Case 04808

Cleanup Review

python scripts/_cache_clear.py is an optional cleanup helper for the Anthropic Usage Report Format skill. It may recursively remove /tmp/.skill_state, /tmp/skill_scratch. Show the exact paths first, explain whether they are skill-private temporary paths or user state, and run the cleanup only after the user confirms.

Anthropic Usage Report Format

Design Principles

1. Icons = signal importance only. Icons appear ONLY at the start of lines with key metrics the user needs to scan quickly. No decorative icons.
2. Key metrics first. CLI %, cache %, high-output call count — these appear at line start with icons.
3. No noise. No latency (ms), no success/error icons, no redundant breakdowns unless errors exist.
4. Neutral labeling. High-output calls are flagged for review, not pre-judged as "should have used CLI" — the cause may be inefficiency, legitimate work, or missing CLI usage.
5. Compact. Every line earns its place.

Report Template

Anthropic Usage — {date} ({time} CET)
══════════════════════════════════════════
Total: {total} calls ({errors} errors)
⚠️ CLI: {cli_pct}% / API: {api_pct}%
{agent_name}:
  🗄️ {cache_pct}% cache ({cache_read} read / {cache_create} create)
  Tokens: {total_in} in ({new_in} new) / {total_out} out
  Cost: ~${cost}
──────────────────────────────────────────
CLI potential:
  ⚠️ {n}× inline work (>{threshold} out, {sum} tok, max {max})
  {n}× standard ({low}-{threshold} out)
  {n}× heartbeat (<{low} out)
──────────────────────────────────────────
Total estimate: ~${total_cost}

Conditional Rules

• Error breakdown: Show API: X (Y ok, Z err) | CLI: X (Y ok, Z err) ONLY when errors > 0
• CLI icon: ⚠️ when CLI < 50%, ✅ when CLI >= 50%
• High output icon: ⚠️ when inline work calls > 0, ✅ (show "No inline work") when 0
• Model line: Show only if multiple models used

Icon Placement Rules

Icons appear ONLY on these lines, ALWAYS at line start:

• ⚠️ or ✅ — CLI/API ratio
• 🗄️ — Cache rate
• ⚠️ or ✅ — Inline work count

NO icons on: header, totals, tokens, cost, heartbeat count, standard count, separator lines.

Language

Default language is English. To localize, edit LANG dict in scripts/format_report.py:

LANG = {
    "total":        "Total",
    "calls":        "calls",
    "errors":       "errors",
    "cache":        "cache",
    "tokens":       "Tokens",
    "cost":         "Cost",
    "total_est":    "Total estimate",
    "cli_potential": "CLI potential",
    "inline_work":  "inline work",
    "standard":     "standard",
    "heartbeat":    "heartbeat",
    "no_calls":     "No API calls today.",
    "no_inline":    "No inline work",
}

Presets available in scripts/:

• lang_cs.py — Czech / Čeština

Implementation

Reference implementation: scripts/format_report.py

To apply to an existing installation:

python3 scripts/patch_format.py /path/to/anthropic-report.py