.

329Substitute the template placeholders:

330

331| Placeholder | Value |

332|---|---|

333| intro_related_work_plan | full JSON object from outline.json |

334| project_idea | contents of idea.md |

335| project_experimental_log | contents of experimental_log.md |

336| citation_checklist | the BibTeX keys from refs.bib |

337| collected_papers | list of {key, title, abstract} from citation_pool.json |

338| paper_count | len(citation_pool.papers) |

339| min_cite_paper_count | from citation_pool.json |

340| cutoff_date | the date you derived in Step 0 |

341

342**Also prepend the Anti-Leakage Prompt** from

343../paper-orchestra/references/anti-leakage-prompt.md.

344

345Run your LLM with the combined prompt against template.tex. The agent's

346job is to fill in the empty Introduction and Related Work sections of the

347template **and leave everything else untouched**. Output: the full

348template.tex with those two sections filled. Save to

349workspace/drafts/intro_relwork.tex.

350

351### 5b. Append §2 to research_brief.md

352

353After intro_relwork.tex is drafted and before the citation coverage check,

354append §2 to workspace/research_brief.md (see skills/shared/research_brief_template.md).

355

356Template:

357

358```markdown

359## §2 · Literature Landscape

360_Written by: literature-review-agent, Step 3_

361

362**What the literature says about the core claim:** <2-3 sentence synthesis>

363

364**Strongest prior work (must address in the paper):**

365- <bibtex_key>: <why this is the strongest comparator or predecessor>

366

367**Gaps confirmed by the literature:** <list>

368

369**Baseline comparisons — verification status:**

370| Baseline | In citation_pool? | Confidence tier |

371|---|---|---|

372

373**Related Work cluster coverage:**

374| Cluster | Papers found | Notes |

375|---|---|---|

376

377**Anything the section-writing agent should know:** <important context>

378```

379

380This synthesises what was actually found — not what the outline assumed.

381

382### 6. Verify ≥90% citation coverage

383

384```bash

385python skills/literature-review-agent/scripts/citation_coverage.py \

386 --tex workspace/drafts/intro_relwork.tex \

387 --pool workspace/citation_pool.json

388# exit 0 if ≥90% of pool is cited; exit 1 otherwise

389```

390

391If the gate fails, re-prompt the writing step explicitly listing the missing

392keys and asking the agent to integrate them where contextually appropriate.

393

394## Critical rules from the prompt

395

396These are excerpted from references/prompt.md. The host agent MUST honor

397them on the writing call:

398

399- **Cite ONLY from collected_papers.** Never invent BibTeX keys, never

400 reference papers not in the pool.

401- **Cite at least min_cite_paper_count of them** in Intro + Related Work

402 combined.

403- **TIMELINE RULE**: Do not treat any papers published after cutoff_date

404 as prior baselines to beat. They are concurrent work only.

405- **EVALUATION RULE**: Do not claim our method beats / achieves SOTA over a

406 specific cited paper UNLESS that paper is explicitly evaluated against in

407 experimental_log.md. Frame other recent papers strictly as concurrent,

408 orthogonal, or conceptual work.

409- **Output format**: return the full code for the updated template.tex,

410 with the two empty sections (Introduction and Related Work) filled in,

411 and **all the other code** (packages, styles, other sections) **identical

412 to the original** template.tex.

413- Wrap output in ``` `latex ... ` ``` fences.

414- Do not change \usepackage[capitalize]{cleveref} to cleverref (there is

415 no cleverref.sty).

416

417## Degraded mode (no web search)

418

419If your host has no web search tool, switch to degraded mode:

420

4211. If the user has placed a pre-built workspace/inputs/refs.bib in the

422 workspace, load it directly into workspace/refs.bib and skip Phase 1

423 and Phase 2.

4242. Otherwise, emit workspace/drafts/intro_relwork.tex containing the

425 template with two TODO markers in the Intro and Related Work sections,

426 and tell the user the pipeline cannot complete Step 3 without web search.

