/health-check — 8-Dimension Project Assessment
What
Runs a data-driven health assessment across 8 dimensions, each graded A-F with
the specific data points that produced the grade, and rolls them into a GPA.
Gut feeling is not a grade: every dimension uses MCP tools or CLI commands, and
every grade below A comes with specific, prioritized, effort-estimated fixes —
"add test classes for OrderService, PaymentProcessor, ShippingCalculator" is
actionable; "improve test coverage" is not.
This skill owns the canonical grading system for the kit. The full rubrics,
GPA scale, and report template live in references/grading-rubric.md — load
that file when running an assessment.
Tone is diagnostic, not punitive: a C grade is an improvement path, not a
failure.
When
- Onboarding to an unfamiliar or new project — set the baseline
- "How healthy is this?", "grade this project", "codebase review", "report card"
- Pre-release quality gate, or monthly/quarterly maintenance review
- After a cleanup sprint (
/de-sloppify) — re-grade to show progress
- Tech-debt prioritization — lowest grades get the next sprint's attention
How
Step 1: Choose Scope
| Scenario |
Dimensions |
| Full assessment (onboarding, pre-release, monthly review) |
All 8 |
| Quick health (mid-sprint checkpoint, before a demo, after a merge) |
1-4 only |
| After major refactor |
1 (Build), 3 (Architecture), 4 (Tests) |
| Post-dependency update |
1 (Build), 7 (Security) |
| After cleanup sprint |
Re-grade only the cleaned dimensions |
Step 2: Run the Dimensions
Read references/grading-rubric.md for the grade thresholds, then collect data
per dimension. For deep code-quality dimensions, delegate to the
code-reviewer agent with the code-review skill.
| # |
Dimension |
Data source |
| 1 |
Build Health |
dotnet build --no-restore — errors + warnings |
| 2 |
Code Quality |
MCP detect_antipatterns — read summary, grade high-confidence only |
| 3 |
Architecture |
MCP get_project_graph + detect_circular_dependencies (projects AND types) |
| 4 |
Test Coverage |
MCP get_test_coverage_map — check applicable first (structural, not line coverage) |
| 5 |
Dead Code |
MCP find_dead_code(scope: "solution") — grade high-confidence; ignore conventionFiltered |
| 6 |
API Surface |
MCP get_public_api + find_references — overexposure, return-type consistency |
| 7 |
Security Posture |
dotnet list package --vulnerable --include-transitive + secrets/auth spot check (deep dive: /security-scan) |
| 8 |
Documentation |
XML doc coverage on public APIs + README currency |
Step 2.5: Triage Gate (before any grade is assigned)
Detector output is evidence, not a grade. Pass every finding through this gate
first — it is what stops a noisy count becoming a wrong letter.
- Read
summary, not the violation list. summary.byId is complete even
when the list is truncated. Never sample a truncated list and extrapolate.
- Drop
suppressed. Record the count and summary.suppressionConfig in the
report so suppression stays visible.
- Set aside
medium. These are review items, not grade inputs. Summarise
them by category; do not fix or count them.
- Grade
high only. These are wrong regardless of context.
- Check invariants. If a signal contradicts something the target repo's
CLAUDE.md documents as deliberate, the invariant wins — verify before grading.
- Read before asserting. Never describe a finding you have not opened. If
you cannot open all of them, report the ones you did and say so.
Fill in the triage table from references/grading-rubric.md as you go. If the
table cannot be filled, the dimension is not ready to grade.
Step 3: Grade and Aggregate
Apply the rubric thresholds exactly — never grade on a curve ("pretty good for
a project this size" is how standards erode; 15 warnings is a C regardless of
project size). Convert to GPA (A=4.0 … F=0.0), averaging only the dimensions
actually graded — a dimension marked "Not assessed" is excluded from the GPA,
never scored as an F.
Step 4: Report
Produce the report card from the template in references/grading-rubric.md:
grades table with key findings, overall GPA, and priority recommendations —
each with named types/files, priority order, and effort estimates. If a
previous report exists, append the trend comparison table.
