docs: Record security audit verdict

[BenWestgate]

What

Record the finished 2026-09-22 python-codex32 security-audit verdict in docs/audits/ while keeping local planning and unfinished review material out of the repository.

The audit records confirmed findings, the falsified candidate, protocol/application hazards, and the release recovery boundary. It explicitly separates:

• accident-safety restore controls (wallet: Require the recorded fingerprint before import #57/wallet: Check existing seed before sharing #81): typed wallet-record fingerprint when available, otherwise an explicit no-record visual authorization before mutation; and
• malicious-tampering resistance (wallet: Encrypted descriptor backup keyed by the seed #55): an authenticated encrypted descriptor backup verified against the recovered seed.

The no-record path is documented as a last-resort waiver/visual authorization, not independent authentication.

Review stack

Base: #33 (fix-hrp-83-limit). Current head: d07024a.

This refresh replayed the four previously reviewed human-authored commits onto the settled runtime/foundation stack. All four stable patch IDs are unchanged. The GitHub diff remains one audit file and all prior review threads are resolved.

Validation

On the stacked candidate through #52:

• 950 tests passed normally;
• 950 tests passed under python -O;
• Ruff check and format check passed;
• strict mypy passed for 21 source files;
• workflow YAML parsed successfully;
• git diff --check passed.

Refs #20.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 00470b1f2b

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".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3bc520a5d3

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".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 30e54cab79

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".

[baslabofhd777-lab]

yes

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c5ec185f0a

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".

[BenWestgate]

Put planning back in .gitignore. this is a nightmare to review documents the world is not meant to see. Create a separate folder for audit results instead which are at least finished documents not subject to endless nit picking..

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dc0d70a46d

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".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f7b86cd9b7

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".

[BenWestgate]

AI-generated review (Claude), posted at the maintainer's request.

ACK f7b86cd, with one cross-PR conflict.

• The accident-safety vs tampering split (#28/#57 vs #55) reads accurately, and the 32-bit grind point is stated without overclaiming.
• #21 and #22 closed today. A one-line "Status: fixed in " under each would stop readers treating them as open. Same for #32 once #33 lands.
• Nit: no trailing newline at EOF.

[BenWestgate]

Follow-up on the ACK nits: I’m not baking transient PR state into this historical verdict yet. The report intentionally records what was true at audited parent c118a83; #21/#22 are already fixed on reviewability-v1, #32 remains pending #33, and the restore finding remains pending #57/#28 integration. When those release-gate fixes land, add a compact “Status after audit” section (rather than rewriting each finding as if it were true at c118a83) and fix the EOF newline in that same doc-only touch. No security conclusion is blocked on those editorial nits.

[BenWestgate]

Release-gate follow-up: the cross-PR conflict noted in the f7b86cd ACK is resolved by current #57 head dcc0d41, which restores the no-record codex32/Bails/Bails-alpha identifier evidence and has been independently re-verified for verify-before-mutate. The published-audit sequencing condition is also satisfied because the cache-bound and secret-rendering fixes are already on reviewability-v1. The remaining review notes are editorial/historical-status nits; #32 should only be marked fixed in the audit after #33 actually lands.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 64bc3e4be3

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".

[BenWestgate]

AI-generated release-gate review, posted at the maintainer's request.

ACK 9c5fd5a.

The current one-file audit accurately records the historical findings while separating the release-gate accident-safety control (#28/#57) from the stronger malicious-tampering design work in #55. The no-record fallback is correctly described as an explicit operator waiver/authorization, not independent authentication, and the required recovery invariant matches that distinction. All prior inline findings are resolved and exact-head Python-package run 417 is green.

Nit only: the Markdown file still lacks a final newline. That is non-blocking and can be folded into the final human integration/handoff rather than creating review churn.

[BenWestgate]

Agent release-gate review at exact head 9c5fd5a: the audit record now accurately separates validated implementation findings from protocol/application hazards, explicitly marks the opaque-HRP candidate as falsified, and distinguishes the release-gate fingerprint/no-record accident-safety control from #55's stronger malicious-tampering design. Its required restore invariant matches the current #57/#28 direction. Exact-head Python-package run 36532244956 succeeded. No documentation blocker found; the missing final newline is non-functional and not worth another churn commit. Review after the underlying #33/#57 behavior is settled so the recorded status remains true.

[BenWestgate]

Codex current-head audit-verdict re-review: ACK d07024a for content.

The one-file verdict accurately separates the four confirmed implementation findings (#21, #22, #32, #26/#30), the two application-level BIP93 hazards, the falsified opaque-HRP candidate, and the stronger malicious-substitution work deferred to #55. Its recovery invariant matches the live restore stack: independent record evidence first, explicit no-record waiver only when independent evidence is absent. All prior inline findings are resolved.

This draft head has no GitHub workflow run attached, so this review is a content/traceability ACK rather than a fresh CI claim. The missing final newline is still a non-blocking editorial nit. No material audit-tracking blocker found.