cf-recommend

Know exactly what to practice next on Codeforces.

A fast, local-first CLI that analyses your Codeforces submission history and recommends the best problems to solve next — laser-focused on your weak topics.

npm install -g cf-recommend

No browser. No account linking. No C++ compiler. Just your handle.

Why cf-recommend?

Every competitive programmer hits the same wall: "what should I solve next?" Existing tools are web-based and generic. cf-recommend is built for people who live in the terminal and want personalised, data-driven practice.

• Weak-topic targeting — it finds the tags you actually struggle with.
• Adaptive difficulty — recommendations centre on a solving level inferred from the problems you really solve, not a static rating.
• Explains itself — every suggestion comes with a reason.
• Local-first — cached in SQLite, instant, works offline.
• Zero-friction install — prebuilt binary via npm.

🎬 Demo

$ cf analyze

  tourist  —  616 problems solved across 683 attempted
  Adaptive solving level: ~1650 (from your solve history)

  Problem-Solving Breakdown  (success rate · solved/attempted)
    Math                  92%   291/315
    Greedy                90%   300/334
    DP                    84%   75/89
    Binary Search         82%   54/66
    Graphs                69%   11/16
    Divide And Conquer    63%   5/8     (weak)

  Weak areas (below 50%) ...

$ cf next --weak-topics-only --count 1

  1. 1167B  rating 1400  ·  Divide And Conquer [weak area]
     Lost Numbers
     Why: You're only 63% proficient in Divide And Conquer — focused practice on a weak area.
     For you: Medium  (feels like ~1475)   Est. 25-40 min
     https://codeforces.com/problemset/problem/1167/B

Install

npm install -g cf-recommend

Installs the cf command (plus a cf-recommend alias). Prebuilt binaries ship for Windows x64, Linux x64, and macOS x64 & arm64 — no compiler needed. Prefer to build it yourself? See Build from source.

Heads up: the command is cf. If you also use the Cloud Foundry CLI (also cf), use the cf-recommend alias instead.

Quick start

cf login <your-handle>     # download & cache your history (public data)
cf analyze                 # see your per-topic breakdown
cf next                    # get problem recommendations
cf weak-topics             # focus areas with insights
cf progress                # recent activity, improvements, streak

Commands

| Command | Description | |---|---| | cf login <handle> | Verify a handle and download its full history | | cf sync | Re-fetch submissions & the problem set | | cf analyze | Per-topic success-rate breakdown | | cf next [options] | Recommend problems to solve next | | cf weak-topics | Your weakest topics, with insights | | cf progress | Recent activity, improvements, and streak | | cf config [key value] | View or change settings (e.g. max-rating) | | cf help | Show usage |

next options

| Flag | Meaning | |---|---| | --count N | How many problems to show (default 10) | | --topic NAME | Filter by topic — dp, greedy, binsearch, graphs, … | | --difficulty A-B | Rating range, e.g. 1400-1600 (single value also works) | | --rating-range A-B | Alias for --difficulty | | --max-rating N | Cap this run's problem rating at N | | --weak-topics-only | Only problems that touch your weak topics |

Topic names are flexible: binsearch, binary-search, and binary search all resolve to the same Codeforces tag.

Set a permanent rating cap

cf config max-rating 1700   # cap every recommendation at 1700
cf config max-rating off    # remove the cap
cf config                   # show current settings

Precedence for next: explicit --difficulty / --rating-range → one-off --max-rating → saved config max-rating default.

Examples

cf next --weak-topics-only --count 10
cf next --topic dp --difficulty 1400-1600 --count 5
cf weak-topics
cf progress

How it works

Adaptive solving level

Recommendations centre on a solving level rather than your raw contest rating. Once you have at least 12 rated solves, it's the 80th percentile of the ratings you've solved, blended 50/50 with your Codeforces rating:

solving_level = round(0.5 * cf_rating + 0.5 * p80(solved problem ratings))

With less history it falls back to your Codeforces rating (or 1400 for a new account). Both cf analyze and cf next show it.

Proficiency

Submissions are aggregated per problem (not per submission), so retries and upsolves don't distort the numbers. A problem counts as solved if it ever got an OK verdict; in-progress submissions are ignored.

success_rate(topic) = solved problems in topic / attempted problems in topic

A topic needs at least 3 attempted problems before its rate is treated as reliable, and topics below 70% are considered weak.

Recommendation score

Each unsolved, rated problem gets a score in [0, 1]:

score = 0.40 * topic_weakness      // 1 - success_rate of its weakest tag
      + 0.30 * rating_fit          // Gaussian around your solving level
      + 0.20 * popularity          // log-scaled solve count (a proxy for "standard")
      + 0.10 * variety             // penalises tags you've drilled very recently

Per-user difficulty

A weak topic makes a problem feel harder, so the effective rating adds up to +200 for your weakest involved tag:

effective_rating = problem_rating + round(weakness * 200)

The Easy / Medium / Hard label and time estimate compare that effective rating to your solving level.

Data & caching

• Data comes from the public Codeforces API (user.info, user.status, problemset.problems), with automatic retry on rate limits.
• Cached in a SQLite database under %LOCALAPPDATA%\cf-recommend\ (Windows) or ~/.cf-recommend/ (Unix).
• analyze, next, weak-topics, and progress read only from the cache, so they work offline. Run cf sync to refresh.

Build from source

Only needed if you're not installing via npm. Requires a C++17 compiler plus the libcurl and sqlite3 development libraries. The resulting binary is cf (cf.exe on Windows).

pacman -S mingw-w64-ucrt-x86_64-gcc \
          mingw-w64-ucrt-x86_64-curl \
          mingw-w64-ucrt-x86_64-sqlite3 \
          mingw-w64-ucrt-x86_64-pkgconf

# from a UCRT64 shell (or with C:\msys64\ucrt64\bin on PATH):
mingw32-make

# Debian/Ubuntu:  sudo apt install g++ libcurl4-openssl-dev libsqlite3-dev
# macOS:          xcode-select --install
make
sudo cp cf /usr/local/bin/   # optional: put it on PATH

cmake -S . -B build && cmake --build build

Contributing

Contributions are very welcome — see CONTRIBUTING.md for dev setup, project layout, and guidelines. Good first areas: smarter ranking heuristics, new filters, and platform testing.

Roadmap

• [ ] Friend / rival comparison (cf compare <handle>)
• [ ] Contest-division awareness in filtering
• [ ] Per-tag "learning path" ordering
• [ ] Export recommendations to a file

License

MIT © nathandemoss