Skip to content

TMP.md

Latest commit

 

History

History
596 lines (428 loc) · 28 KB

TMP.md

File metadata and controls

596 lines (428 loc) · 28 KB

TMP.md — Codebase Audit & Improvement Plan

Created: 2026-03-28. Last updated: 2026-04-07. Row #27 wrong-fix-reverted 2026-04-07.

Implementation Progress

  • Phase 0 (CI infra) — DONE — 6b5593c (#28, #1, #2, #3, #4)
  • Phase 1 (WASM safety + XSS) — DONE — 39d6998 (#6, #6b, #7, #29)
  • Phase 2 (build hardening) — DONE — 4b4ca12 (#30, #10, #11), 87cb02c (#13, #36)
  • Phase 3 (a11y) — DONE — 970402a (#31, #32), ba11ecf (#33)
  • Phase 4 (assets) — DONE — 3133ed5 (#34, #15, #16 partial)
  • Phase 5 (CSP) — DONE — 6d3de24 (#17)
  • Phase 6 (tests) — DONE — 3ff7a0a (#18, #20)
  • Phase 7 (landing split) — DONE (#21)
  • Follow-up #16 — DONE — dda7c67 (favicon + apple-touch-icon link tags)
  • Merges into development: d0b491c, eb0c994, 6d94c95

Auditors: Claude Opus 4.6 (initial audit + plan), Codex gpt-5.4 xhigh (independent review x2) Status: Pre-implementation — all findings reconciled across three independent passes.

Issue Tracker

Original Issues (#1-#27)

# Category Severity Issue Status Source Notes
1 CI High Deploy workflow missing cargo test --test content_check Done Claude Covered by ci.yml. Commit 6b5593c
2 CI High No cargo clippy in CI — linting regressions can ship Done Claude ci.yml runs clippy SSR + WASM. Commit 6b5593c
3 CI Medium No cargo fmt --check in CI — formatting not enforced Done Claude Commit 6b5593c
4 CI Medium No cargo audit in CI — security advisories not caught (PRD SR-003) Done Claude Added to deploy.yml. Commit 6b5593c
5 CI Low No Lighthouse CI (PRD TR-003) Deferred Claude Disproportionate at current scale
6 WASM Medium theme_toggle.rs — double-unwrap on local_storage() panics if localStorage blocked Done Claude Graceful None handling + try/catch on inline bootstrap. Commit 39d6998
7 WASM Medium doc_search.rs:119 — resp.text().unwrap() panics on Response body failure Done Claude resp.ok() guard + chained validation. Commit 39d6998
8 WASM Low web_sys::window().unwrap() appears ~11 times across island code Open Claude Multiple island files
9 WASM Low 5 islands use Closure::forget() / std::mem::forget() — intentional, never cleaned up Open Claude DocSearch, CodeBlockActivator, MobileMenu, TocHighlighter, ThemeToggle
10 Build Medium build.rs add_heading_ids() uses string-based HTML parsing — fragile if pulldown-cmark output changes Done Claude Rewritten: IDs injected in cmark event stream; add_heading_ids removed. Commit 4b4ca12
11 Build Medium build.rs generate_rust_module() minimal escaping (only " to \") — title, description, and headings all affected Done Claude, Codex Raw strings r#"..."# used for title, description, heading text. Commit 4b4ca12
12 Build Low build.rs ~30 unwrap()/expect() calls — acceptable for build script Skipped Claude Build script panics are diagnostic; skipped per plan
13 Build Low prerender.rs doesn't check HTTP status codes — 500 response written as-is Done Claude http_get() parses status code, Err on non-200. Commit 87cb02c
14 Assets Medium No OG images (PRD PR-005 mentions WebP) Deferred Claude Requires design assets
15 Assets Low No CNAME file for custom domain (PRD DR-001) Done Claude public/CNAME → commitbee.buzz. Commit 3133ed5
16 Assets Low No apple-touch-icon or favicon.ico Partial Claude Placeholder 1×1 PNGs added to public/images/. <link> tags in src/app.rs deferred (off-limits worktree). Commit 3133ed5
17 Security Medium CSP allows 'unsafe-inline' and 'unsafe-eval' — weakens policy significantly Done Claude unsafe-eval dropped; unsafe-inline retained for scripts (Leptos hydration) and styles (Tailwind v4) with documented justification. Commit 6d3de24
18 Testing Medium Only 3 content-validation tests — no Rust unit tests, component tests, or WASM tests Done Claude 3→7 tests: order uniqueness, description length, delimiter-on-own-line, internal links. Commit 3ff7a0a
19 Testing Low No PRD TR-001 through TR-005 test requirements implemented Partial Claude TR-001 covered by content tests; TR-002..005 need browser infra
20 Testing Low No internal link validation between doc pages Done Claude internal_links_resolve() test added. Commit 3ff7a0a
21 Code Low landing.rs 415 lines in single file — could split into section modules Done Claude Split into src/pages/landing/{hero,problem,pipeline,differentiators,comparison,install,docs_transition}.rs
22 Code Low Scroll-reveal IntersectionObserver in inline script runs indefinitely (never disconnected) Open Claude src/app.rs inline script
23 Code Low doc_search.rs — serde_json imported but could be more explicit about dependency chain Open Claude src/components/doc_search.rs
24 Docs Done All 12 doc pages updated from stubs to comprehensive content Done Claude content/docs/*.md
25 Docs Done Test count corrected from 308 to 424 across all pages Done Claude content/docs/architecture.md
26 Docs Done LLM provider TOML syntax fixed (was [provider] sections, now flat keys) Done Claude content/docs/llm-providers.md
27 Docs Wrong-fix-undone Secret pattern count "corrected" (24 → 25) was wrong — true count is 24 Reverted Claude Audit was incorrect; true count is 24 (verified against src/services/safety.rs::builtin_patterns() and DOCS.md). Also fixed architecture.md crate-tree comment (same wrong value). Reverted in 416d055

