Add the LLD drill log: log_lld_drill / get_lld_drill_log

[codenamesubho]

Adds the persistence step for the new LLD-drill skill: two tools that maintain one running, append-only LLD_SOLUTIONS_DIR/DRILL_LOG.md.

A drill is a short focused rep — one pattern, one class hierarchy, one "how would you extend this" — not a full mock interview. Giving each one a Mock Solutions/ folder, a problem.md and a rubric score would be noise, so they all append to a single file shaped like revision.md: newest at the bottom, entries split by a --- rule.

The contract the skill wires to

log_lld_drill(topic, content_markdown, problem_id="", duration_minutes=0, gaps="")
get_lld_drill_log(limit=5)        # limit=0 returns the whole file

content_markdown is the entire entry body, authored by the skill; the tool only stamps the dated header (## <date> · <topic> · <n> min) and appends. Bodies start their headings at ### and contain no bare --- rule, since ## and --- are the entry separators.

Feeding the existing trackers

• gaps (semicolon-separated, same vocabulary as log_session) go into index["weak_areas"] through a new _bump_weak_areas helper, now shared with log_session rather than duplicated.
• problem_id counts the drill as an attempt via _record_practice, whose doc_path is now Path | None. None means "this practice produced no per-problem file", so logging a drill preserves any doc save_practice_doc linked — covered by a test that calls get_practice_doc after a drill.
• Deliberately not written: an index["sessions"] record. That would appear in get_progress_summary's session table and in get_session_detail as an id with no revision.md entry behind it.

One consequence worth flagging: a drill also refreshes last_practiced, so it pushes that problem's STALE_DAYS revision clock out. Documented in the README; easy to change if drills shouldn't count for staleness.

Also

• _lld_kind returns drill-log for DRILL_LOG.md, so scans report it as server-generated (like feedback.md) instead of as a corpus-wide index. Import tools are unaffected — they still require kind=solution.
• Module docstring and README updated: storage layout, tool table, a new "LLD drills" section, tool count 26→29.

Testing

• test_lld_tools.py — 19 new hermetic checks covering create-then-append, the header/duration stamp, gap aggregation, doc_path preservation, limit clipping, the empty-drill guard, and drill-log classification. 114 checks pass, zero failures.
• test_server.py — new names added to EXPECTED_TOOLS; handshake and tool listing pass, 29 tools exposed.

🤖 Generated with Claude Code