Literature review
Use when surveying a field or choosing among competing methods — searches and critically appraises sources, maps the state of the art and gaps, and produces a justified ranked comparison of methods with the oracles an implementation must match.From its SKILL.md
npx -y skills add aaronbao1/scienceskills --skill literature-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.2 KB, as published. Nobody here has run it
Literature Review
Find what is known, judge it critically, and decide which method to use — with the evidence to defend the choice. The output drives implementation, so it must be precise.
Produce a method dossier
- Search. Use
deep-researchto fan out across sources; cover seminal work, recent state of the art, and dissenting results. Record queries so the search is reproducible. - Appraise. For each source, note its claim, evidence quality, sample or benchmark, and limitations. Distinguish strong results from weak or unreplicated ones.
- Map. Synthesize the state of the art and the gaps — what is unsolved or contested.
- Compare and rank. Lay competing methods side by side against explicit criteria (accuracy, assumptions, cost, robustness, fit to your problem). Rank them and state the recommended method with the reason it wins for this context.
- Extract oracles. Pull the equations, algorithms, hyperparameters, and reported
numbers the chosen method specifies — these become the test oracles for
faithful-implementation.
Output: a method dossier — appraised sources, a state-of-the-art and gap map, a ranked comparison with the decision, and the extracted oracles.
Composes with
deep-researchfor the search and adversarial verification of claims.- The dossier's oracles feed
faithful-implementation; its decision feedsresearch-design. - For argument-driven or primary-source work, defer source criticism to
argumentation-and-sources.
Domain lenses
- ML/AI: prefer results with released code and reported variance; note benchmark leakage.
- Computational science: capture the governing equations and validation cases.
- Data science: capture identification assumptions and estimator definitions.
Red flags (stop)
- A method chosen by popularity or recency rather than fit and evidence.
- Citing a claim without recording the evidence behind it.
- No gap analysis — a review that only summarizes cannot justify new work.
Consult
Before starting, read your playbook of learned heuristics:
skills/skill-forge/insights/literature-review/playbook.md (skip if absent).
Capture (run at session end)
When you finish a task that used this skill, record what happened so skill-forge can learn:
- Snapshot this session:
python3 -m eval.harness.capture snapshot literature-review - Reflect against the five signals — user correction/redo · abandonment · approval · hard failure
(tool/hook errors) · self-assessed struggle — then append ONE generalized insight (no
project-specifics): pipe a JSON record to
python3 -m eval.harness.capture insight literature-reviewwith fields{ts, session_id, context, signals, what_worked, what_failed, lesson, proposed_edit?, confidence}.
Record the lesson, not the incident. If a specific line of this SKILL.md caused a failure, include a
proposed_edit of {old, new, reason}.