New Issues from Codex Cross-Review (#28-#36)

These issues were discovered by Codex gpt-5.4 during independent codebase audit and plan review, and confirmed by Claude during reconciliation.

# Category Severity Issue Status Source Notes
28 CI Critical No PR validation workflow — deploy.yml only runs on push to development, no ci.yml on pull_request Open Codex .github/workflows/ — highest-leverage CI gap
29 Security High Raw docs HTML injected via inner_html + CSP allows unsafe-inline = XSS vector if docs contain malicious content Open Codex build.rs:175,233 → src/pages/docs.rs:48 → src/app.rs:58
30 Build Medium Frontmatter parsing bug — after_first.find("---") matches --- inside YAML content, not just delimiter lines Done Codex Fixed in both build.rs and tests/content_check.rs. Commit 4b4ca12
31 A11y Medium Search modal lacks role="dialog", aria-modal, focus trap — background content remains tabbable while open Open Codex src/components/doc_search.rs:189,195,203
32 A11y Medium Mobile menu overlay lacks dialog/focus management — no scroll lock, keyboard focus can escape into page content Open Codex src/components/mobile_menu.rs:39,45,63
33 A11y Medium Copy buttons invisible to keyboard users — opacity-0 hover-only, tabbable but not visible Open Codex build.rs:213, src/pages/landing.rs:331
34 Build Medium Search index only copied to target/site if directory already exists — first build can miss it; client fails silently Done Codex create_dir_all before write. Commit 3133ed5
35 Build Low Tailwind @source "../src/**/*.rs" doesn't scan build.rs output — classes emitted by build script may be missing from CSS Open Codex style/tailwind.css:11, build.rs:213
36 Code Low Route discovery for prerendering uses recursive search for first routes.txt under target/ — can pick stale artifact Open Codex src/bin/prerender.rs:60,205

Reconciliation: Claude vs Codex Findings

Items Both Auditors Agree On

  • All 4 CI gaps (#1-#4) are real and should be fixed
  • Closure::forget() (#9) is intentional — document only
  • build.rs unwrap() usage (#12) is appropriate — skip
  • OG images (#14) require design assets — defer
  • CNAME (#15) and apple-touch-icon (#16) are correct and straightforward
  • landing.rs refactor (#21) is low priority
  • Scroll-reveal observer (#22) is effectively idle — document only
  • serde_json clarity (#23) — not a real issue

Codex Corrections Accepted

Original Correction Rationale
#1 severity High Should be part of a larger CI workflow fix (#28) The bigger gap is no PR validation at all, not just missing test step
#6 fix scope Must also guard inline theme bootstrap script in app.rs:64-71 The inline localStorage.getItem('theme') has the same unwrap assumption as the island
#7 fix scope Should also check resp.ok() / status before calling text() A 404/500 on search_index.json would silently fail
#11 fix scope Must fix title and description fields too, not just headings All three are emitted as normal string literals with only " escaping
#13 fix Parse numeric status code instead of contains("200") More precise validation for the prerender HTTP check
#8 (tracked) Remove from active tracking Replacing unwrap() with expect() in a helper doesn't materially improve safety — window is always available in browser WASM context
#17 approach Should experiment with tightening CSP before documenting as-is Particularly unsafe-eval — may not be required by Leptos 0.8
#23 (tracked) Remove from active tracking Codex and Claude agree this is not a real issue