427

428## Resources

429

430- references/prompt.md — verbatim Literature Review Agent prompt from App. F.1

431- references/discovery-pipeline.md — Phase 1 + Phase 2 explained in detail

432- references/verification-rules.md — Levenshtein cutoff, year alignment, dedup

433- references/citation-density-rule.md — the ≥90% integration rule

434- references/s2-api-cookbook.md — Semantic Scholar URLs, fields, rate limits

435- references/cross-index-verification.md — Crossref + OpenAlex corroboration, confidence tiers, arXiv false-positive note

436- references/exa-search-cookbook.md — optional Exa backend for Phase 1 (research-paper-focused web search)

437- references/tavily-search-cookbook.md — optional Tavily backend for Phase 1 (LLM-optimized web search)

438- scripts/pre_dedup_candidates.py — **NEW** dedup Phase 1 candidates before Phase 2 (saves 30-40% S2 quota)

439- scripts/s2_cache.py — **NEW** persistent S2 response cache (eliminates re-verification on re-runs)

440- scripts/validate_pool.py — **NEW** validate & auto-fix citation_pool.json schema (authors format)

441- scripts/sync_keys.py — **NEW** sync cite keys in .tex with canonical bibtex_keys after bibtex_format.py

442- scripts/levenshtein_match.py — fuzzy title match (ratio > 70)

443- scripts/check_cutoff.py — date cmp w/ month → day-1 default

444- scripts/dedupe_by_id.py — dedup verified pool by S2 paperId

445- scripts/bibtex_format.py — build refs.bib from JSON pool

446- scripts/citation_coverage.py — ≥90% citation coverage gate

447- scripts/s2_search.py — **NEW** Semantic Scholar title-search helper; reads SEMANTIC_SCHOLAR_API_KEY from env (optional — falls back to unauthenticated)

448- scripts/exa_search.py — optional Exa Phase 1 backend (reads EXA_API_KEY from env)

449- scripts/tavily_search.py — optional Tavily Phase 1 backend (reads TAVILY_API_KEY from env)

450- scripts/crossref_client.py — **NEW** Crossref title/DOI lookup for cross-index corroboration (no key; reads CROSSREF_MAILTO / PAPER_ORCHESTRA_MAILTO)

451- scripts/openalex_client.py — **NEW** OpenAlex title/DOI lookup for cross-index corroboration (no key; reads OPENALEX_MAILTO / PAPER_ORCHESTRA_MAILTO)

452- scripts/cross_verify.py — **NEW** cross-corroborate the S2-verified pool against Crossref + OpenAlex; flags hallucinated citations (WARN gate)

453- skills/shared/research_brief_template.md — **NEW** §2 schema; append after intro_relwork.tex is drafted

454

In the file

SKILL.md2,555 words

Files24

LicenceMIT

Why you can read it

Nothing in a skill executes. The client loads the text and the model follows it, so a skill can be audited the way a runbook is — by reading it.

What it costs in context

Skills are not billed by the call. They are paid for in context: every token the instructions occupy is a token your code, your diff and your conversation cannot use. Here is what this one takes and when it takes it.

≈170

always loaded

The name and description, so the model knows the skill exists and when to reach for it.

35,505

on trigger

The instruction body and 23 supporting files, read only when the skill fires.

17.8%

of a 200k window

Ten skills this size would take about 178% of the window before you open a file.

050k100k150k200k context window

35.7k tokens, estimated from the bundle at four bytes to the token, held for the rest of the session once it triggers. Heavy. Teams tend to install this one per project rather than globally, and load it only when the job comes up.

Servers bill, skills cost

A server charges by the month. A skill charges once per session, in context, and then keeps charging it for as long as the session lives.

Before and after

The same question, put to the same model twice: once as it comes, and once with these instructions loaded.

No worked example has been published for this skill yet.

Adoption

Installsnone yet

Ratingno reviews yet

The procedure it runs

The procedure has not been published here. It is in the skill’s own SKILL.md, which its author has not sent to the marketplace yet.

