[skills-eval] dotnet-diag: 7 skills, 49% pass — 1 P0, 1 P1, 3 to strengthen, 2 keep

> ## 🔄 EDITED 2026-07-19 — correction per #909
>
> **This finding has been revised.** The errored trials that drove the **FIX-RELIABILITY (P0)** flag(s) below were shown in [#909](https://github.com/dotnet/skills/issues/909) to be **judge-side infrastructure failures** — a disabled/throttled PAT (`CAPIError 400 organization disabled`) and Vally's `session.idle` judge timeout — **not** fixture nondeterminism. Verdict scores already exclude errored trials, so re-running the classifier with those judge errors dropped reclassifies the affected skill(s):
>
> | Skill | ~~Was~~ | Now |
> | --- | --- | --- |
> | `analyzing-dotnet-performance` | ~~FIX-RELIABILITY (P0)~~ | **KEEP-POLISH (keep)** |
> | `android-tombstone-symbolication` | ~~FIX-RELIABILITY (P0)~~ | **FIX-REGRESSION (P0, real)** |
>
> Struck-through text below is the **superseded** finding — in particular the "pin fixture SDK/tool versions" remediation does **not** apply. The title's P0 count has been updated. No eval re-run is required to correct the scores; recovering the lost judgments (optional) means **re-judging** those trials, not re-running the skill.

## Context — cross-family skill evaluation: `dotnet-diag`

This issue is **self-contained**: it captures everything a skill author needs to act on the `dotnet-diag` plugin without opening the full report.

**What this measures.** Every runnable skill in `dotnet/skills` was run through [Vally](https://github.com/microsoft/evaluate) (0.7) on a **cross-family matrix**: 5 executor model families — `opus-4.8`, `gpt-5.5`, `sonnet-4.6`, `haiku-4.5`, `mai-flash` — each judged by a **different family** (judge ≠ executor; default judge = latest Opus, or GPT when Opus is the executor). For every skill, a **skilled** run is compared against a **baseline** (no-skill) run and scored per executor. This removes single-model and self-judging bias, so a skill that only helps one model family — or only its own family's judge — is visible.

- **Data source:** cross-family CI grid (run `29228914412` + backfills) — 5 executors × 85 runnable skills, **419 scored cells** over 84 skills.
- **Row grain:** one row per **skill**, aggregated across its (up to 5) executor cells. `avgN` is the mean trial count behind the cells (trials 1–17; higher = more statistically trustworthy). `thin-N` flags directional-only rows.

### `dotnet-diag` at a glance (portfolio scorecard)

| Plugin | Skills | Cells | Pass | Impact | Tie-trials | Err | avg ΔTok | avg ΔTurns | avg ΔTools | Headline |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | --- |
| **dotnet-diag** | 7 | 35 | 49% | 0.362 | 33 | 2 | +35,515 | +0.74 | +2.50 | Half-inert (33 ties); regressions in symbolication |

**Insights.** 7 skill(s); mean impact **0.36**; 4/7 help ≥1 frontier model. Exemplars (win incl. frontier): `dotnet-trace-collect`. Weakest: `android-tombstone-symbolication` (0.16).

~~**Address first:** `analyzing-dotnet-performance` — reliability; `android-tombstone-symbolication` — reliability; `microbenchmarking` — cost.~~

**Address first:** `android-tombstone-symbolication` — **regression** (reclassified from reliability); `microbenchmarking` — cost. _(`analyzing-dotnet-performance` reclassified out of P0 — see EDIT above.)_