Codex Corrections Rejected / Modified

Codex Suggestion Decision Rationale
Downgrade #1, #2 from High Partially accepted — reframe as part of #28 CI gates matter but the real fix is a separate PR workflow
#5 should not be deferred (PRD TR-003) Rejected — still defer PRD requirement exists but Lighthouse CI is disproportionate for a static site at current scale
#10 severity should be lower Accepted — keep Medium but acknowledge the rewrite is more invasive than originally described
#17 should not be "document only" Accepted — change to "experiment first, then document"
#18 fix doesn't address the issue statement Partially accepted — content tests are still valuable; component/WASM tests require browser infrastructure
#19 should be higher than Low Rejected — the PRD requirements are aspirational; current test coverage is appropriate for project stage

What's Already Done

  • All 12 documentation pages updated from sparse stubs to comprehensive content matching source DOCS.md
  • Architecture page: test count 308 → 424, added full crate structure, design decisions, error philosophy
  • Commands page: added short flags, config set-key/get-key, usage patterns
  • Configuration page: full TOML reference, config priority, environment variables
  • LLM Providers page: fixed TOML syntax, added recommended models, thinking mode, secure key storage
  • All other pages expanded with complete content from source

Implementation Plan

Summary: Prioritized Execution Order

Phase Issues Effort Impact Priority
0 #28, #1, #2, #3, #4 3 hours Prevents broken deploys on PRs Critical
1 #6, #6b, #7, #29 4-6 hours Prevents runtime panics + XSS hardening High
2 #30, #13, #10, #11 6-10 hours Build correctness + parser bugs Medium
3 #31, #32, #33 3-4 hours Accessibility compliance Medium
4 #15, #16, #34, #36 2 hours Asset completeness + build robustness Low
5 #17 2 hours Security — CSP experiment + document Low
6 #20, #18 2-4 hours Testing improvements Low
7 #21 2-3 hours Code organization Lowest

Explicitly Deferred or Skipped

Issue Action Reason
#5 Defer Lighthouse CI is disproportionate for static site at current scale; manual testing sufficient
#8 Remove Replacing unwrap() with expect() helper doesn't materially improve safety
#9 Document only Closure::forget() is correct for long-lived listeners in Leptos islands
#12 Skip Build script unwrap() is appropriate; panics are diagnostic
#14 Defer Requires design assets (1200x630 OG images), not code
#19 Partial (TR-001 only) TR-002 through TR-005 need browser testing infrastructure
#22 Document only IntersectionObserver is effectively idle after reveal
#23 Remove Both auditors agree this is not a real issue
#35 Document Note in CLAUDE.md that build.rs classes need explicit CSS rules in tailwind.css (current workaround already in place)

Phase 0: CI Infrastructure (Issues #28, #1-4) — 3 hours

The highest-leverage improvement. Create a proper CI workflow that runs on all PRs, then add validation gates.

Issue #28: Create ci.yml PR validation workflow [S]

File: .github/workflows/ci.yml (new file)

name: CI

on:
  push:
    branches: [development]
  pull_request:
    branches: [development]

jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Install Rust toolchain
        uses: dtolnay/rust-toolchain@nightly
        with:
          targets: wasm32-unknown-unknown

      - name: Cache cargo registry and build
        uses: actions/cache@v4
        with:
          path: |
            ~/.cargo/registry
            ~/.cargo/git
            target
          key: ${{ runner.os }}-cargo-${{ hashFiles('**/Cargo.lock') }}
          restore-keys: |
            ${{ runner.os }}-cargo-

      - name: Check formatting
        run: cargo fmt --check

      - name: Validate content
        run: cargo test --test content_check

      - name: Lint (SSR)
        run: cargo clippy --features ssr -- -D warnings

      - name: Lint (WASM)
        run: cargo clippy --features hydrate --target wasm32-unknown-unknown -- -D warnings

This runs on both push and PR. The deploy workflow (deploy.yml) remains separate and only handles build + deploy on push.

Issue #1: Content validation — covered by ci.yml step above.

Issue #2: Clippy — covered by ci.yml steps above (both SSR and WASM targets).

Issue #3: Formatting check — covered by ci.yml step above.

