Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate an explorable HTML report of Claude Code session usage (tokens, cache, subagents, skills, expensive prompts) from ~/.claude/projects transcripts.
.claude/skills/anthropics-session-report/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -23% | 0% |
Produce a self-contained HTML report of Claude Code usage and save it to the current working directory.
24h, 30d, or all). The script analyze-sessions.mjs lives in the same directory as this SKILL.md — use its absolute path:sh node <skill-dir>/analyze-sessions.mjs --json --since 7d > /tmp/session-report.json For all-time, omit --since.
/tmp/session-report.json. Skim overall, by_project, by_subagent_type, by_skill, cache_breaks, top_prompts.sh cp <skill-dir>/template.html ./session-report-$(date +%Y%m%d-%H%M).html
<script id="report-data" type="application/json"> with the full JSON from step 1. The page's JS renders the hero total, all tables, bars, and drill-downs from this blob automatically.<!-- AGENT: anomalies --> block with 3–5 one-line findings. Express figures as a % of total tokens wherever possible (total = overall.input_tokens.total + overall.output_tokens). One line per finding, exact markup:html <div class="take bad"><div class="fig">41.2%</div><div class="txt"><b>cc-monitor</b> consumed 41% of the week across just 3 sessions</div></div> Classes: .take bad for waste/anomalies (red), .take good for healthy signals (green), .take info for neutral facts (blue). The .fig is one short number (a %, a count, or a multiplier like 12×). The .txt is one plain-English sentence naming the project/skill/prompt; wrap the subject in <b>. Look for: a project or skill eating a disproportionate share, cache-hit <85%, a single prompt >2% of total, subagent types averaging >1M tokens/call, cache breaks clustering.
<!-- AGENT: optimizations --> block (at the bottom of the page) with 1–4 <div class="callout"> suggestions tied to specific rows (e.g. "/weekly-status spawned 7 subagents for 8.1% of total — scope it to fewer parallel agents").top_prompts already includes subagent tokens and rolls task-notification continuations into the originating prompt.top_prompts to 100 entries and cache_breaks to 100 before embedding (they should already be capped).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→fail | 5,543 | 3,863 | -30% | 1 | 1 | 0% | 394 | 984 | +150% | 0 | 0 | — |
case-15 | pass→fail | 10,340 | 3,462 | -67% | 1 | 1 | 0% | 2,096 | 1,623 | -23% | 0 | 0 | — |
case-01 | fail→fail | 6,419 | 4,948 | -23% | 1 | 1 | 0% | 330 | 1,047 | +217% | 0 | 0 | — |
case-02 | fail→fail | 22,147 | 5,631 | -75% | 1 | 1 | 0% | 5,382 | 1,079 | -80% | 0 | 0 | — |
case-03 | fail→fail | 13,295 | 4,465 | -66% | 1 | 1 | 0% | 3,259 | 1,024 | -69% | 0 | 0 | — |
case-05 | fail→fail | 24,795 | 4,184 | -83% | 1 | 1 | 0% | 6,157 | 971 | -84% | 0 | 0 | — |
case-06 | fail→fail | 7,688 | 5,680 | -26% | 1 | 1 | 0% | 1,686 | 1,130 | -33% | 0 | 0 | — |
case-07 | fail→pass | 14,120 | 63,689 | +351% | 1 | 1 | 0% | 2,823 | 1,472 | -48% | 0 | 0 | — |
case-08 | fail→pass | 9,338 | 3,820 | -59% | 1 | 1 | 0% | 2,024 | 1,482 | -27% | 0 | 0 | — |
case-09 | fail→pass | 8,768 | 3,059 | -65% | 1 | 1 | 0% | 1,572 | 1,282 | -18% | 0 | 0 | — |
case-10 | fail→fail | 6,527 | 1,763 | -73% | 1 | 1 | 0% | 1,048 | 1,066 | +2% | 0 | 0 | — |
case-11 | fail→fail | 7,780 | 2,613 | -66% | 1 | 1 | 0% | 1,357 | 1,244 | -8% | 0 | 0 | — |
case-12 | pass→fail | 7,852 | 2,139 | -73% | 1 | 1 | 0% | 1,419 | 1,097 | -23% | 0 | 0 | — |
case-13 | fail→pass | 11,597 | 4,544 | -61% | 1 | 1 | 0% | 2,338 | 1,674 | -28% | 0 | 0 | — |
case-14 | fail→fail | 11,431 | 8,101 | -29% | 1 | 1 | 0% | 2,495 | 1,164 | -53% | 0 | 0 | — |
case-16 | pass→pass | 7,122 | 1,411 | -80% | 1 | 1 | 0% | 1,389 | 971 | -30% | 0 | 0 | — |
case-17 | pass→pass | 5,445 | 1,447 | -73% | 1 | 1 | 0% | 966 | 1,033 | +7% | 0 | 0 | — |
case-18 | pass→pass | 7,542 | 2,663 | -65% | 1 | 1 | 0% | 1,486 | 1,207 | -19% | 0 | 0 | — |
case-19 | pass→pass | 11,315 | 1,930 | -83% | 1 | 1 | 0% | 2,033 | 1,142 | -44% | 0 | 0 | — |
case-20 | fail→fail | 3,375 | 7,048 | +109% | 1 | 1 | 0% | 529 | 1,982 | +275% | 0 | 0 | — |
case-21 | fail→fail | 23,193 | 4,901 | -79% | 1 | 1 | 0% | 6,164 | 947 | -85% | 0 | 0 | — |
case-22 | fail→fail | 8,382 | 6,369 | -24% | 1 | 1 | 0% | 1,578 | 1,932 | +22% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 14 counted toward the lift figure. The other 8 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +9 percentage points is the difference between those two pass rates over the 14 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.