Append one day to data/progress.txt — the hand-written practice log that
site/build-review-plan.js compiles into the review plan's data on every build. It is the only copy. Nothing else in the repo records that a problem was
practised, so a line this skill writes badly is a day that quietly leaves the schedule.
Invocation: /lc-log <what you practised> — e.g. /lc-log 23 ok, 24 again!!, 25.
With no argument, ask what was practised today; do not invent a session.
Prime directives
- Never invent an attempt. Only what the user actually says they practised goes in.
Not what they were "supposed to" do, not the carried-over
todobucket's contents as though they were attempted, not a problem filed by/lc-pythonor/lc-javain the same session. Filing a solution is not practising it. - A number must start its own comma-chunk, or it is silently lost. This is the format's sharpest edge and it has already cost this log 93 attempts — see The two shapes that lose data.
- The annotation is the whole signal.
againbeatsokwhen a note says both, and the bang count sorts. Copy the user's own wording; do not normaliseagain!!down toagain. - Never downgrade or rewrite a previous day. The log is an append-only record of what happened. A problem that went badly on Tuesday still went badly on Tuesday.
- Run the gate before reporting done. A line the parser misreads produces no error — it produces a smaller schedule. Step 5 is not optional.
The format, as the parser actually reads it
One physical line per day, newest first at the top of the file, separated by two blank lines:
20260921: top 100 like(dp):, 32(ok), 300 | LCA:, 236(again!!) | others:, 354(2D LIS) |
| Piece | Rule |
|---|---|
YYYYMMDD: |
must be a real calendar date — an impossible one is warned about and the whole day is dropped |
| |
groups the day into sessions. Cosmetic to the parser (it becomes a comma), structural to the reader |
, |
the real separator. A . also splits, but that is typo-tolerance, not a convention — write commas |
123(note) |
the annotation. Commas and nested parens inside it are safe: 70(ok*, o(1) space!!) is one entry |
topo_sort, weekly_331 |
named drills. Parsed and ignored — they have no LC number to schedule, which is fine and intended |
a line starting --- |
a separator that ends the current day. Do not write one mid-session |
The annotation vocabulary, in the order the classifier tries it — again, then todo,
then ok:
| Written | Classified | Why it matters |
|---|---|---|
141(ok) |
ok | graduated this pass |
994(ok*) |
ok | the * is the user's own shading, carried through as the note |
128(again!!!) |
again, emphasis 3 | bangs are counted anywhere in the note and sort it above a bare again |
44(again !!) |
again, emphasis 2 | spaces inside the note are fine |
139(ok, but again) |
again | again is checked first, on purpose |
79(todo) |
todo | intended, not attempted — still records the intent |
300 |
none | practised, no verdict given. Legitimate; do not invent one |
The two shapes that lose data
Both are already in the log, both parse without a single warning, and both are the reason this skill exists rather than a habit.
A number sharing a chunk with its bucket label
Each comma-chunk used to have to start with the digits. A chunk that starts with a label looked exactly like a named drill to the site's parser, so until Sep 2026 it was dropped whole — taking the number with it:
20260920: top 100 like(linked list): 23, 24(again!!!), 25 | ...
^^ LC 23 was dropped. 24 and 25 survived.
93 attempts across the log sit in a chunk shaped like that. The two readers of the log
used to disagree about them — script/suggest_review.py strips the label first
(_strip_label), and site/build-review-plan.js did not until Sep 2026, when it gained the
same stripLabel because the roadmap stamps its done state from what that parser reads. Both
read the chunk now; a reader that does not exist yet will not, so the shape is still avoided.
The fix is one character — end the label with a comma so the number starts its own chunk, which no reader has to be taught:
20260920: top 100 like(linked list):, 23, 24(again!!!), 25 | ...
That is the shape this skill writes. It reads identically to a human and parses losslessly.
A number written after its description
^(\d+) matches the first digits in the chunk, so a description that itself starts with a
digit hands the parser the wrong number. Neither reader recovers this one:
2D LIS (354) -> logged as LC 2, note "D LIS" discarded. 16 entries in the log do this.
354(2D LIS) -> logged as LC 354, note "2D LIS". correct.