Example
User: /health-check
Claude: Running full 8-dimension assessment...
| Dimension | Grade | Key Finding |
|-----------|-------|-------------|
| Build Health | A | 0 errors, 2 warnings |
| Code Quality | B | 3 high-confidence findings in 4.2K lines; 31 medium untriaged |
| Architecture | A | Clean direction, 0 cycles |
| Test Coverage | Not assessed | Integration-driven suite — structural metric invalid |
| Dead Code | B | 5 unused methods (79 convention-discovered, not counted) |
| API Surface | B | 2 overexposed service types |
| Security | A | 0 vulnerable packages |
| Documentation | D | 12/30 public APIs documented |
Overall GPA: 3.1 (B) — averaged over 7 graded dimensions.
Triage: 44 AP005 raw → all log-and-rethrow wrappers (medium); 2 AP004 real.
Priority: (1) `SystemSeeder` → `TimeProvider`, ~15 min; (2) XML docs on the 8
endpoint classes, ~1 day; (3) review the 44 catch blocks or suppress by path.
Related
references/grading-rubric.md — canonical rubrics, GPA scale, report template/de-sloppify — cleanup pipeline for the issues a health check surfaces/security-scan — deep 6-layer scan behind Dimension 7/code-review — per-change review (this skill grades the whole project)/verify — pass/fail pipeline for a change set, not a graded assessment
1---2name: health-check3description: Multi-dimensional health assessment for .NET projects with letter grades (A-F) using Roslyn MCP tools. Evaluates 8 dimensions: build health, code quality, architecture, test coverage, dead code, API surface, security posture, and documentation. Produces a structured report card with actionable recommendations. Load this skill when: "health check", "how healthy is this", "project health", "code quality report", "grade this project", "assess codebase", "quality audit", "technical assessment", "codebase review", "report card".4---5
6# /health-check — 8-Dimension Project Assessment
7
8## What
9
10Runs a data-driven health assessment across 8 dimensions, each graded A-F with
11the specific data points that produced the grade, and rolls them into a GPA.
12Gut feeling is not a grade: every dimension uses MCP tools or CLI commands, and
13every grade below A comes with specific, prioritized, effort-estimated fixes —
14"add test classes for OrderService, PaymentProcessor, ShippingCalculator" is
15actionable; "improve test coverage" is not.
16
17This skill owns the **canonical grading system** for the kit. The full rubrics,
18GPA scale, and report template live in `references/grading-rubric.md` — load
19that file when running an assessment.
20
21Tone is diagnostic, not punitive: a C grade is an improvement path, not a
22failure.
23
24## When
25
26- Onboarding to an unfamiliar or new project — set the baseline
27- "How healthy is this?", "grade this project", "codebase review", "report card"
28- Pre-release quality gate, or monthly/quarterly maintenance review
29- After a cleanup sprint (`/de-sloppify`) — re-grade to show progress
30- Tech-debt prioritization — lowest grades get the next sprint's attention
31
32## How
33
34### Step 1: Choose Scope
35
36| Scenario | Dimensions |
37|----------|------------|
38| Full assessment (onboarding, pre-release, monthly review) | All 8 |
39| Quick health (mid-sprint checkpoint, before a demo, after a merge) | 1-4 only |
40| After major refactor | 1 (Build), 3 (Architecture), 4 (Tests) |
41| Post-dependency update | 1 (Build), 7 (Security) |
42| After cleanup sprint | Re-grade only the cleaned dimensions |
43
44### Step 2: Run the Dimensions
45
46Read `references/grading-rubric.md` for the grade thresholds, then collect data
47per dimension. For deep code-quality dimensions, delegate to the
48`code-reviewer` agent with the `code-review` skill.
