Deep Inspect
Use when a dotnet-inspect change needs expensive evidence outside normal PR CI, or when preparing release certification; coordinate Deep Inspect lanes (full slow tests, cross-platform certification, IL round-trip sweep, corpus sensors, package discovery, validity scans, and analysis census).
Get the complete skill folder, including its scripts and reference files.
When to use this skill
Use Deep Inspect when an AI agent needs a reusable SKILL.md workflow for this job: Use when a dotnet-inspect change needs expensive evidence outside normal PR CI, or when preparing release certification; coordinate Deep Inspect lanes (full slow tests, cross-platform certification, IL round-trip sweep, corpus sensors, package discovery, validity scans, and analysis census).
When not to use it
Skip Deep Inspect when the task is outside the analysis category, or when a more specific skill in this directory already covers the same workflow with clearer triggers.
How to install
Where will you use it?
mkdir -p ~/.claude/skills/deep-inspect # Save SKILL.md to ~/.claude/skills/deep-inspect/SKILL.md
- Claude Code personal install: place the skill and its bundled files at ~/.claude/skills/deep-inspect/SKILL.md. For a shared project, commit the folder at .claude/skills/deep-inspect/ instead.
- Current Claude Code detects SKILL.md changes in watched directories during the session. If you just created a top-level skills directory, run /reload-skills. Check that the skill is listed before testing its trigger.
- For Claude chat or Cowork in the Desktop app, manage uploaded skills in Customize > Skills; copying files into a local Claude Code folder is not a Desktop install. Other agents may use different install locations.
Full install guide for Claude, Cursor, and Codex →
Try a starter prompt
After installing the skill, paste this starter prompt into your agent to verify it triggers on a small task.
Use the Deep Inspect skill for this task: Use Deep Inspect when this documented workflow matches the task. Confirm any required inputs or files before making changes.
What supports this recommendation?
70/100 provisional content score. This measures documentation, practical detail and source signals. It is not a measured task success rate or a security certification.
- Documentation
- 77/100
- Practical detail
- 68/100
- Source evidence
- 60/100
- Source signals
- 72/100
Model compatibility evidence
- claudeuntested
No model-specific signal or recorded compatibility test was found.
- chatgptuntested
No model-specific signal or recorded compatibility test was found.
- geminiuntested
No model-specific signal or recorded compatibility test was found.
- copilotuntested
No model-specific signal or recorded compatibility test was found.
- llamauntested
No model-specific signal or recorded compatibility test was found.
- perplexityuntested
No model-specific signal or recorded compatibility test was found.
- mistraluntested
No model-specific signal or recorded compatibility test was found.
- grokuntested
No model-specific signal or recorded compatibility test was found.
Inferred compatibility is based on content signals. Check the app, its available tools and setup requirements before running a skill.
What this skill does
# Deep Inspect
This is a repo-local maintainer skill (CI/certification harness), not a shipped end-user capability; it is not embedded in the dotnet-inspect binary.
Use this skill when a change needs expensive evidence outside normal PR CI. Deep Inspect is opt-in for risky PRs. Its `test`, `platform-test`, and decompiler-corpus jobs run daily to certify a commit for release. Publish consumes that certification evidence rather than rerunning the slow suites. The `authored-corpus` ratchet runs on a separate daily schedule as a regression gate rather than release-certification or publish evidence. The comprehensive Inspect Web lane is also daily regression evidence rather than release certification. Every lane can be dispatched on demand during the day.
