hardwood: Skill for Claude Code

Search the parquet-dev mailing list

The lists.apache.org browse UI is a JavaScript app and its stats.lua JSON gets truncated/hallucinated when read through a summarizing web fetch. The reliable path is to download the raw monthly mbox archive and grep it locally. This skill encodes that path.

Archive URL (per calendar month):

https://lists.apache.org/api/mbox.lua?list=dev&domain=parquet.apache.org&d=YYYY-MM

Caching rule

• Past months are immutable — download once and reuse the cached copy.
• The current month is live — always re-download it, because new messages arrive continuously.

mlfetch.sh implements exactly this: it re-fetches YYYY-MM if it equals the current month (date +%Y-%m) or the cache file is missing/empty, otherwise it reuses the cache. Cache lives at ${PARQUET_ML_CACHE:-${XDG_CACHE_HOME:-$HOME/.cache}/parquet-dev-ml}.

Procedure

All commands assume the skill directory:

SKILL="/workspace/.claude/skills/parquet-dev-ml"

1. Fetch the month(s) you need. Pass one or more YYYY-MM. It prints the local mbox path(s):

   "$SKILL/mlfetch.sh" 2026-07                 # current month → always re-downloaded
   "$SKILL/mlfetch.sh" 2026-05 2026-06 2026-07 # older months served from cache
   

   If you don't know when a thread happened, fetch the last few months and grep across all of them. Widen the range rather than guessing a single month.

2. List subjects / find the thread. Grep the mbox(es) directly:

   CACHE="${PARQUET_ML_CACHE:-$HOME/.cache/parquet-dev-ml}"
   grep -hi '^Subject:' "$CACHE"/parquet-dev-2026-07.mbox | sort -u        # every subject
   grep -hi '^Subject:.*thrift' "$CACHE"/parquet-dev-*.mbox | sort -u      # by keyword
   

   Note the wording of subjects: a [DISCUSS]/[VOTE]/[ANNOUNCE] prefix and phrasing may differ from how the user remembers it (e.g. the user's "Inline parquet.thrift" was really [DISCUSS] Inline parquet.thrift into parquet-java). Match on a distinctive substring, not the exact remembered title.