T29: Bílá kniha course catalogue (fit.catalog.*) #28

Open

dzenda wants to merge 4 commits from t29-bila-kniha into t27-fitwiki-connection

dzenda commented

2026-11-14 18:43:14 +01:00

Owner

T29: Bílá kniha (course catalogue)

Stacked on #27 (the part 2 plan lives there).

New MCP functions over the public faculty catalogue bk.fit.cvut.cz:

fit.catalog.course { code, lang? } returns credits, completion (Z/ZK/KZ), extent (e.g. 2P+1R+1C), semester, instructors and other facts, the course website, study-plan roles, and the text sections (annotation, lecture and seminar syllabus, literature, requirements), each capped at 4,000 chars. Codes may omit the plan suffix (BI-LA1 → BI-LA1.21, the newest suffix wins).
fit.catalog.search { query, lang?, limit? } matches course codes by prefix and course names word by word, accent-insensitively, with light stemming for Czech.
English or Czech pages (/en/, /cz/). The index (6 h) and up to 200 course pages are cached per isolate.
Safety:
- only digit detail paths taken from the index (00/…/p123.html) are fetched, on bk.fit.cvut.cz only;
- manual redirects, a 15 s timeout, an identified user agent;
- byte caps of 3 MB for the index and 500 KB for a page.
New shared helper src/lib/html-text.ts: entity decoding, HTML → text (keeping line structure), innerBlocks, stripTags and clip. It runs in linear time on malformed input, and T28 (FIT Wiki) reuses it.
Search keywords avoid "zkouška" and "zápočet" so exam-date and grade questions still go to Sirius and Grades.
PLAN.md: T28 is marked blocked until a FIT Wiki session is connected locally (Needs you).

Verified live against bk.fit.cvut.cz: BI-PA1 → 7 credits, Z,ZK, 2P+2R+2C, 5 sections, 23 study plans. The Czech BI-LA1 page has Czech section titles, and searching "databáze" finds the database courses.

Review (Astra)

Fixed [P2]: searching a course code was stemmed, so "BI-PA1" also matched BI-PA2. Code-like words now match code prefixes exactly; only name words are stemmed.
Fixed [P2]: quadratic regexes on malformed upstream HTML (e.g. 490k < without >) could exhaust the Worker's CPU budget. Every parser now uses linear indexOf scanners and bounded quantifiers, with timing tests for pathological pages.

Checks

lint, typecheck and test: 332 passing.

🤖 Generated with Claude Code

## T29: Bílá kniha (course catalogue) Stacked on #27 (the part 2 plan lives there). New MCP functions over the public faculty catalogue **bk.fit.cvut.cz**: - `fit.catalog.course { code, lang? }` returns credits, completion (Z/ZK/KZ), extent (e.g. 2P+1R+1C), semester, instructors and other facts, the course website, study-plan roles, and the text sections (annotation, lecture and seminar syllabus, literature, requirements), each capped at 4,000 chars. Codes may omit the plan suffix (`BI-LA1` → `BI-LA1.21`, the newest suffix wins). - `fit.catalog.search { query, lang?, limit? }` matches course codes by prefix and course names word by word, accent-insensitively, with light stemming for Czech. - English or Czech pages (`/en/`, `/cz/`). The index (6 h) and up to 200 course pages are cached per isolate. - Safety: - only digit detail paths taken from the index (`00/…/p123.html`) are fetched, on `bk.fit.cvut.cz` only; - manual redirects, a 15 s timeout, an identified user agent; - byte caps of 3 MB for the index and 500 KB for a page. - New shared helper `src/lib/html-text.ts`: entity decoding, HTML → text (keeping line structure), `innerBlocks`, `stripTags` and `clip`. It runs in **linear time** on malformed input, and T28 (FIT Wiki) reuses it. - Search keywords avoid "zkouška" and "zápočet" so exam-date and grade questions still go to Sirius and Grades. - PLAN.md: T28 is marked blocked until a FIT Wiki session is connected locally (Needs you). Verified live against bk.fit.cvut.cz: BI-PA1 → 7 credits, Z,ZK, 2P+2R+2C, 5 sections, 23 study plans. The Czech BI-LA1 page has Czech section titles, and searching "databáze" finds the database courses. ### Review (Astra) - Fixed [P2]: searching a course code was stemmed, so "BI-PA1" also matched BI-PA2. Code-like words now match code prefixes exactly; only name words are stemmed. - Fixed [P2]: quadratic regexes on malformed upstream HTML (e.g. 490k `<` without `>`) could exhaust the Worker's CPU budget. Every parser now uses linear `indexOf` scanners and bounded quantifiers, with timing tests for pathological pages. ### Checks lint, typecheck and test: 332 passing. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dzenda added 2 commits

2026-11-14 18:43:14 +01:00

T29: Bílá kniha provider (fit.catalog.course, fit.catalog.search) 9d707c8012

- bk.fit.cvut.cz index + course pages (en/cs): credits, completion, extent, instructors,
  study-plan roles, annotation/syllabus/literature/requirements (bounded), cached per isolate
- only digit detail paths from the index are fetched; manual redirects; identified UA
- shared src/lib/html-text.ts (entities, HTML → text, clip)
- PLAN: T28 blocked on a connected FIT Wiki session (Needs you)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

T29: linear-time HTML parsing for untrusted pages; exact course-code search (Astra review) 3b39be645a

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dzenda added 2 commits

2026-11-14 18:43:46 +01:00

biome: skip third-party HTML fixtures (test/fixtures/bk) 4eecaf74a0

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

PLAN: link T29 PR 81ec24d92b

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dzenda referenced this pull request

2026-11-14 18:45:54 +01:00

T33: docs + examples for FIT Wiki and the catalogue #29

This pull request can be merged automatically.

You are not authorized to merge this pull request.

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin t29-bila-kniha:t29-bila-kniha

git switch t29-bila-kniha

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.