## Lanes
| Lane | Use for | Runs | | ---- | ------- | ---- | | `test` | Daily/on-demand release certification or blocking proof before risky merges | Full decompiler tests, full analysis tests, vendored ILAssembler restore, full IL round-trip sweep. | | `platform-test` | Daily/on-demand release certification across Windows, macOS, and Ubuntu | Reduced cross-platform suite: CLI, CSharpText, artifact, fast decompiler, NuGetFetch offline, metadata, services, query, and Research tests, plus `ilasm`/`ildasm`/`mdv` setup. | | `census` | Observational broad signal and triage | Real-world corpus sensor, validity predicate scan, uncapped validity sweep, assertion scan, analysis corpus sensor, paydirt recall. | | `package-sweep` | Weekly/on-demand discovery over current top NuGet packages | Product-backed package acquisition plus bounded per-library fully-raised, validity, defect-class, and promotion-candidate reporting. | | `authored-corpus` | Daily/on-demand regression ratchet against checksum-verified authored source | Restores the pinned authored-source corpus and fails on quality regression or measurement-integrity loss. | | `inspect-web` | Daily/on-demand comprehensive Browser/Wasm regression evidence | Runs generated-facade version invariance, mutation controls, and Mono/CoreCLR multi-facade and managed-operation canaries. | | `nightly` | Opt-in next-SDK/compiler validation | Builds with the .NET daily SDK and checks opt-in compiler lowering drift; intentionally excluded from `all`. | | `all` | Release-candidate deep read | The `test`, `platform-test`, decompiler-corpus, `census`, `authored-corpus`, and `inspect-web` lanes. |
Run manually:
```bash gh workflow run deep-inspect.yml -f lane=test gh workflow run deep-inspect.yml -f lane=platform-test gh workflow run deep-inspect.yml -f lane=census gh workflow run deep-inspect.yml -f lane=package-sweep gh workflow run deep-inspect.yml -f lane=authored-corpus gh workflow run deep-inspect.yml -f lane=inspect-web gh workflow run deep-inspect.yml -f lane=nightly gh workflow run deep-inspect.yml -f lane=all ```
Inspect recent runs and artifacts:
```bash gh run list --workflow deep-inspect.yml --limit 10 gh run view <run-id> --log-failed gh run download <run-id> -D /tmp/deep-inspect-<run-id> ```
## Local equivalents
For the test lane:
```bash dotnet build dotnet-inspect.slnx -c Release dotnet run --project tests/DotnetInspect.Cli.Tests -c Release source eng/activate-iltools.sh dotnet run --project tests/ILInspector.Decompiler.Tests -c Release -- --gate no-corpus dotnet run --project tests/ILInspector.Analysis.Tests -c Release bash eng/restore-ilassembler.sh dotnet run --project tests/DotnetInspector.ILRoundtrip.Tests -c Release dotnet run --project tests/ILInspector.Decompiler.Tests -c Release -- --gate corpus ```
The corpus command runs as a separate workflow job and can take hours. Omit it only when intentionally reproducing the non-corpus `test` job rather than the complete dispatched `test` lane.
A successful daily or manually dispatched certification requires the `test`, `platform-test`, and decompiler-corpus jobs at one exact `main` SHA. Publish requires that certification run ID. Publishing a later descendant remains an explicit operator decision. Its exact main-push `ci-required` result must succeed, but main-push CI does not run the PR-only substantive test jobs, and the certification does not claim to cover intervening changes.
The `platform-test` lane is workflow-owned and runs the reduced cross-platform suite on Windows, macOS, and Ubuntu. When reproducing a platform-only break locally, mirror the exact project list and tool activation from `.github/workflows/deep-inspect.yml` for the affected platform.
For the census lane, prefer the workflow so artifacts are retained. If running locally, use the same scripts/baselines as `deep-inspect.yml` and preserve the generated snapshots/cards under `/tmp` or `artifacts/` for review.
The package sweep runs every Monday at 09:00 UTC and can also be dispatched manually. It is owned by `@richlander`, is discovery-only, and never gates a pull request. Each run resolves the latest stable versions for ranks 1-10, records exact package/version/TFM provenance, and samples at most 250 methods and 25 semantic-validity candidates per selected library. Promote a package to an existing pinned corpus only after a reported defect or unsupported shape is accepted for ongoing coverage.
## Reading results
- Treat `test`, `platform-test`, and decompiler-corpus failures as blockers: reproduce locally, identify the first failing proof, and fix it before certification can authorize publish. - Treat `census` output as triage signal unless a command exits nonzero by design. Compare snapshots against committed baselines and route meaningful drift to issues or follow-up PRs. - Treat `inspect-web` failures as Browser/Wasm regression blockers. Ordinary PRs run the fast boundary modes; changes to the facade generators, canaries, or managed bridge owners run the complete modes before merge. - Do not add broad/corpus-style tests to PR CI. Mark them `[Trait("Speed", "Slow")]` and keep them in Deep Inspect / full local runs.
Intended uses
- Use Deep Inspect when this documented workflow matches the task.
Implementation notes
- Review the source instructions and adapt inputs before running the workflow.
Try it, then tell us how it went
A copied prompt does not tell us whether the task worked. Report your result after checking the output. We collect your tool and outcome, never your files or prompt.