Agent skills
The repo ships its own recipes for coding agents under .claude/skills/ —
plain markdown, no dependencies. Each one does a single job this repo has got wrong by
hand before: filing a solution, logging a session, graduating an AGAIN,
picking what to drill next, keeping the site's data valid.
› /lc-python 239 Sliding_Window slug sliding-window-maximum file Sliding_Window/sliding_window_maximum.py smoke 3/3 docstring examples pass ✓ README row inserted under ## Sliding Window › /lc-log 239 ok, 76 again!! ✓ 20260930 appended to data/progress.txt › /l3-core next 5 1. 76 again Sliding Window 2. 124 — Tree …
How they fit together
The skills are not a toolbox so much as a loop. Each one writes exactly one record and names the one it leaves for the next — so a solution, a practice attempt and a mastery verdict never end up in each other's file.
↺ tomorrow, /l3-core reads the log you just wrote
Directory
Grouped by what the skill touches: your solutions, your practice record, or this site. Each
card links to the skill's own page — the failure modes it exists to prevent, its steps,
and a worked example — and to the SKILL.md an agent actually reads.
Finds the problem's real slug, writes the house layout (docstring → V0 → IDEA → time/space), smoke-tests it against the docstring's own examples, and inserts the README row in the right table set.
/lc-python 4038 Hash_tableThe package its pattern owns, a clean javadoc header, // V markers that match the method names, a compile and a run — and the [Java] link added to the row the problem already has.
/lc-java 25 LinkedListFiles what a problem taught you into the sheet whose Scope line owns it — as an example, a variation, a new pattern or a whole sheet — then fixes the anchors and 中文 sections the edit moved.
/lc-cheatsheet 1650 into binary_tree as variationScores a solution the way an interviewer does on the six-point scale (SH / H / LH / LNH / NH / SNH), writes the debrief packet, names the line that sets the complexity, and says what to drill next.
/lc-coach mock interview me on LC 239Appends today to the practice log in the shape the review plan's parser actually reads — labels comma-terminated, annotations verbatim — so no problem number is silently dropped.
/lc-log 23 ok, 24 again!!, 25Moves README's status cell after a re-solve. Promotes only on all four: re-derived unaided, invariant stated, complexity line named, edges handled — and always keeps the star run.
/lc-again 128Blind 75 plus the NeetCode 150 problems marked MUST, held fixed so the number means something: where each stands in the log, the ok share, and the next five to drill.
/l3-core next 5Searches recent LeetCode Discuss posts by keyword, reads the ones that describe a real question, maps each to an LC number, checks it against the practice log, and writes a 繁體中文 summary doc.
/lc-discussion google l3Writes a visualizer against the contract the other 36 obey — the shared palette, the canvas wrapper, the structured step trace, and the repaint a theme switch depends on.
/lc-algo-demo monotonic-stackAdds a roadmap topic or a quiz question so the build-time validators accept it first time — no implied edge, no duplicate id, answers that parse as complexity expressions.
/lc-site-data roadmap monotonic-stackWorks the 繁體中文 overlay one section at a time — prose only, every code fence kept as a marker, parked translations adapted rather than started from nothing.
/lc-zh-translate faq/javaFiles an interview question into the FAQ its Scope line owns, under the numbered section it belongs to — and writes the Chinese half in the same change.
/lc-faq-add kafka rebalancingNo skill matches that. Try readme, log or translate.
lc- recipes and have no page of their own:
Install
Inside a clone of this repo there is nothing to install — the skills are already at
.claude/skills/ and Claude Code loads them. To use them anywhere else, build the
command here.
Copied into your user-level skills folder, so every repo gets them.
git clone --depth 1 https://github.com/yennanliu/CS_basics.git /tmp/cs_basics
mkdir -p ~/.claude/skills
cp -r /tmp/cs_basics/.claude/skills/{lc-python,lc-coach,lc-log} ~/.claude/skills/Under the hood
No runtime, no server. A skill is a directory an agent reads, and this repo checks every one of them in CI.
SKILL.md carries a frontmatter description the agent matches your request against, an Invocation line, numbered steps, a Do not list and a worked example.
Renaming .claude/skills/lc-log/ renames /lc-log. Plain English works too — “log today's practice” loads the same skill.
script/check_skills.py --install --run installs each skill into a temp dir, re-checks it alone, and runs the commands it tells an agent to type.