### How to read the table

Each skill is scored **skilled vs. baseline** on these axes:

| Signal | Column | What it means | Good |
| --- | --- | --- | --- |
| Breadth | `Families ✓` (n/5) | How many of the **5 model families** the skill helps (per-family pass count, max `5/5`) | 4–5 / 5 |
| Where | `Passed on` | *Which* families passed. **Frontier = latest Opus + latest GPT** are **bold**; Sonnet 4.6 / Haiku 4.5 / MAI Flash are mid/low-weight | frontier ✓ |
| Magnitude | `Impact` (−1…+1) | How strongly the judge prefers *skilled* over *baseline* | ≥ 0.4 |
| Decisiveness | `Ties▵` | Trials where the judge saw **no difference** → skill is *inert* | low |
| Safety | `Loss▵` | Trials where **skilled was WORSE** than baseline → skill misfires | ~0 |
| Reliability | `Err` | Trials that errored/crashed in setup or judging | 0 |
| Efficiency | `ΔTok` / `ΔTurns` / `ΔTools` | Extra tokens / agent turns / tool calls vs baseline | ≤ 0 |
| Confidence | `avgN` | Mean trials behind the verdict; low N = directional only | ≥ 3 |
| Invocation | `Call%` | Share of skilled trials where the model actually **invoked the skill** | ~100% |

> **`Families ✓` is cell-level (max `5/5`); `Ties▵`/`Loss▵` are trial-level tallies summed across *all* families** (including the ones where the skill failed). A high `Families ✓` next to non-zero `Loss▵` is not a contradiction — see `Passed on` and the Action text for where losses landed.

**Action buckets** (each skill has one primary action; `[flags]` note secondary concerns):

| Bucket | Priority | Meaning |
| --- | --- | --- |
| FIX-RELIABILITY | 🔴 P0 | Errored trials / no verdict — stabilize the harness before trusting the score |
| FIX-DISCOVERY | 🔴 P0 | Model doesn't invoke it (`Call% < 50%`) — a triggering/description problem |
| FIX-REGRESSION | 🔴 P0 | Skilled is worse than baseline on many trials — the skill misfires |
| ADD-DECISIVENESS | 🟠 P1 | Called ~100% but ties dominate, ~0 impact — inert; needs sharper behavioral steps |
| TRIM-COST | 🟠 P1 | Passes but with heavy token/turn overhead — trim verbosity |
| EXEMPLAR | 🟢 keep | Broad, strong, reliable win — use as a template |
| EFFICIENT-WIN | 🟢 protect | Wins *and* cuts turns/tools — the ideal shape |
| KEEP-POLISH | 🟢 | Solid majority win; minor polish + more trials |
| STRENGTHEN | 🟡 P2 | Marginal/mixed lift — sharpen triggers & success criteria |

