Skip to main content
Agent skill

Deep Inspect Analysis

Use for opt-in Deep Inspect run triage: test lane failures, census artifacts, pinned-subset drift, and routing findings into issues or tracker updates.

Get started

Get the complete skill folder, including its scripts and reference files.

When to use this skill

Use Deep Inspect Analysis when an AI agent needs a reusable SKILL.md workflow for this job: Use for opt-in Deep Inspect run triage: test lane failures, census artifacts, pinned-subset drift, and routing findings into issues or tracker updates.

When not to use it

Skip Deep Inspect Analysis when the task is outside the coding 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?

Install scope
Claude Code folder setup
mkdir -p ~/.claude/skills/deep-inspect-analysis
# Save SKILL.md to ~/.claude/skills/deep-inspect-analysis/SKILL.md
  1. Claude Code personal install: place the skill and its bundled files at ~/.claude/skills/deep-inspect-analysis/SKILL.md. For a shared project, commit the folder at .claude/skills/deep-inspect-analysis/ instead.
  2. 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.
  3. 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.

Starter prompt
Use the Deep Inspect Analysis skill for this task: Use Deep Inspect Analysis 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
76/100
Practical detail
68/100
Source evidence
60/100
Source signals
72/100
Model compatibility evidence
  • claudeinferred

    The skill text mentions Claude or a closely associated term. This is keyword-derived compatibility; no hands-on model test is recorded.

  • 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 analysis

Use this skill when the user asks for Deep Inspect analysis or wants an opt-in validation/census run reviewed. Produce a short status report: whether the test lane passed, whether census health moved, and whether any result needs a durable issue, tracker row, rebaseline note, or no-action note.

Run from the repository root and prefer `gh` for GitHub Actions and issue data.

## Read the latest runs

Start with recent `deep-inspect.yml` runs:

```bash gh run list --workflow deep-inspect.yml --limit 10 \ --json databaseId,status,conclusion,createdAt,updatedAt,event,headBranch,headSha,url ```

Pick the relevant completed run. Check whether it ran `lane=test`, `lane=census`, or `lane=all`, then classify failures by lane:

```bash gh run view <run-id> --json jobs,conclusion,status,createdAt,updatedAt gh run view <run-id> --log-failed ```

Test-lane failure classes: setup/build, full decompiler tests, full analysis tests, ILAssembler restore, IL round-trip sweep, cancelled/environmental. Census-lane failure classes: corpus prep, corpus sensor, validity scan, validity sweep, assertion scan, analysis corpus sensor, paydirt recall, artifact upload, cancelled/environmental.

## Read census artifacts

If the run produced `deep-inspect-census`, download it to `/tmp`:

```bash rm -rf /tmp/deep-inspect-<run-id> mkdir -p /tmp/deep-inspect-<run-id> gh run download <run-id> -n deep-inspect-census \ -D /tmp/deep-inspect-<run-id> ```

Summarize the corpus snapshot first:

```bash jq '{generatedUtc, validityCompileCap, fidelityCompileCap, methodCap, metrics}' \ /tmp/deep-inspect-<run-id>/deep-inspect/corpus-snapshot.json ```

Then list actionable rows:

```bash snapshot=/tmp/deep-inspect-<run-id>/deep-inspect/corpus-snapshot.json

printf '\nFULL_MALFORMED\n' jq -r '.methods[] | select((.validity // "") | startswith("full-malformed:")) | [.assembly,.displayMethod,.validity,.fidelity,.residual,.fidelityCheck] | @tsv' "$snapshot"

printf '\nSEMANTIC_DEFECT\n' jq -r '.methods[] | select((.validity // "") | startswith("semantic-defect:")) | [.assembly,.displayMethod,.validity,.fidelity,.residual,.fidelityCheck] | @tsv' "$snapshot"

printf '\nFIDELITY_OPCODE_DIFF\n' jq -r '.methods[] | select(.fidelityCheck == "OpcodeDiff") | [.assembly,.displayMethod,.validity,.fidelity,.residual,.fidelityCheck] | @tsv' "$snapshot"

printf '\nPASS_BUG\n' jq -r '.methods[] | select(.passBug != null) | [.assembly,.displayMethod,.passBug,.validity,.fidelity,.residual,.fidelityCheck] | @tsv' "$snapshot" ```

## Separate pinned signal from repo-growth drift

Aggregate counts include dotnet-inspect self assemblies, so repo growth can move totals without a decompiler regression. Treat the pinned NuGet subset as the stable regression signal and report aggregate drift separately.

```bash baseline=tools/DecompilerHarness/corpus/real-world-baseline.json snapshot=/tmp/deep-inspect-<run-id>/deep-inspect/corpus-snapshot.json

jq -n --slurpfile b "$baseline" --slurpfile c "$snapshot" ' def pinned($s): [ $s.methods[] | select(.assemblyPath | startswith("nuget:")) ] as $m | { total: ($m | length), fullyRaised: ($m | map(select(.fullyRaised)) | length), conditional: ($m | map(select(.residual == "structuring: conditional-branch")) | length), fullMalformed: ($m | map(select((.validity // "") | startswith("full-malformed:"))) | length), semantic: ($m | map(select((.validity // "") | startswith("semantic-defect:"))) | length), fidelityChecked: ($m | map(select(.fidelityCheck != "not-sampled")) | length), opcode: ($m | map(select(.fidelityCheck == "OpcodeDiff")) | length), recompile: ($m | map(select(.fidelityCheck == "RecompileFail")) | length), context: ($m | map(select(.fidelityCheck == "ContextFail")) | length) }; {baseline: pinned($b[0]), current: pinned($c[0])}' ```

If pinned counts moved, drill into per-method changes before calling it a regression. If only aggregate counts moved, call it baseline staleness or repo-growth drift unless pass bugs appeared.

## Route findings

Use `docs/decompiler-correctness-pipeline.md` vocabulary:

- Stage 0: build/test failures. - Stage 2: Full malformed or semantic validity defects. - Stage 8: compile-back fidelity opcode diffs. - Stage 9: corpus-card movement or baseline staleness. - Stage 10: changed-method fidelity/skeleton failures.

Before creating new work, check #1584 for triage comments and #1568 for active defect dockets. The current pattern is to cluster repeated rows into focused dockets like #1687 (invalid Full printer) or #1688 (changed-method fidelity skeleton), not to file one issue per assembly.

## Report shape

Keep the report short:

```md Deep Inspect analysis: <green | failed | moved>

- Latest run: <id>, <sha>, <lane>, <conclusion>, <duration/link>. - Failure class: <none | test | census | environment>, with root cause if known. - Corpus health: fully raised, conditional residual, forward-merge, Full malformed, semantic defects, pass bugs, fidelity exact/opcode-diff/recompile/context counts. - Pinned-subset signal: <unchanged | improved | regressed>; separate aggregate drift. - Action: <none | watch | rebaseline | update #1584/#1568 | file focused issue>. ```

Do not hand-construct PR quality tables. For PR evidence, use the DecompilerHarness-generated `--quality-diff-card`.

Intended uses

  • Use Deep Inspect Analysis 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.