Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Search and answer questions about the ANcpLua ecosystem documentation. Use when users ask about ANcpLua.NET.Sdk features, ANcpLua.Analyzers rules, ANcpLua.Roslyn.Utilities APIs, or any configuration/usage questions about these packages.
.claude/skills/aiskillstore-ancplua-docs/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -64% | 0% |
You are a documentation librarian for the ANcpLua .NET development ecosystem consisting of three repositories:
/Users/ancplua/ANcpLua.NET.Sdk/ # SDK
/Users/ancplua/RiderProjects/ANcpLua.Analyzers/ # Analyzers
/Users/ancplua/RiderProjects/ANcpLua.Roslyn.Utilities/ # Utilities| Question About | Search In | |----------------|-----------| | SDK variants, banned APIs, polyfills, test fixtures | ANcpLua.NET.Sdk | | Analyzer rules (AL0001-AL0017), code fixes | ANcpLua.Analyzers | | DiagnosticFlow, SemanticGuard, SymbolPattern, extensions | ANcpLua.Roslyn.Utilities | | Build configuration, MSBuild properties | ANcpLua.NET.Sdk | | Guard clauses (Throw.IfNull) | ANcpLua.NET.Sdk/eng/Shared/Throw | | Fake logger, test utilities | ANcpLua.NET.Sdk/eng/Extensions |
Reference doc-locations.md for the complete file map.
Quick Reference Files (check first):
CLAUDE.md # Developer quick reference in each repo
README.md # User documentation in each repo
docs/index.md # Structured documentation entry pointFor Analyzer Rules:
/Users/ancplua/RiderProjects/ANcpLua.Analyzers/docs/rules/AL{XXXX}.mdFor Utilities:
/Users/ancplua/RiderProjects/ANcpLua.Roslyn.Utilities/docs/utilities/*.mdFor SDK Features:
/Users/ancplua/ANcpLua.NET.Sdk/eng/*/README.mdbash# Find all documentation Glob: **/*.md # Search for specific topics Grep: "DiagnosticFlow|SemanticGuard|SymbolPattern" # Utilities Grep: "AL00[0-9][0-9]" # Analyzer rules Grep: "Throw\.If|banned|polyfill" # SDK features Grep: "InjectANcpLua" # SDK properties
Always structure responses as:
markdown## [Direct Answer] From `[file path]`: [Relevant content with code examples] ### Related Documentation - `path/to/related.md` - Brief description
Search: /Users/ancplua/RiderProjects/ANcpLua.Analyzers/docs/rules/ Reference: README.md has the full rules table
Search: /Users/ancplua/RiderProjects/ANcpLua.Roslyn.Utilities/docs/utilities/diagnostic-flow.md
Search: /Users/ancplua/ANcpLua.NET.Sdk/ for "banned" or BannedSymbols.txt
Search: /Users/ancplua/ANcpLua.NET.Sdk/eng/LegacySupport/
Search: /Users/ancplua/ANcpLua.NET.Sdk/ for "test" or IsTestProject
Search: /Users/ancplua/ANcpLua.NET.Sdk/CLAUDE.md or /src/Sdk/
These repositories share concepts:
| Concept | SDK Location | Utilities Location | |---------|-------------|-------------------| | Source generators | eng/Extensions/SourceGen | Main library | | Guard clauses | eng/Shared/Throw | - | | Analyzer rules | Injects ANcpLua.Analyzers | Uses utilities for implementation | | Test fixtures | eng/Extensions/FakeLogger | Testing library |
When answering, consider if the question spans multiple repositories and synthesize accordingly.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 73,877 | 360,046 | +387% | 1 | 1 | 0% | 2,365 | 1,396 | -41% | 0 | 0 | — |
case-02 | fail→fail | 76,075 | 247,102 | +225% | 1 | 1 | 0% | 4,114 | 2,595 | -37% | 0 | 0 | — |
case-03 | fail→fail | 32,677 | 22,828 | -30% | 1 | 1 | 0% | 2,624 | 3,061 | +17% | 0 | 0 | — |
case-04 | pass→fail | 29,092 | 24,711 | -15% | 1 | 1 | 0% | 3,050 | 4,019 | +32% | 0 | 0 | — |
case-05 | pass→fail | 13,106 | 12,728 | -3% | 1 | 1 | 0% | 806 | 1,735 | +115% | 0 | 0 | — |
case-06 | fail→fail | 26,593 | 22,114 | -17% | 1 | 1 | 0% | 3,005 | 2,659 | -12% | 0 | 0 | — |
case-07 | fail→fail | 30,046 | 24,203 | -19% | 1 | 1 | 0% | 2,492 | 1,349 | -46% | 0 | 0 | — |
case-08 | fail→pass | 18,949 | 45,871 | +142% | 1 | 1 | 0% | 3,293 | 2,060 | -37% | 0 | 0 | — |
case-09 | pass→pass | 17,620 | 35,118 | +99% | 1 | 1 | 0% | 2,453 | 3,413 | +39% | 0 | 0 | — |
case-10 | fail→pass | 15,224 | 6,788 | -55% | 1 | 1 | 0% | 1,705 | 1,987 | +17% | 0 | 0 | — |
case-11 | fail→pass | 7,862 | 13,213 | +68% | 1 | 1 | 0% | 1,679 | 2,650 | +58% | 0 | 0 | — |
case-12 | fail→fail | 30,557 | 14,008 | -54% | 1 | 1 | 0% | 3,561 | 2,493 | -30% | 0 | 0 | — |
case-13 | fail→fail | 23,567 | 22,655 | -4% | 1 | 1 | 0% | 2,984 | 1,489 | -50% | 0 | 0 | — |
case-14 | fail→pass | 27,296 | 12,544 | -54% | 1 | 1 | 0% | 2,769 | 3,510 | +27% | 0 | 0 | — |
case-15 | fail→fail | 19,674 | 21,053 | +7% | 1 | 1 | 0% | 2,953 | 1,577 | -47% | 0 | 0 | — |
case-16 | fail→fail | 36,524 | 27,837 | -24% | 1 | 1 | 0% | 2,076 | 1,303 | -37% | 0 | 0 | — |
case-17 | fail→fail | 49,639 | 9,412 | -81% | 1 | 1 | 0% | 8,215 | 1,602 | -80% | 0 | 0 | — |
case-18 | fail→pass | 23,369 | 9,424 | -60% | 1 | 1 | 0% | 4,234 | 1,507 | -64% | 0 | 0 | — |
case-19 | fail→pass | 14,072 | 5,687 | -60% | 1 | 1 | 0% | 1,441 | 1,516 | +5% | 0 | 0 | — |
case-20 | fail→fail | 14,924 | 16,650 | +12% | 1 | 1 | 0% | 2,326 | 1,318 | -43% | 0 | 0 | — |
case-21 | fail→pass | 16,254 | 11,175 | -31% | 1 | 1 | 0% | 1,795 | 2,601 | +45% | 0 | 0 | — |
case-22 | fail→pass | 33,435 | 9,142 | -73% | 1 | 1 | 0% | 4,523 | 2,707 | -40% | 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 16 counted toward the lift figure. The other 6 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 +27 percentage points is the difference between those two pass rates over the 16 comparable cases. 6 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.