49
50| # | Dimension | Data source |
51|---|-----------|-------------|
52| 1 | Build Health | `dotnet build --no-restore` — errors + warnings |
53| 2 | Code Quality | MCP `detect_antipatterns` — read `summary`, grade high-confidence only |
54| 3 | Architecture | MCP `get_project_graph` + `detect_circular_dependencies` (projects AND types) |
55| 4 | Test Coverage | MCP `get_test_coverage_map` — **check `applicable` first** (structural, not line coverage) |
56| 5 | Dead Code | MCP `find_dead_code(scope: "solution")` — grade high-confidence; ignore `conventionFiltered` |
57| 6 | API Surface | MCP `get_public_api` + `find_references` — overexposure, return-type consistency |
58| 7 | Security Posture | `dotnet list package --vulnerable --include-transitive` + secrets/auth spot check (deep dive: `/security-scan`) |
59| 8 | Documentation | XML doc coverage on public APIs + README currency |
60
61### Step 2.5: Triage Gate (before any grade is assigned)
62
63Detector output is evidence, not a grade. Pass every finding through this gate
64first — it is what stops a noisy count becoming a wrong letter.
65
661. **Read `summary`, not the violation list.** `summary.byId` is complete even
67 when the list is truncated. Never sample a truncated list and extrapolate.
682. **Drop `suppressed`.** Record the count and `summary.suppressionConfig` in the
69 report so suppression stays visible.
703. **Set aside `medium`.** These are review items, not grade inputs. Summarise
71 them by category; do not fix or count them.
724. **Grade `high` only.** These are wrong regardless of context.
735. **Check invariants.** If a signal contradicts something the target repo's
74 CLAUDE.md documents as deliberate, the invariant wins — verify before grading.
756. **Read before asserting.** Never describe a finding you have not opened. If
76 you cannot open all of them, report the ones you did and say so.
77
78Fill in the triage table from `references/grading-rubric.md` as you go. If the
79table cannot be filled, the dimension is not ready to grade.
80
81### Step 3: Grade and Aggregate
82
83Apply the rubric thresholds exactly — never grade on a curve ("pretty good for
84a project this size" is how standards erode; 15 warnings is a C regardless of
85project size). Convert to GPA (A=4.0 … F=0.0), averaging only the dimensions
86actually graded — a dimension marked "Not assessed" is excluded from the GPA,
87never scored as an F.
88
89### Step 4: Report
90
91Produce the report card from the template in `references/grading-rubric.md`:
92grades table with key findings, overall GPA, and priority recommendations —
93each with named types/files, priority order, and effort estimates. If a
94previous report exists, append the trend comparison table.
95
96## Example
97
98```
99User: /health-check
100
101Claude: Running full 8-dimension assessment...
102
103| Dimension | Grade | Key Finding |
104|-----------|-------|-------------|
105| Build Health | A | 0 errors, 2 warnings |
106| Code Quality | B | 3 high-confidence findings in 4.2K lines; 31 medium untriaged |
107| Architecture | A | Clean direction, 0 cycles |
108| Test Coverage | Not assessed | Integration-driven suite — structural metric invalid |
109| Dead Code | B | 5 unused methods (79 convention-discovered, not counted) |
110| API Surface | B | 2 overexposed service types |
111| Security | A | 0 vulnerable packages |
112| Documentation | D | 12/30 public APIs documented |
113
114Overall GPA: 3.1 (B) — averaged over 7 graded dimensions.
115
116Triage: 44 AP005 raw → all log-and-rethrow wrappers (medium); 2 AP004 real.
117
118Priority: (1) `SystemSeeder` → `TimeProvider`, ~15 min; (2) XML docs on the 8
119endpoint classes, ~1 day; (3) review the 44 catch blocks or suppress by path.
120```
121
122## Related
123
124- `references/grading-rubric.md` — canonical rubrics, GPA scale, report template
125- `/de-sloppify` — cleanup pipeline for the issues a health check surfaces
126- `/security-scan` — deep 6-layer scan behind Dimension 7
127- `/code-review` — per-change review (this skill grades the whole project)
128- `/verify` — pass/fail pipeline for a change set, not a graded assessment
codewithmukesh