Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When a user uploads or references a data file (CSV, JSON, XLSX, TXT, LOG) or any file larger than 100KB, immediately estimate token cost. If over 30k tokens, chunk and summarize. Use when large files are detected to prevent context overflow and maintain agent performance.
.claude/skills/adityapeshave-sophos-dont-be-greedy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 850% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 538% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 324% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 206% | 0% |
Prevents context overflow by enforcing size-aware data loading. Large files can exceed context windows and crash agent workflows. This skill measures files before loading, chunks oversized data, and returns compact summaries with safe previews so downstream processing can continue without context exhaustion.
Before loading ANY data file:
bashpython scripts/estimate_size.py "<file_path>"
This returns byte count and estimated token count.
| Estimated Tokens | Action | |------------------|--------| | < 10,000 | Run quick inspection, load directly | | 10,000 - 30,000 | Run quick inspection, consider filtering | | > 30,000 | Chunk and summarize before loading |
For files under 10k tokens:
bashpython scripts/quick_inspect.py "<file_path>"
Return stats and load file directly.
For files over 30k tokens:
bashpython scripts/chunker.py "<file_path>" python scripts/summarize.py "<chunk_file>"
Return overall summary + per-chunk summaries + safe preview of first rows.
Always provide:
cat on unknown or large filesInput: User says "Analyze this sales data" and uploads a 50MB CSV file
Workflow:
scripts/estimate_size.py sales.csv - Output: 50MB, ~13M tokensscripts/chunker.py sales.csvscripts/summarize.py on representative chunksInput: User asks "Check my config.json for issues"
Workflow:
scripts/estimate_size.py config.json - Output: 2KB, ~512 tokensscripts/quick_inspect.py config.jsonInput: User uploads a 500KB application.log
Workflow:
scripts/estimate_size.py application.log - Output: 500KB, ~128K tokensscripts/chunker.py application.log| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→fail | 18,450 | 6,166 | -67% | 1 | 1 | 0% | 3,317 | 1,231 | -63% | 0 | 0 | — |
case-02 | fail→fail | 2,057 | 6,679 | +225% | 1 | 1 | 0% | 337 | 1,402 | +316% | 0 | 0 | — |
case-01 | fail→fail | 4,871 | 6,586 | +35% | 1 | 1 | 0% | 847 | 1,286 | +52% | 0 | 0 | — |
case-03 | fail→fail | 4,006 | 7,201 | +80% | 1 | 1 | 0% | 688 | 1,126 | +64% | 0 | 0 | — |
case-04 | fail→pass | 8,898 | 15,629 | +76% | 1 | 1 | 0% | 434 | 4,125 | +850% | 0 | 0 | — |
case-05 | fail→pass | 6,974 | 19,418 | +178% | 1 | 1 | 0% | 477 | 3,043 | +538% | 0 | 0 | — |
case-06 | fail→fail | 3,243 | 4,802 | +48% | 1 | 1 | 0% | 509 | 1,717 | +237% | 0 | 0 | — |
case-07 | fail→pass | 5,403 | 4,237 | -22% | 1 | 1 | 0% | 371 | 1,572 | +324% | 0 | 0 | — |
case-08 | fail→fail | 4,445 | 6,619 | +49% | 1 | 1 | 0% | 661 | 1,145 | +73% | 0 | 0 | — |
case-09 | fail→pass | 12,215 | 4,337 | -64% | 1 | 1 | 0% | 1,589 | 1,630 | +3% | 0 | 0 | — |
case-10 | fail→fail | 11,854 | 18,418 | +55% | 1 | 1 | 0% | 2,576 | 4,698 | +82% | 0 | 0 | — |
case-11 | fail→pass | 3,435 | 3,581 | +4% | 1 | 1 | 0% | 516 | 1,581 | +206% | 0 | 0 | — |
case-12 | fail→pass | 13,407 | 18,024 | +34% | 1 | 1 | 0% | 2,595 | 3,825 | +47% | 0 | 0 | — |
case-13 | fail→pass | 4,028 | 9,750 | +142% | 1 | 1 | 0% | 674 | 2,637 | +291% | 0 | 0 | — |
case-14 | pass→pass | 8,133 | 7,037 | -13% | 1 | 1 | 0% | 1,710 | 2,306 | +35% | 0 | 0 | — |
case-15 | pass→pass | 2,950 | 2,385 | -19% | 1 | 1 | 0% | 516 | 1,300 | +152% | 0 | 0 | — |
case-16 | pass→pass | 3,030 | 3,285 | +8% | 1 | 1 | 0% | 499 | 1,525 | +206% | 0 | 0 | — |
case-17 | fail→fail | 6,848 | 7,922 | +16% | 1 | 1 | 0% | 1,092 | 1,855 | +70% | 0 | 0 | — |
case-19 | fail→fail | 4,649 | 5,080 | +9% | 1 | 1 | 0% | 871 | 1,130 | +30% | 0 | 0 | — |
case-20 | fail→fail | 3,809 | 6,093 | +60% | 1 | 1 | 0% | 546 | 1,217 | +123% | 0 | 0 | — |
case-21 | fail→fail | 8,459 | 15,008 | +77% | 1 | 1 | 0% | 1,496 | 4,509 | +201% | 0 | 0 | — |
case-22 | fail→fail | 6,775 | 6,687 | -1% | 1 | 1 | 0% | 1,238 | 1,273 | +3% | 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 11 counted toward the lift figure. The other 11 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 +32 percentage points is the difference between those two pass rates over the 11 comparable cases.
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.