### Per-skill actions

| Skill | Families ✓ | Passed on (frontier **bold**) | Impact | Ties▵ | Loss▵ | Err | avgN | Call% | ΔTok | ΔTurns | ΔTools | Action |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `dotnet-trace-collect` | 4/5 | **Opus**, Sonnet, Haiku, MAI | 0.53 | 8 | 8 | 0 | 17 | 97% | +30,164 | +1.48 | +1.6 | **EXEMPLAR** · ✅ **Template-worthy.** What's good: broad (passes 4/5, incl. frontier Opus). Keep as-is; lift its structure (crisp triggers + imperative steps) into weaker siblings. ⚠️ Caveat: frontier miss on GPT — confirm it isn't only lifting weaker models on cases frontier already handles. *([frontier-miss])* |
| `clr-activation-debugging` | 2/5 | **Opus**, MAI | 0.44 | 2 | 5 | 0 | 6.8 | 94% | +39,418 | +1.09 | +2.36 | **STRENGTHEN** · 🟡 **Marginal/mixed** (passes 2/5 — **Opus**, MAI). Diagnosis: model-dependent — helps some families, not others. Try: (1) sharpen the trigger so it fires only where it wins; (2) add 1–2 opinionated, concrete steps that change behaviour; (3) add trials to separate signal from noise. *([frontier-miss])* |
| `microbenchmarking` | 4/5 | **Opus**, **GPT**, Sonnet, MAI | 0.44 | 1 | 0 | 0 | 1 | 100% | +237,705 | +5.6 | +9.2 | **TRIM-COST** · 🟠 **Passes but heavy (+237,705 tok).** Value is real; cost isn't justified. Trim: (1) cut redundant/whole-file reads — point to specific sections; (2) replace "explore everything" with a targeted checklist; (3) move deep reference material behind links instead of inlining it. *([costly], [thin-N])* |
| `dump-collect` | 2/5 | Sonnet, Haiku | 0.37 | 8 | 8 | 0 | 9 | 80% | +18,779 | +1.11 | +1.6 | **STRENGTHEN** · 🟡 **Marginal/mixed** (passes 2/5 — Sonnet, Haiku). Diagnosis: mostly *ties* — too generic/non-prescriptive. Try: (1) sharpen the trigger so it fires only where it wins; (2) add 1–2 opinionated, concrete steps that change behaviour; (3) add trials to separate signal from noise. If frontier models never benefit, scope it explicitly to weaker models or reconsider its value. *([both-frontier-miss])* |
| `analyzing-dotnet-performance` | 3/5 | **GPT**, Haiku, MAI | 0.33 | 7 | 4 | 1 | 9.6 | 96% | +5,866 | -0.73 | +6.33 | ~~**FIX-RELIABILITY** · 🔴 **1 errored trial(s)** — verdict can't be trusted until setup is deterministic. This is an infra/harness fix, not a content one: (1) capture the failing trial's stderr; (2) pin tool/SDK versions in the fixture; (3) add a setup smoke-check before scoring. *([reliability], [frontier-miss])*~~

🔄 **EDIT (see #909):** **KEEP-POLISH** · 🟢 Solid majority win (3/5 — GPT, Haiku, MAI). The 1 errored trial was a **judge-side `session.idle` timeout** (sonnet46), not fixture flakiness. Investigate the Opus miss; minor polish. *([frontier-miss])* |
| `apple-crash-symbolication` | 1/5 | Sonnet | 0.26 | 5 | 1 | 0 | 2.8 | 67% | -75,504 | -3.33 | -3.37 | **STRENGTHEN** · 🟡 **Marginal/mixed** (passes 1/5 — Sonnet). Diagnosis: mostly *ties* — too generic/non-prescriptive. Try: (1) sharpen the trigger so it fires only where it wins; (2) add 1–2 opinionated, concrete steps that change behaviour; (3) add trials to separate signal from noise. If frontier models never benefit, scope it explicitly to weaker models or reconsider its value. *([both-frontier-miss])* |
| `android-tombstone-symbolication` | 1/5 | Haiku | 0.16 | 2 | 13 | 1 | 6.6 | 88% | -7,821 | -0.08 | -0.2 | ~~**FIX-RELIABILITY** · 🔴 **1 errored trial(s)** — verdict can't be trusted until setup is deterministic. This is an infra/harness fix, not a content one: (1) capture the failing trial's stderr; (2) pin tool/SDK versions in the fixture; (3) add a setup smoke-check before scoring. *([reliability], [both-frontier-miss], [frontier-regressed])*~~

🔄 **EDIT (see #909):** **FIX-REGRESSION** · 🔴 Still P0 — but the real problem is **regression, not reliability**. The 1 errored trial was a judge-side disabled-PAT failure in the **mai** cell; removing it *unmasks* losses on **both frontier models (Opus, GPT)** — cells that had **no** errored trials (aggregate loss ~39%). Add explicit stop-conditions and narrow the trigger. *([both-frontier-miss], [frontier-regressed])* |
---

_{Generated from the cross-family Call-to-Action report (`CALL-TO-ACTION.md` §4–§5; companion `IMPACT-ANALYSIS.md`). Regenerate the underlying tables with `node deep-metrics.mjs "$env:TEMP\cf-ci" agg-ci` → `node gen-cta-tables.mjs agg-ci`. Numbers are directional where `avgN` is low; treat single-trial cells as hypotheses to confirm with more runs.}