Add Plan.md — Own.NET Audit orchestrator plan

[PhysShell]

Plan for a decoupled audit orchestrator that runs a fleet of mature,
ready-made tools (Roslyn analyzer packs, CodeQL, Infer#, own-check,
runtime profilers) over the legacy .NET Framework 4.7.2 / WPF / DevExpress
project and produces a categorized "health anamnesis" report ranked by
where it hurts most.

Key design choices, grounded in the existing repo culture:

• Orchestrator, not a new analyzer — no regex detectors; categories with
  no reliable tool are marked NO-TOOL/skipped, never faked (honest-skip /
  minimal-FP, mirroring OWN050 and own-check --stats).
• SARIF as the single normalized format; reuse parse_sarif and generalize
  oracle_compare's agreement buckets into a cross-tool confidence score.
• Three layers: static fleet (build-free vs build-required tiers) →
  runtime (FlaUI + PerfView/procdump/ClrMD + SematixTrace, leak harness,
  duplicate-immutable detector, PropertyChanged-storm profiler) →
  AI reviewer strictly downstream and falsifiable.
• Lives in an isolated audit/ subtree with zero coupling to ownlang/, so it
  lifts out as a standalone project; own-check is consumed only via its CLI.
• Feeds confirmed findings back into corpus/ to train the analyzer.

Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01QDPpNT9Uh8RoTrKPvgcoRE

Summary by CodeRabbit

• Documentation
  • Added an RFC in Russian for the “Own.NET Audit” audit-orchestration workflow for a legacy desktop app, including scope boundaries, core principles, reporting formats, and phased delivery milestones.
  • Defined the intended end-to-end approach across static and runtime validation, unified findings/reporting, coverage and scoring semantics, and guidelines for using AI only as a downstream explainer/ranker with falsifiable outputs.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f4edd2116d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: dd364388-fbbb-4464-b1dd-04a338b4a928

📥 Commits

Reviewing files that changed from the base of the PR and between f4edd21 and 68a3272.

📒 Files selected for processing (1)

• Plan.md

🚧 Files skipped from review as they are similar to previous changes (1)

• Plan.md

---

📝 Walkthrough

Walkthrough

Adds Plan.md as an RFC for Own.NET Audit, defining scope, a three-layer architecture, static and runtime evidence collection, AI reviewer constraints, repository structure, and phased delivery criteria.

Changes

Own.NET Audit RFC

Layer / File(s)	Summary
Goals, scope constraints, and foundational principles Plan.md	Introduces the RFC title, defines non-goals and scope boundaries, and states the principles for SARIF-only normalization, honest coverage reporting, determinism, and downstream AI use.
Three-layer architecture and category coverage Plan.md	Describes the static, runtime, and AI reviewer layers and introduces the category-to-tool coverage framing used to separate detectability by layer.
Static analyzer setup and tiering Plan.md	Defines audit-only analyzer activation through a temporary git worktree, injected build props and targets, and the build-free versus build-required tool tiering with partial reporting behavior.
Static fleet execution, normalization, and scoring Plan.md	Specifies per-tool SARIF generation, runner expectations, DevExpress baseline suppression with coverage accounting, taxonomy mapping, cross-tool agreement scoring, report rollups, and the per-finding JSON shape.
Runtime evidence collection and correlation Plan.md	Introduces the runtime harness stack, deterministic leak-harness checks, duplicate-immutable detection, PropertyChanged-storm profiling, and correlation of runtime evidence into the unified findings.
AI reviewer role and audit tree layout Plan.md	Defines the downstream-only AI reviewer responsibilities, the feedback loop into OwnLang and corpus artifacts, the intended audit subtree layout, and the decoupling constraints around orchestration languages and imports.
Phases, risks, and open items Plan.md	Lists the delivery phases, phase-1 completion criteria, explicit non-goals, accepted customer decisions, assumptions, risks, and remaining open items.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐇 A rabbit read the RFC by moonlight glow,
With SARIF streams and health reports in tow.
No verdicts from AI, just clues in neat rows,
Phase one hops first, as the audit road grows.
Hop-hop! The plan is set, precise and bright ✨

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: adding a Plan.md for the Own.NET Audit orchestrator.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/brave-gauss-elvquy

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (1)

Plan.md (1)

122-127: 🩺 Stability & Availability | 🔵 Trivial | ⚡ Quick win

Make worktree lifecycle failure-safe in the documented flow.