Rule: the LC number first, the description inside the parens.
Fixing the 16 number-after-description entries is a separate job from writing today's line — this skill does not rewrite past days (directive 4).
The steps
1. Read the last few days
head -20 data/progress.txt
Two things to take from them:
- the date. New entries go at the top of the file. Use today's real date — check it rather than assuming, and never post-date or back-date without being asked.
- the open buckets. A day's line carries forward the buckets still in play
(
top 100 like(dp):,LCA:,DP:,LC weekly:,others:). They are the user's own running agenda, not decoration. Carry forward the ones still open, keep their wording, and drop a bucket only when the user says it is finished.
2. Sort what the user said into buckets
Put each problem in the bucket it belongs to, matching the labels already in use. A problem
with no obvious bucket goes in others:. A new bucket is fine when the user names a new
theme — keep the label short and reuse it the next day.
Keep the user's own verdicts verbatim. If they said "24 fought me again", that is
24(again); if they said "again, three times now", that is 24(again!!!) only if they gave
three bangs or said so — do not inflate the emphasis.
3. Write the line
YYYYMMDD: <label>:, <n>(<note>), <n>, ... | <label>:, <n>(<note>) | others:, <n> |
- every label ends
:,(directive 2) - every problem is
<number>or<number>(<note>), number first (directive 2) - sessions separated by
|, and the line ends with a trailing|like its neighbours - one physical line, however long — the parser handles wrapping, but the log writes one line
- two blank lines before the entry below it
Insert at the top of the file, above the current newest day.
4. Do not touch anything else
This skill writes exactly one new line in one file. Not README's status column — that is
/lc-again's
job and it has its own rules about what earns a promotion. Not a solution file. Not
data/again_problems.txt.
5. Run the gate
The parser never errors on a badly shaped line — it just reads fewer problems. So check the numbers, not the exit code:
node site/build-review-plan.js
npm test --prefix site
build-review-plan.js prints the day count, the problem count and the again count, and
site/test/build-review-plan.test.js asserts the real log parses with zero warnings.
Then confirm today's line round-trips — every number the user gave, and nothing extra:
node -e "
const p = require('./site/build-review-plan.js');
const { payload } = p.buildPayload(require('fs').readFileSync('data/progress.txt','utf8'));
const d = payload.days[payload.days.length - 1];
console.log(d.date, d.items.map(i => i.id + (i.note ? '(' + i.note + ')' : '')).join(' '));"
If a number the user gave is missing from that output, the line is wrong — fix the line, not the expectation.
6. Report
Close with the date written, the problems in it grouped by bucket, the round-trip output, and anything inferred: a bucket chosen for a problem the user did not place, a carried-over bucket dropped because it looked finished, a verdict left blank because none was given.
Do not
- ❌ invent an attempt, or promote a
todointo an attempt (directive 1) - ❌ write a label and a number in one chunk —
others: 678(todo)loses LC 678 - ❌ write
2D LIS (354)— it logs LC 2 - ❌ edit or "tidy" a previous day's line
- ❌ inflate or flatten the user's emphasis —
again!!is notagain - ❌ touch README's status column,
data/again_problems.txt, or any solution file - ❌ report done without the round-trip in step 5
- ❌ commit or push unless asked — the practice log gets its own commit
Worked example
/lc-log 23 ok, 24 again!!!, 25, and 138 came back again on 2026-09-21:
| Step | What it produced |
|---|---|
| 1 | read the top of the log: yesterday's open buckets were top 100 like(linked list), top 100 like(heap), top 100 like(dp), others, DP |
| 2 | 23/24/25/138 all sort into top 100 like(linked list); the rest of yesterday's buckets carry forward untouched |
| 3 | 20260921: top 100 like(linked list):, 23(ok), 24(again!!!), 25, 138(again) | top 100 like(heap):, 295 | ... written at the top |
| 4 | README, again_problems.txt and the solution files left alone |
| 5 | build-review-plan.js → 822 days, 799 problems; round-trip prints 20260921 23(ok) 24(again!!!) 25 138(again) — all four present |
| 6 | flagged: 25 recorded with no verdict because none was given; LC weekly: dropped, as the user said 2289 was finally done |