AlgoSu ADR

Goal

Eliminate the pain point where AI analysis receives only user code, forcing the AI to reverse-engineer the problem.
Wave B: Inject problem information (title + description) into group analysis to improve multi-submission comparison accuracy.
Wave A: Auto-populate description via Programmers crawling on problem registration, fundamentally resolving the missing AI context.
Wave S (spike): Validate that Programmers can be parsed with httpx+BeautifulSoup4 alone, before committing to Wave A.

Sprint 238 (ADR-030 S-5) implemented the <problem_context> injection pattern for individual analysis, but the group analysis endpoint (/group-analysis) was never updated with the same pattern.
The Problem entity already had sourceUrl/sourcePlatform fields (from Sprint 97 Programmers migration), but no auto-crawling logic on registration — leaving description empty in most cases.
Crawling feasibility could not be assumed without a spike, so Wave S (Scout) was placed before Wave A.

On group analysis request, query Problem Service GET /internal/{problem_id} using problem_id.
On query failure (network error, problem not found): fallback — log warning and proceed without problem info to preserve service availability.
<problem_context> block follows the ADR-030 S-5 isolation pattern — prompt injection defense included.
Both GROUP_SYSTEM_PROMPT and GROUP_SQL_SYSTEM_PROMPT updated with injection defense guidelines.
Outbound key: INTERNAL_KEY_PROBLEM (CLAUDE.md Sprint 239 Q-5 SSOT).

Scout spike result: BOJ service fully shut down on 2026-04-28, Programmers confirmed as SSR (no JS rendering required).
Playwright prohibited (Docker image ~400MB increase, browser runtime not allowed in problem service) → axios + cheerio lightweight crawling.
On problem registration: if description is empty and sourceUrl/sourcePlatform are present, trigger async backfill (fire-and-forget, non-blocking).
ALLOWED_HOSTS whitelist: school.programmers.co.kr only + https: protocol enforcement → SSRF defense.

Running Wave S (Scout) before Wave A to validate crawling feasibility proved correct.
Operational changes like BOJ shutdown can only be confirmed by actual HTTP requests, not static analysis.

BOJ acmicpc.net/problem/1000: HTTP 200 but returns "scoring service under maintenance" page — shutdown confirmed.
Programmers school.programmers.co.kr/learn/courses/30/lessons/{N}: SSR confirmed, HTML received via httpx direct request.
- Parseable fields: title (.challenge-title), difficulty ([data-challenge-level]), description (div.markdown), constraints, I/O example tables.

File	Change
`src/config.py`	Added `problem_service_url` / `problem_service_key` env vars
`src/prompt.py`	`build_group_user_prompt()` — added `problem_title`/`problem_description` params, `<problem_context>` isolated block injection
`src/main.py`	`/group-analysis` endpoint — Problem Service query + fallback logic
`tests/test_prompt.py`	5 new tests (context inject / no-inject / sanitize)
`tests/test_main.py`	2 new tests (Problem Service success / failure scenarios)

File	Change
`src/crawler/crawler.service.ts`	ALLOWED_HOSTS whitelist + axios+cheerio Programmers SSR parsing
`src/crawler/crawler.module.ts`	Module registration
`src/problem/problem.module.ts`	CrawlerModule import
`src/problem/problem.service.ts`	Async crawling trigger after `create()` (with `.catch()`)
`src/crawler/crawler.service.spec.ts`	5 new tests (platform detection, parsing, error handling, SSRF blocking)
`src/problem/problem.service.spec.ts`	2 new tests (crawling trigger, early return)

Severity	Finding	Fix
P1 SSRF	`crawler.service.ts` — unvalidated `sourceUrl` passed directly to axios	`ALLOWED_HOSTS` whitelist + `https:` protocol enforcement
P2 envelope unwrap	`main.py` — `{data: problem}` response accessed directly for fields	`prob_data.get("data", prob_data)` unwrap before extraction
P2 catch handler	`problem.service.ts` — fire-and-forget missing catch → unhandled rejection	Added `.catch()` with Error / non-Error branching

CI: 38 checks PASS / 0 FAIL
AI Analysis: 353 passed · TOTAL 99% · main.py 100% · prompt.py 100%
Problem: 205 passed · Statements 99.02% · Branches 96.47% ≥ 96% · Functions 98.61% ≥ 98% · Lines 99.13%
ESLint: 0 errors (tsc pre-existing baseUrl deprecation at tsconfig.json:12 is a known issue from Sprint 246, unrelated to changes)
Critic (Codex gpt-5.5): All 3 findings (P1·P2·P2) resolved → CLEAN
PR #462 squash merge → a7731fa (2026-07-15T08:58:11Z)