Issue #4: Add cargo audit to deploy workflow [S]

Add to .github/workflows/deploy.yml before the build step:

- name: Install cargo-audit
  run: cargo install cargo-audit
- name: Audit dependencies
  run: cargo audit

Note: Only in deploy workflow since audit failures should block deployment but not necessarily block PR merging (some advisories are informational). Consider cargo binstall cargo-audit to speed up CI.

Phase 1: WASM Runtime Safety + XSS Hardening (Issues #6, #6b, #7, #29) — 4-6 hours

Issue #6: Fix theme_toggle.rs double-unwrap [S]

File: src/components/theme_toggle.rs

Lines 14-15: window.local_storage().unwrap().unwrap(). Line 38: same pattern.

Fix: Replace with graceful fallback:

// Before (line 14-15)
let window = web_sys::window().unwrap();
let storage = window.local_storage().unwrap().unwrap();

// After
let storage = web_sys::window()
    .and_then(|w| w.local_storage().ok())
    .flatten();

When storage is None, fall through to match_media system preference check (existing fallback). In the toggle closure (line 38), use if let Some(storage) — theme toggles visually but is not persisted.

Issue #6b (NEW): Guard inline theme bootstrap script in app.rs [S]

File: src/app.rs, lines 65-71

The inline script does localStorage.getItem('theme') without guarding against localStorage being unavailable (same scenario as #6 — Safari ITP, private browsing).

Fix: Wrap in try-catch:

(function(){
    try {
        var t = localStorage.getItem('theme');
        if (t === 'dark' || (!t && matchMedia('(prefers-color-scheme:dark)').matches)) {
            document.documentElement.classList.add('dark');
        }
    } catch(e) {
        if (matchMedia('(prefers-color-scheme:dark)').matches) {
            document.documentElement.classList.add('dark');
        }
    }
})();

Issue #7: Fix doc_search.rs resp.text().unwrap() panic + add status check [S]

File: src/components/doc_search.rs, line 117-119

Fix: Check response status before reading body:

// Before
if let Ok(resp) = resp {
    if let Ok(json) = wasm_bindgen_futures::JsFuture::from(resp.text().unwrap()).await

// After
if let Ok(resp) = resp {
    if resp.ok()
        && let Ok(text_promise) = resp.text()
        && let Ok(json) = wasm_bindgen_futures::JsFuture::from(text_promise).await
        && let Some(text) = json.as_string()
        && let Ok(entries) = serde_json::from_str::<Vec<SearchEntry>>(&text)
    {
        set_index.set(Some(entries));
    }
}

This adds: (1) status check via resp.ok(), (2) guarded text() call, (3) chained validation.

Issue #29: XSS hardening — docs HTML + CSP [M]

File: src/app.rs:58, build.rs:175,233, src/pages/docs.rs:48

The raw HTML pipeline: build.rs renders markdown to HTML → generate_rust_module() embeds as r##"..."## → docs.rs injects via inner_html=page.html_content → CSP allows unsafe-inline.

Since docs content is authored at build time (not user input), the practical risk is low. However, combined with the permissive CSP, any malicious content in a markdown file would execute.

Mitigation strategy (in order of impact):

  1. Add html-sanitize or ammonia in build.rs to strip <script>, onclick, onerror etc. from rendered HTML before embedding
  2. Alternatively, tighten CSP to remove unsafe-inline for scripts (requires hash-based approach for the 3 known inline scripts + Leptos hydration)
  3. Minimum: add a build-time test that verifies no <script> tags exist in rendered doc HTML

Phase 2: Build Pipeline Hardening (Issues #30, #13, #10, #11) — 6-10 hours

Issue #30: Fix frontmatter parsing — delimiter matching [S]

File: build.rs:124-126, tests/content_check.rs:25-27

Current code: after_first.find("---") matches --- anywhere in the remaining content, including inside YAML values. If a YAML value contains --- (e.g., a description with an em-dash separator), the frontmatter would be truncated.

Fix: Match --- on its own line, not embedded in text:

// Before
let end = after_first.find("---").expect("Missing closing frontmatter delimiter");

// After
let end = after_first
    .find("\n---\n")
    .or_else(|| after_first.find("\n---\r\n"))
    .or_else(|| {
        // Also handle --- at end of string (no trailing newline)
        after_first
            .find("\n---")
            .filter(|&i| i + 4 >= after_first.len())
    })
    .expect("Missing closing frontmatter delimiter");

Same fix needed in tests/content_check.rs:25-27.

