status

Render the current Claude Supercharger session state. Arguments: $ARGUMENTS

Read these files (silently — do not show their raw content) and produce a dashboard:

STEP 0 — determine the CURRENT session id (do this first; everything per-session depends on it). Do NOT guess by "newest file" — under concurrent sessions that grabs another project's session (the exact bug this step fixes). Derive it from the current working directory's own transcript directory:

ENC="-$(pwd | sed 's|/|-|g; s|^-||')"                 # cwd → CC's project-dir encoding
SID=$(ls -t "$HOME/.claude/projects/$ENC"/*.jsonl 2>/dev/null | head -1 | xargs -r basename | sed 's/\.jsonl$//')

$SID is THIS session. Use it for every per-session read below (.main-tokens-$SID, .tool-history-$SID, .memory-restored-$SID, .repetition-flag-$SID). If $SID is empty (no transcript yet), fall back to the globally-newest .main-tokens-* and note it.

Session cost — compute from the TRANSCRIPT, do NOT trust .main-tokens-$SID.cost_usd. That accumulator only counts from when v2.7.63 deployed, so a session spanning the upgrade is badly under-reported (observed: $22.83 vs a real $2343.54). A one-shot command can afford to re-sum the transcript for ground truth:

python3 - "$HOME/.claude/projects/$ENC/$SID.jsonl" <<'PY'
import json,sys
P={'opus':(5.0,6.25,0.5,25.0),'sonnet':(3.0,3.75,0.3,15.0),'haiku':(0.8,1.0,0.08,4.0),'fable':(10.0,12.5,1.0,50.0)}
t=0.0
try:
    for ln in open(sys.argv[1]):
        try: d=json.loads(ln)
        except: continue
        if d.get('type')!='assistant': continue
        u=(d.get('message') or {}).get('usage') or {}
        if not u: continue
        m=((d.get('message') or {}).get('model') or '').lower()
        k='opus' if 'opus' in m else 'fable' if ('fable' in m or 'mythos' in m) else 'haiku' if 'haiku' in m else 'sonnet'
        ip,cw,cr,op=P[k]
        t+=(u.get('input_tokens',0)*ip+u.get('cache_creation_input_tokens',0)*cw+u.get('cache_read_input_tokens',0)*cr+u.get('output_tokens',0)*op)/1e6
except Exception: pass
print(f'{t:.2f}')
PY

Use that value for Cost (session). (The budget cap still uses the accumulator, which is accurate for sessions that start after v2.7.63.)

Files to read (skip any that don't exist):

• ${CLAUDE_PLUGIN_DATA}/scope/.economy-tier (Tier: minimal|lean|standard; absent → standard)
• ${CLAUDE_PLUGIN_DATA}/scope/.mcp-profile (MCP profile: light|dev|research|full; absent → "—")
• ${CLAUDE_PLUGIN_DATA}/scope/.profile (Hook profile: standard|fast|minimal; absent → standard)
• ${CLAUDE_PLUGIN_DATA}/scope/.disabled-hooks
• ${CLAUDE_PLUGIN_DATA}/scope/.tool-history-$SID (last 10 entries)
• ${CLAUDE_PLUGIN_DATA}/scope/.repetition-flag-$SID
• ${CLAUDE_PLUGIN_DATA}/scope/.memory-restored-$SID (per-session; mtime → "compaction X min ago"). The bare .memory-restored (no suffix) is a DEPRECATED global flag not written since v2.7.47 — do NOT use it; if .memory-restored-$SID is absent, this session hasn't compacted → "this session".
• .claude/supercharger/lessons.jsonl (count + 3 most recent lesson fields)
• Claude file-memory dir ~/.claude/projects/$ENC/memory/ (reuse $ENC from STEP 0): count the entry files (*.md excluding MEMORY.md — MEMORY.md is the index, not an entry) and take the NEWEST mtime across the dir as "last updated". This is the curated, recalled memory the user maintains — NOT the git-gated .claude/supercharger-memory.md session file (which is stale-by-design and no longer surfaced here as of v2.7.75). If the dir is absent, render —.
• .supercharger.json (role, economy, profile, budget, hints)
• ${CLAUDE_PLUGIN_DATA}/audit/$(date -u +%Y-%m-%d).jsonl (count of today's events)
• ${CLAUDE_PLUGIN_DATA}/scope/.subagent-costs-*.jsonl (per-subagent cost rollup — aggregate by agent_name, show top 3 by cost_usd)
• ${CLAUDE_PLUGIN_DATA}/scope/.blocked-commands (the block log; last 3 lines for "Recent blocks" — format: [ts] category — reason — command)