Prose, not code

These steps are written for a model to follow, not executed by a runtime. It can still be told to skip one, and it will say so when it does.

Servers it uses

None. This skill calls no MCP servers at all.

Everything it needs is in the instructions, so it works in a project with nothing connected — the model reads the file and changes how it works with what it can already reach.

It writes no files and reaches no network. All it changes is how the model reasons and writes.

What it asks for

Writes filesno

Network accessno

Read from the allowed-tools line of this skill’s own SKILL.md. A skill grants no permissions of its own — it can only ask for tools your client already has.

What it will not do

Every skill is narrow, and the useful ones say where they stop. These are the jobs this one is the wrong tool for.

What this skill is not for has not been published here. Nothing is implied by that: it is a section the author has not filled in.

If this is not it

15 other skills in Research

What is in the bundle

24 files, 142.7 kB on disk. Mostly text — the instructions the model reads — with 15 scripts in it that your client would run only if the instructions tell it to.

SKILL.md19.9 kB
references/citation-density-rule.md2.9 kB
references/cross-index-verification.md4.6 kB
references/discovery-pipeline.md4.9 kB
references/exa-search-cookbook.md9.3 kB
references/prompt.md3.3 kB
references/s2-api-cookbook.md4.2 kB
references/tavily-search-cookbook.md10.1 kB
references/verification-rules.md5.5 kB
scripts/bibtex_format.py5.3 kB
scripts/check_cutoff.py2.3 kB
scripts/citation_coverage.py3.2 kB
scripts/cross_verify.py12.6 kB
scripts/crossref_client.py6.1 kB
scripts/dedupe_by_id.py3.2 kB
scripts/exa_search.py6.0 kB
scripts/levenshtein_match.py2.2 kB
scripts/openalex_client.py6.1 kB
scripts/pre_dedup_candidates.py5.0 kB
scripts/s2_cache.py3.6 kB
scripts/s2_search.py7.2 kB
scripts/sync_keys.py4.1 kB
scripts/tavily_search.py6.2 kB
scripts/validate_pool.py4.9 kB

What is not in it

A skill installs nothing and depends on nothing: it is a folder your client reads. This one carries 15 scripts beside the text, so the bundle is 24 files you can review in full before installing. The MIT licence covers the templates and examples as well as the instructions.

Install

Installing copies the bundle into your project. Nothing runs at install time — the files sit on disk until the model reads them.

# Literature Review Agent · 35.7k tokens when loaded
npx mcprush@latest skill add ar9av/literature-review-agent

Writes to .claude/skills/literature-review-agent/ in the current project. Add --global to put it in your home directory instead, for every project.

Which clients pick it up on their own

A skill is a folder of text. A client with a skills folder reads it without being told; everywhere else the same text works, it is just handed to the model rather than found.

Claude Code.claude/skills/

Claude Desktop

ChatGPT

Cursor.cursor/skills/

VS Code.github/skills/

Codex CLI.agents/skills/

Gemini CLI.gemini/skills/

Grok.grok/skills/

Zed.agents/skills/

Windsurf.windsurf/skills/

Agent SDK.claude/skills/

HTTP / API

This release

Versionnot versioned

Publishedno release date on file

PriceFree

Referencear9av/literature-review-agent

Versions

Its author publishes no version number, so there is nothing here to pin to: what you install is the folder as it stands today. Instructions change more often than APIs do — a skill can be rewritten entirely without anything it depends on moving.

No earlier releases have been published to the marketplace.

Pinning

Nothing to pin to: this skill carries no version number of its own. What you install is what the folder holds on the day you install it.

Reviews

—no reviews yet · no installs yet

Nobody has reviewed this skill yet. The rating is the mean of the reviews written here, so there is none until somebody writes the first.

Who can post

Only accounts that have had the skill installed for fourteen days, so a review is written after living with it rather than after reading it. Publishers may reply once.

Who wrote it

Ar9av

Publishes on mcprush.

0 servers listed1 skill listednot claimed

Profile

Publisher

Servers0

Claim this skill