If build fails mid-run, git worktree remove may be skipped; document try/finally (or shell trap) cleanup so temporary trees don’t accumulate.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@Plan.md` around lines 122 - 127, The documented worktree flow is not
failure-safe because a failed build can skip cleanup and leave temporary
worktrees behind. Update the instructions around git worktree add and git
worktree remove to explicitly use a try/finally-style cleanup or a shell trap so
the temporary target-audit tree is always removed, and reference the documented
build sequence in Plan.md so readers can apply the fix consistently.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@Plan.md`:
- Line 387: The planned tree still includes ndepend.ps1, which conflicts with
the accepted no-NDepend direction. Update the plan entry that lists
roslyn_pack.ps1, codeql.sh, infersharp.sh, owncheck.py, and ndepend.ps1 by
removing ndepend.ps1 entirely or marking it clearly as deprecated so the plan
aligns with the free-stack-only decision.
- Line 507: The Plan.md content has a dangling fenced code block marker, leaving
the Markdown invalid. Remove the standalone triple-backtick at the end of the
section and ensure the fenced block is properly opened and closed around the
intended content so the Markdown renders and lints cleanly.
- Line 52: Several fenced code blocks in the plan docs are missing language
identifiers, which trips MD040. Update each affected fence in the plan document
to include the appropriate language tag (for example text, bash, or tree) based
on the block content, and make sure the same fix is applied to all reported
occurrences so the markdown stays lint-clean and unambiguous.

---

Nitpick comments:
In `@Plan.md`:
- Around line 122-127: The documented worktree flow is not failure-safe because
a failed build can skip cleanup and leave temporary worktrees behind. Update the
instructions around git worktree add and git worktree remove to explicitly use a
try/finally-style cleanup or a shell trap so the temporary target-audit tree is
always removed, and reference the documented build sequence in Plan.md so
readers can apply the fix consistently.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 19c39715-0703-44c2-b0a0-ee49eaf49335

📥 Commits

Reviewing files that changed from the base of the PR and between a21bcaa and f4edd21.

📒 Files selected for processing (1)

• Plan.md

[PhysShell]

Адресовал все ревью-треды в 68a3272 (все 5 зарезолвил):

• MSBuild props injection (Codex P2) — верный catch: инъекция копировалась как OwnAudit.Directory.Build.props, а MSBuild авто-импортит только Directory.Build.props/.targets, так что аналайзеры молча не подхватывались. Теперь кладём под распознаваемым именем + trap-очистка worktree на EXIT (это же закрывает nitpick про незачищенные деревья при падении билда) + заметка про сохранение существующего Directory.Build.props target'а.
• Таксономия OWN001 (Codex P2) — catch вскрыл более глубокую ошибку, чем сам репорт. OWN001 действительно зонтичный (IDisposable / [resource: subscription token] / [resource: timer]), плоский OWN001→кат.1 недосчитывал бы кат. 2/3 — развёл по тегу [resource: …]. Попутно: мой OWN014 был подписан subscription-leak, хотя по P-004/lifetimes.md это region-escape (vm→App / SystemEvents) — переименовал.
• ndepend.ps1 (CodeRabbit major) — убрал из дерева §7, противоречил принятому «только free-стек, без NDepend».
• Markdown lint (MD040 + висячий ```) — проставил языки во все fenced-блоки, удалил непарный закрывающий маркер в конце файла.

Всё это правки консистентности самого плана, поведение не меняется. Документ остаётся внутренне согласованным с §8/§11.

---

Generated by Claude Code