Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when working with deepseek-v4-pro or deepseek-v4-flash in thinking mode on multi-step or plan-driven tasks. Provides rules to prevent stale references, unverified plan assumptions, and vague plan output.
.claude/skills/hmbown-v4-best-practices/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 12% | 0% |
Rules for multi-step V4 thinking-mode workflows. Each rule prevents a specific, observable failure class.
Before referencing a file path, function, or type in code or plan output, call grep_files or read_file to confirm it exists in the workspace.
# Bad: edit_file path="src/config/loader.rs" (assumed from memory)
# Good: grep_files pattern="pub fn load_config" → confirms src/config/mod.rs:42
# then reference src/config/mod.rs:42Failure avoided: edit_file errors on non-existent paths; LSP diagnostics on hallucinated symbols.
Before executing a plan that touches 3+ files, spawn a deepseek-v4-flash sub-agent (thinking off) to read the target files and confirm path/symbol assumptions still hold.
agent type="verifier" model="deepseek-v4-flash"
prompt: "Read these files and confirm: [list assumptions]. Report mismatches."Failure avoided: multi-step edits fail partway because file structure changed since the plan was drafted.
In plan-mode output, replace vague location pointers with path:line references drawn from a prior grep_files result.
# Bad: "Update the retry logic in the client module"
# Good: "Update retry loop at crates/tui/src/client.rs:187"Failure avoided: agent-mode execution cannot locate the intended edit target when plan directions are imprecise.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,626 | 3,709 | -78% | 1 | 1 | 0% | 3,651 | 680 | -81% | 0 | 0 | — |
case-02 | fail→fail | 2,307 | 4,178 | +81% | 1 | 1 | 0% | 347 | 694 | +100% | 0 | 0 | — |
case-03 | fail→fail | 2,220 | 3,607 | +62% | 1 | 1 | 0% | 281 | 616 | +119% | 0 | 0 | — |
case-04 | pass→pass | 2,716 | 2,084 | -23% | 1 | 1 | 0% | 506 | 857 | +69% | 0 | 0 | — |
case-05 | pass→pass | 7,898 | 5,878 | -26% | 1 | 1 | 0% | 1,683 | 1,657 | -2% | 0 | 0 | — |
case-06 | pass→pass | 3,444 | 2,710 | -21% | 1 | 1 | 0% | 650 | 964 | +48% | 0 | 0 | — |
case-07 | fail→pass | 25,931 | 5,523 | -79% | 1 | 1 | 0% | 1,313 | 1,662 | +27% | 0 | 0 | — |
case-08 | fail→pass | 5,351 | 2,180 | -59% | 1 | 1 | 0% | 1,055 | 834 | -21% | 0 | 0 | — |
case-09 | fail→fail | 3,368 | 1,753 | -48% | 1 | 1 | 0% | 631 | 795 | +26% | 0 | 0 | — |
case-10 | fail→pass | 7,156 | 2,465 | -66% | 1 | 1 | 0% | 1,453 | 890 | -39% | 0 | 0 | — |
case-11 | fail→pass | 4,071 | 3,560 | -13% | 1 | 1 | 0% | 769 | 1,093 | +42% | 0 | 0 | — |
case-12 | fail→pass | 4,546 | 2,688 | -41% | 1 | 1 | 0% | 979 | 1,096 | +12% | 0 | 0 | — |
case-13 | fail→fail | 5,870 | 3,172 | -46% | 1 | 1 | 0% | 1,061 | 1,100 | +4% | 0 | 0 | — |
case-14 | fail→pass | 8,930 | 5,881 | -34% | 1 | 1 | 0% | 1,845 | 1,624 | -12% | 0 | 0 | — |
case-15 | fail→pass | 7,951 | 4,005 | -50% | 1 | 1 | 0% | 1,446 | 1,263 | -13% | 0 | 0 | — |
case-16 | fail→pass | 7,310 | 1,257 | -83% | 1 | 1 | 0% | 1,515 | 619 | -59% | 0 | 0 | — |
case-17 | fail→pass | 7,669 | 4,994 | -35% | 1 | 1 | 0% | 1,394 | 857 | -39% | 0 | 0 | — |
case-18 | fail→pass | 13,143 | 3,770 | -71% | 1 | 1 | 0% | 2,004 | 1,110 | -45% | 0 | 0 | — |
case-19 | fail→pass | 4,490 | 2,174 | -52% | 1 | 1 | 0% | 936 | 846 | -10% | 0 | 0 | — |
case-20 | fail→pass | 9,509 | 1,344 | -86% | 1 | 1 | 0% | 1,949 | 690 | -65% | 0 | 0 | — |
case-21 | fail→pass | 5,695 | 2,092 | -63% | 1 | 1 | 0% | 1,118 | 844 | -25% | 0 | 0 | — |
case-22 | fail→pass | 4,306 | 2,209 | -49% | 1 | 1 | 0% | 857 | 920 | +7% | 0 | 0 | — |
case-23 | fail→pass | 5,318 | 1,509 | -72% | 1 | 1 | 0% | 995 | 740 | -26% | 0 | 0 | — |
case-24 | fail→fail | 5,756 | 1,810 | -69% | 1 | 1 | 0% | 1,053 | 802 | -24% | 0 | 0 | — |
case-25 | fail→pass | 7,978 | 3,072 | -61% | 1 | 1 | 0% | 1,271 | 1,034 | -19% | 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. 25 cases were attempted, and 21 counted toward the lift figure. The other 4 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 +64 percentage points is the difference between those two pass rates over the 21 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.