agentleFS
Xuzhen-LiXuzhen-Li/literature-review-skills/skills/review-project-setup/SKILL.md

review-project-setup

Skill2 starsChanged 35 days agoMITView on GitHub

Creates the review project folder, interviews the user for topic, field, review type, audience, and timeline, and writes project.yaml plus LOG.md. Trigger on "start a new review on X", "new literature review", or "set up the review project". Do not use it to choose the final angle, write a search, or draft the manuscript.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 3 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Ask, in one message, for: working title.

    line 33
  2. Very specific:

    Create every directory below, plus 04-library/uploads/.

    line 39
  3. Very specific:

    Write LOG.md exactly as:

    line 64
  4. Very specific:

    Do not set topic_lock. The working title is not the locked angle.

    line 75
  5. Very specific:

    Every gate key exists and is YAML false, not the string "false".

    line 89
  6. Distinctive:

    No search string, hit count, or citation appears in the yaml.

    line 93
  7. Distinctive:

    Do not invent a journal, an impact factor, or a sample of papers.

    line 98
  8. Distinctive:

    Do not ask for, type, or store a password, captcha, or second factor.

    line 99
  9. Very specific:

    Do not change narrative_core_cap after the project file is confirmed.

    line 100
  10. Very specific:

    Do not delete a project that already has a LOG.md with decisions.

    line 101
  11. Very specific:

    Write only inside <project>/.

    line 102

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Review project setup

Opening

On first use in a project, say this verbatim:

I am review project setup. I create the project folder and project.yaml from your answers. You do not need a library login: I search free databases and download open-access PDFs myself, and I only ask you for the few essential papers I cannot get.

When to use

No project.yaml exists yet, or the user asked to start a review. This skill builds the empty folder layout and records the interview. It does not lock the topic.

Inputs (paths) / Preconditions

  • A parent directory the user names. Create <project>/ inside it. Do not scatter files outside that folder.
  • The user's answers. If an answer is missing, ask once, then stop. Do not fill the gap from general knowledge.
  • Copy the key names from references/project.yaml.example.

Procedure (numbered steps)

  1. Do not ask about library logins. The default setting is download_access: open_access: the agent searches the free APIs (OpenAlex, Europe PMC, PubMed, Semantic Scholar, and arXiv; Crossref on request; bioRxiv and medRxiv preprints come in through Europe PMC) and downloads open-access PDFs itself. This needs no browser homepage and no gate.
  2. Optional extras (only if the user brings them up or asks for Web of Science, Scopus, or their Zotero library): set use_subscription_databases: true or use_zotero: true as requested. For subscription databases, open https://www.webofscience.com/ in the browser and let the user sign in (never enter a password, captcha, or second factor yourself). Then ask them to reply with I have confirmed institutional login (sets download_access: institutional) or I have no download permission on this computer (sets download_access: unavailable). Verify the reply with python3 <suite>/skills/review-orchestrator/scripts/check_gate_phrase.py --gate institutional_access --reply "<user line>" (or --gate no_download_permission). On a match, set gates.institutional_access to true. If the user never sends the sentence, keep open_access and move on; no later step waits for this gate.
  3. Ask, in one message, for: working title; field; review type (narrative, systematic, scoping, or umbrella); audience (who will read it); timeline; target journal or undecided; whether PubMed will be searched; and the project directory path.
  4. Restate the four review types in one line each before accepting the answer if the user is unsure:
    • narrative: a scoped argument, not an exhaustive search; the reading set is capped by narrative_core_cap
    • systematic: a pre-specified question, search, and eligibility screen (PRISMA 2020)
    • scoping: how much and what kind of evidence exists (JBI PCC; report with PRISMA-ScR)
    • umbrella: the included records are reviews, not primary studies
  5. Create every directory below, plus 04-library/uploads/. Create LOG.md and project.yaml. Do not create evidence/evidence_table.csv here; paper-deep-reading writes it with the header in shared/evidence_table_schema.md.
<project>/
  project.yaml
  LOG.md
  00-scope/
  01-search/
  02-records/
  03-screening/
  04-library/
  05-reading/
  evidence/
  06-exemplars/
  07-outline/
  08-figures/
  09-draft/
  10-polish/
  11-journal/
  12-permissions/
  13-submission/
  14-revision/
  1. Write project.yaml from references/project.yaml.example. Set narrative_core_cap to 500 unless the user already typed a different positive integer in this interview. Do not change that number later. Set search_pubmed from their answer. Set download_access: open_access, use_subscription_databases: false, and use_zotero: false unless step 2 changed them. Set every gate to boolean false, except institutional_access when step 2 passed. Use today's date for created and updated. Set language: en. Quote any string that contains a colon.
  2. Write LOG.md exactly as:
# Decision log

Append-only. One decision per line. Do not delete lines.

YYYY-MM-DD | review-project-setup | project initialized | agent
  1. Show the user the yaml. Ask: "Confirm this project file? Reply yes or no." On yes, append YYYY-MM-DD | review-project-setup | project_confirmed | human. On no, edit only the fields they correct, except narrative_core_cap and the gate keys, and ask again.
  2. Do not set topic_lock. The working title is not the locked angle.

Human gate(s)

None is required. institutional_access is an optional extra, used only when the user wants subscription databases or institutional downloads (sentences in shared/gates.md). A checker script does not replace the user's sentence. Confirming the yaml is a checkpoint, not a lock. Topic lock belongs to review-scoping-and-gaps.

Outputs (exact paths + formats)

  • <project>/project.yaml — keys in references/project.yaml.example
  • <project>/LOG.md — header plus the lines above
  • The fourteen numbered directories, plus evidence/

Quality checklist

  • Every gate key exists and is YAML false, not the string "false". institutional_access is true only after the optional exact sentence.
  • narrative_core_cap is a positive integer, 500 unless the user set it in the interview.
  • download_access is open_access unless the optional gate set institutional or unavailable.
  • review_type is one of the four allowed words.
  • No search string, hit count, or citation appears in the yaml.
  • LOG.md has the initialized line and, after a yes, the confirmed line.

Guardrails

  • Do not invent a journal, an impact factor, or a sample of papers.
  • Do not ask for, type, or store a password, captcha, or second factor.
  • Do not change narrative_core_cap after the project file is confirmed.
  • Do not delete a project that already has a LOG.md with decisions. Ask the user before writing over project.yaml.
  • Write only inside <project>/.

Next step

review-scoping-and-gaps, after project_confirmed is in the log. Until then, stay here.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.