Step 0 — Load config

Suggest the freshest recently-asked coding interview problem the user has NOT solved yet, from an opt-in live community feed. The feed is the only part of this skill that touches the network, and it is OFF by default.

Step 0 — Load config

Look for dsa-coach.config.json in the current working directory. If present, parse it and use its values. For any missing key, fall back to the documented default below. If the file is absent entirely, use all defaults.

Config keys + defaults this skill uses:

• language (default "ts") — solution file extension (for the local solved-scan).
• solutionDirs (default { "leetcode": "leetcode", "geeksforgeeks": "geeksforgeeks", "hackerrank": "hackerrank", "interviewbit": "interviewbit" }) — platform key → folder name.
• liveQuestions (default { "enabled": false, "endpoint": "https://www.hellointerview.com/trpc/intelligence/getQuestionsAndUpvotes", "interviewTypes": ["CODING", "AI_ENABLED_CODING"], "maxPages": 3 }) — the live feed. enabled is the master opt-in; when false this skill makes NO network call.

Step 0a — Opt-in gate (HARD STOP if disabled)

If liveQuestions.enabled is not exactly true, do NOT fetch anything. Print this and stop:

fresh-problem is opt-in and currently disabled. This skill fetches recently-asked interview questions live from an external community feed (default: HelloInterview). To enable it, add this to dsa-coach.config.json and run the skill again:

"liveQuestions": { "enabled": true }

Fetched questions are used only for your own practice — nothing is saved to your repo or shared.

Make no further tool calls. The gate is absolute: no network access happens unless enabled === true.

Step 1 — Scan local solved problems (offline)

Build the set of problems the user has already solved by scanning their own repo (same approach as suggest-problem). For each folder in solutionDirs:

git ls-files -- "<dir>/**/*.<language>" "<dir>/*.<language>"

If git ls-files returns nothing (not a git repo / untracked), fall back to:

ls -1 <dir>/**/*.<language> <dir>/*.<language> 2>/dev/null

Record each solved file's name and, for LeetCode files named <NNN>_<slug>.<language>, its LeetCode number <NNN>.

Step 2 — Source adapter: fetch + normalize (THE ONLY SOURCE-SPECIFIC SECTION)

This is the one place that knows the current source's wire format. To switch sources later, rewrite ONLY this section; everything below stays the same.

Run the fetch with python3 (already used elsewhere in this plugin). Substitute <endpoint> and <maxPages> from liveQuestions, and pass liveQuestions.interviewTypes as a JSON-array string for the third argument (e.g. '["CODING","AI_ENABLED_CODING"]'):

python3 - "<endpoint>" "<maxPages>" '<interviewTypesJson>' <<'PY'
import json, sys, urllib.parse, urllib.request
# Exit 0 in every case so this bash step never errors; the agent reads stdout
# to detect failure (a line containing "_error", or no output at all).
endpoint, maxpages, types = sys.argv[1], int(sys.argv[2]), json.loads(sys.argv[3])
out = []
for page in range(1, maxpages + 1):
    payload = {"0": {"json": {
        "company": None, "interviewTypes": types, "interviewLevels": None,
        "regions": None, "search": None, "sort": "recentAndPopular",
        "page": page, "pageSize": 25, "atLeastOneUpvote": True},
        "meta": {"values": {"company": ["undefined"], "interviewLevels": ["undefined"],
        "regions": ["undefined"], "search": ["undefined"]}}}}
    q = urllib.parse.urlencode({"batch": "1", "input": json.dumps(payload, separators=(',', ':'))})
    req = urllib.request.Request(endpoint + "?" + q,
        headers={"Accept": "application/json", "User-Agent": "dsa-coach"})
    try:
        data = json.load(urllib.request.urlopen(req, timeout=25))
        qs = data[0]["result"]["data"]["json"]["questions"]
    except Exception as e:
        print(json.dumps({"_error": str(e)})); sys.exit(0)
    if not qs:
        break
    for x in qs:
        ups = x.get("upvotes") or []
        recent = max(ups, key=lambda u: (u.get("approxAskedOn") or ""), default=None)
        out.append({
            "title": x.get("title"),
            "type": x.get("defaultInterviewType"),
            "companies": sorted({u.get("company") for u in ups if u.get("company")}),
            "level": (recent.get("level") if recent else None),
            "askedOn": (recent.get("approxAskedOn") if recent else None),
            "link": x.get("detailsLink"),
            "statement": x.get("description"),
        })
for r in out:
    print(json.dumps(r))
PY

Each output line is one normalized question: {title, type, companies[], level, askedOn, link, statement}. The list is already ordered freshest-first (recentAndPopular).

Error handling: If the only output is a line containing _error, or there is no output at all, tell the user the feed could not be reached right now (show the error briefly) and stop — do not fall back to the static curricula (that is suggest-problem's job).

Step 3 — Filter (source-agnostic)

Working from the normalized list, in order:

1. Drop solved problems.
   • If title matches Leetcode <NNN>. (or LeetCode #<NNN>), extract <NNN> and drop it if a solved LeetCode file with that number exists.
   • Otherwise, compare the title/slug against solved filenames (case-insensitive, ignoring punctuation). Drop on a confident match.
   • When unsure, keep it and mark it (possibly solved — not in your folders by name) in the output.
2. Company filter (only if $ARGUMENTS is non-empty). Keep only questions whose companies[] contains the requested company (case-insensitive substring match). If nothing matches, say so and suggest re-running without the company filter — then stop.

Step 4 — Present

Pick the FIRST remaining (freshest) question. Present it like a fresh interview problem. Do NOT give hints, approach, or solution unless the user explicitly asks (same rule as suggest-problem).

Format:

---

Recently asked · [companies, comma-joined] · [level] · ~[askedOn date] · community feed

Problem: [title] Type: [type] Maps to: [LeetCode #NNN if the title is a Leetcode problem, else "not in your bundled curricula (fresh)"] Asked at: [companies, comma-joined]

Problem Statement: [statement]

Reference: [endpoint host + link, e.g. https://www.hellointerview.com]

After presenting, add one line:

This was pulled live for your practice and not saved. Solved it? Use add-problem <file> to record it.

$ARGUMENTS