mine

name	mine
description	Resume mining for an active job application. Searches the hypha archive for achievement statements (shipped/led/launched/prevented/saved/mentored/owned), diffs against ~/Projects/job-hunt/data/cv.md, and surfaces metrics that didn't make it to the current CV. Use when the user says "mine my history for [company]", "before I apply to [company]", "/mine <company>", "what did I drop from my CV".
argument-hint	<company-or-job-id>
allowed-tools	["Bash","Read","Grep"]

/mine

The implementation of the hirenaledios-mine skill that's been a stub. Per ~/.claude/skills/hirenaledios-mine/SKILL.md:

"The highest-leverage 20-minute block of any ≥ 4.0/5 application. Without this step, TEALS + the $500M federal-fines metric almost didn't ship on Uncommon Schools (retro §2.1)."

Hypha makes this real. Cross-source: emails to managers describing accomplishments + Drive docs containing old resumes + calendar events anchoring project timelines.

Arguments

User invoked with: $ARGUMENTS — a company name or job-id slug.

Instructions

1. Locate the application directory:

   APP_DIR="${HOME}/Projects/job-hunt/data/applications/$ARGUMENTS"
   if [[ ! -d "$APP_DIR" ]]; then
     # Try fuzzy match
     APP_DIR=$(ls -d ~/Projects/job-hunt/data/applications/* 2>/dev/null | grep -i "$ARGUMENTS" | head -1)
   fi
   

   If none found, tell the user the application directory doesn't exist and offer to create it.

2. Ensure the daemon:

   bash ${CLAUDE_PLUGIN_ROOT}/scripts/ensure-daemon.sh
   

3. Read the current CV as the diff baseline:

   CV="${HOME}/Projects/job-hunt/data/cv.md"
   

   Use the Read tool to load it.

4. Mine achievement statements from the archive. Run these searches in parallel via Bash and capture the JSON:

   for verb in shipped led launched prevented saved mentored owned drove built improved reduced increased; do
     bash ${CLAUDE_PLUGIN_ROOT}/scripts/hypha-query.sh search "$verb" gmail.message 2>/dev/null | head -c 5000 > /tmp/hypha-mine-$verb.json
   done
   

5. Mine old resumes from Drive:

   bash ${CLAUDE_PLUGIN_ROOT}/scripts/hypha-query.sh search "resume" file.document
   bash ${CLAUDE_PLUGIN_ROOT}/scripts/hypha-query.sh search "CV" file.document
   

6. Read the JD if it exists ($APP_DIR/jd.md or $APP_DIR/job.md) so the diff is JD-aware — surface only metrics that match the role's mission-fit.

7. Synthesise a mine.md document at $APP_DIR/mine.md with this structure:

   • JD-relevant metrics surfaced from archive but missing from cv.md — the load-bearing list. Each entry has: the metric, the source (email subject + date, OR drive doc title), and a one-line "why it fits this JD."
   • Old resume versions worth re-reading — list with paths or hypha node ids.
   • Risk metrics — anything that contradicts current CV claims (retro hygiene; the resume should be consistent).
   • Three concrete additions to make to cv.md for this application.

8. Do not write to cv.md. That's a manual decision per the user's YAGNI + no-CV-drift discipline. The skill outputs a proposal at mine.md, the user decides.

Why this matters

This is the use case the competitive analysis flagged as the second-strongest cross-source query (after the relationship dossier). It's the hypha-shaped problem you literally already documented: the M2 stub waiting for an implementation. Hypha is M2.

Common failure modes

• No ~/Projects/job-hunt: explain that this skill is wired to the user's job-hunt project layout. Offer to skip and run a generic mining (search the archive only) without the diff.
• No CV at data/cv.md: still useful — surface the metrics; user can decide what to do with them.
• Archive is empty / not yet ingested: point to the ingest command and stop.