/dig - Session Goldminer v2

Mine Claude Code session data for timelines, gaps, and repo attribution. No query needed.

Usage

/dig                    # Current repo, 10 most recent
/dig [N]                # Current repo, N most recent
/dig --all              # All repos, ALL sessions (auto-detect count)
/dig --all [N]          # All repos, N most recent
/dig --deep             # Deep scan — ALL .jsonl files (not just most-recent per project)
/dig --deep [N]         # Deep scan, N most recent
/dig --all --deep       # All repos, deep scan — full coverage
/dig --timeline         # Day-by-day grouped (current repo)
/dig --all --timeline   # Day-by-day grouped (all repos, ALL sessions)

Deep Mode (#216)

Standard /dig deduplicates by session basename — only showing the most recent file per session. This misses historical sessions, subagent sessions, and worktree sessions.

--deep scans ALL .jsonl files across all project directories. Output includes extra fields:

• toolCalls — number of tool invocations in the session
• fileSizeKB — session file size in KB
• Coverage metadata: "X of Y sessions" showing how many were returned vs found

Timezone (#214)

dig.py v2 auto-detects timezone instead of hardcoding GMT+7:

1. MAW_DISPLAY_TZ env var (e.g., Asia/Bangkok, 7, America/New_York)
2. TZ env var
3. System timezone
4. Fallback: UTC

Output metadata includes detected timezone name and offset.

Step 0: Timestamp

date "+🕐 %H:%M %Z (%A %d %B %Y)"
ENCODED_PWD=$(pwd | sed 's|^/|-|; s|[/.]|-|g')
PROJECT_BASE=$(ls -d "$HOME/.claude/projects/${ENCODED_PWD}" 2>/dev/null | head -1)
export PROJECT_DIRS="$PROJECT_BASE"

# Strip -wt* suffix to find parent project dir
PARENT_ENCODED=$(echo "$ENCODED_PWD" | sed 's/-wt-[^/]*$//')
if [ "$PARENT_ENCODED" != "$ENCODED_PWD" ]; then
  PARENT_BASE=$(ls -d "$HOME/.claude/projects/${PARENT_ENCODED}" 2>/dev/null | head -1)
  [ -n "$PARENT_BASE" ] && export PROJECT_DIRS="$PROJECT_DIRS:$PARENT_BASE"
fi

# nullglob-safe worktree scan (both parent and self)
for base in "$PROJECT_BASE" "$PARENT_BASE"; do
  [ -z "$base" ] && continue
  for wt in "$base"-wt-*(N); do  # (N) = zsh nullglob qualifier
    [ -d "$wt" ] && export PROJECT_DIRS="$PROJECT_DIRS:$wt"
  done
done

Encodes pwd the same way Claude does (replace / and . with -, prepend -) to match the .claude/projects/ directory naming (e.g. github.com → github-com). Detects parent project from worktree dirs (strips -wt-* suffix) and includes both parent and self worktrees. The (N) qualifier prevents zsh crashes when no worktrees exist.

With --all (all repos):

export PROJECT_DIRS=$(ls -d "$HOME/.claude/projects/"*/ | tr '\n' ':')

Step 2: Extract Session Data

Run the dig script. Pass 0 for --all (no limit), or N if user specified a count, default 10:

python3 ~/.claude/skills/dig/scripts/dig.py [N] [--deep]
# N=10 (default), N=0 (scan all sessions), N=50 (50 most recent)
# --deep: scan ALL .jsonl files (not just most-recent per basename)

Deep Mode Display

When --deep is used, add extra columns to the table:

| # | Date | Session | Min | Repo | Msgs | Tools | Size | Focus |

And show coverage in the footer:

**Coverage**: [returned]/[found] sessions | **Timezone**: [tz_name] (GMT+[offset])

Step 3: Display Timeline

Read the JSON output and display as a table. Sessions are chronological (oldest first). Gap rows (type: "gap") span the session column with · · · prefix:

## Session Timeline

| # | Date | Session | Min | Repo | Msgs | Focus |
|---|------|---------|-----|------|------|-------|
|   |      | · · · sleeping / offline | | | | |
| 1 | 02-21 | 08:40–09:08 | 28m | oracle-skills-cli | 5 | Wire /rrr to read pulse data |
|   |      | · · · 45m gap | | | | |
| 2 | 02-21 | 09:55–10:23 | 28m | homelab | 3 | oracle-pulse birth + CLI flag |
|   |      | · · · no session yet | | | | |

**Dirs scanned**: [list PROJECT_DIRS]
**Total sessions found**: [count]

Column rendering rules:

• Gap rows: | | | · · · [label] | | | | | — number + date empty, label in Session col
• Date: MM-DD short format (strip year)
• Session: HH:MM–HH:MM using startGMT7 and endGMT7 (strip date, keep time only)
• Min: [durationMin]m
• Repo: use repoName field from dig.py output (resolved via ghq)
• Msgs: realHumanMessages count

"Msgs" = real typed human messages (not tool approvals).

With --timeline: Group by Date

When --timeline flag is present, group sessions by date instead of a flat table. Use --all to see all repos (recommended for timeline).

Step 1: Run dig.py with 0 for --all (scans all sessions), or user-specified count

Step 2: Group sessions by date from startGMT7. Render each day as:

## Feb 22 (Sun) — [vibe label]

                  · · ·   sleeping / offline
08:48–09:11    23m   homelab        Update Fleet Runbook + Explore black.local
09:11–11:30   139m   homelab        Set Up KVM OpenClaw Node on black.local
09:37–12:51   194m   Nat-s-Agents   /recap → supergateway → CF ZT → arra-oracle-v3 dig
                  · · ·   45m gap
12:51–13:03    12m   Nat-s-Agents   Dig All + Design arra-oracle-v3 ← current
                  · · ·   no session yet

## Feb 21 (Sat) — Long day: Fleet + Brewing + Skills

06:19–08:38   139m   homelab        Moltworker Gateway + MBP Node
08:40           (bg)  openclaw       ClawHub Build Script (idle long)
09:23–16:08   405m   homelab        Debug MBP Node 401 — Gateway Token Auth

Rendering rules:

• Day header: ## MMM DD (Day) — [vibe label] — infer vibe from session summaries (e.g. "Infrastructure Day", "Brewing + Skills")
• Session rows: HH:MM–HH:MM [N]m REPO Summary — use repoName for repo, summary for focus
• Gap rows: · · · [label] between sessions when gap > 30 min
• Sidechain: prefix (bg) for sessions with isSidechain: true
• Current: append ← current marker on the last session of the current day (today only)
• Sort: days newest-first, sessions within each day oldest-first (chronological)
• Date format: startGMT7 time portion only (HH:MM), endGMT7 time portion (HH:MM)
• Repo width: pad repo names to align columns

Step 3: Show summary footer:

**Days**: [count] | **Sessions**: [count] | **Total time**: [sum of durationMin]m

Trace Connection

After displaying the timeline, log the discovery to Oracle so it's searchable later:

arra_trace({
  query: "dig session [N]",
  project: "[repo-name]",
  foundFiles: [],
  foundCommits: [list of commit hashes from timeline],
  foundIssues: []
})

This connects /dig discoveries to /trace — "When did we first work on X?" becomes answerable.

ARGUMENTS: $ARGUMENTS