Issue #13: Check HTTP status in prerender [S]

File: src/bin/prerender.rs, function http_get()

Parse the numeric status code from the HTTP response line:

let status_line = response.lines().next().unwrap_or("");
let status_code: u16 = status_line
    .split_whitespace()
    .nth(1)
    .and_then(|c| c.parse().ok())
    .unwrap_or(0);
if status_code != 200 {
    return Err(format!("HTTP {status_code} for {path}: {status_line}").into());
}

Issue #10: Replace string-based add_heading_ids() in build.rs [M]

File: build.rs, function add_heading_ids()

Current implementation uses string find() and strip_prefix() to locate <hN> tags — fragile. Rewrite to inject heading IDs during the pulldown-cmark event stream processing inside render_markdown_with_syntax_highlighting().

Strategy: When encountering Event::Start(Tag::Heading { level, .. }), track heading text content. On Event::End(TagEnd::Heading(_)), emit the full <hN id="slug"> tag. This eliminates add_heading_ids() entirely.

Note: This is more invasive than originally described — requires buffering heading content before emitting the opening tag. The extract_headings() function already does this, so the logic can be shared.

Issue #11: Improve escaping in generate_rust_module() for all fields [S]

File: build.rs, function generate_rust_module()

The fix applies to title, description, and heading text — not just headings. All three are emitted as normal string literals with only " to \" escaping.

Fix: Use raw strings for all text fields:

// Before (for title, description, headings)
writeln!(code, "        title: \"{}\",", page.title.replace('"', "\\\"")).unwrap();

// After
writeln!(code, "        title: r#\"{}\"#,", page.title).unwrap();

Raw strings r#"..."# don't need quote escaping. The only edge case is if content contains "# which is essentially impossible in headings/titles/descriptions. Apply to all three fields.

Phase 3: Accessibility (Issues #31, #32, #33) — 3-4 hours

Issue #31: Search modal dialog semantics [M]

File: src/components/doc_search.rs

Add to the search modal container:

  • role="dialog" attribute
  • aria-modal="true" attribute
  • aria-label="Search documentation" attribute
  • Focus trap: on Tab/Shift+Tab when modal is open, wrap focus within the modal
  • On close: restore focus to the search trigger button

Issue #32: Mobile menu dialog semantics [M]

File: src/components/mobile_menu.rs

Add to the menu overlay:

  • role="dialog" attribute
  • aria-modal="true" attribute
  • aria-label="Navigation menu" attribute
  • Scroll lock: set document.body.style.overflow = "hidden" when open, restore on close
  • Focus trap: prevent Tab from escaping into background content

Issue #33: Copy buttons keyboard visibility [S]

File: style/tailwind.css (or inline in build.rs:213 and src/pages/landing.rs:331)

Current: opacity-0 group-hover:opacity-100 — invisible to keyboard users who Tab to the button.

Fix: Add focus:opacity-100 alongside group-hover:opacity-100:

opacity-0 group-hover:opacity-100 focus:opacity-100

This makes the button visible when focused via keyboard navigation.

Phase 4: Build Robustness + Assets (Issues #34, #36, #15, #16) — 2 hours

Issue #34: Ensure search index is always written [S]

File: build.rs:464-466

Current code only writes search_index.json to target/site/ if that directory exists. First build creates target/site/ later, so the search index can be missing.

Fix: Create target/site/ directory before writing:

let site_dir = Path::new("target/site");
fs::create_dir_all(site_dir).unwrap();
fs::write(site_dir.join("search_index.json"), &json).unwrap();

Issue #36: Prerender route discovery robustness [S]

File: src/bin/prerender.rs:60,205

Current code recursively searches for the first routes.txt under target/. This can pick up a stale artifact from a previous build.

Fix: Use the OUT_DIR environment variable (set by cargo for the commitbee-web crate) or the known build output path:

// Instead of recursive search, use known location
let routes_file = find_routes_file();
// Prefer target/build/commitbee-web-*/out/routes.txt (most recent)

Or simpler: sort found paths by modification time and use the newest.

Issue #15: Add CNAME file [S]

Create public/CNAME with the content:

commitbee.buzz

Prerender copies public/ to dist/, so this deploys automatically.

Issue #16: Add apple-touch-icon and favicon.ico [S]

Create:

  • public/images/apple-touch-icon.png — 180x180 PNG
  • public/images/favicon.ico — multi-size ICO (16x16, 32x32, 48x48)

Update src/app.rs <head> section:

<link rel="apple-touch-icon" href="/images/apple-touch-icon.png"/>
<link rel="icon" href="/images/favicon.ico" sizes="any"/>

Keep existing SVG favicon link (Safari/Chrome support it).

Phase 5: CSP Experiment + Documentation (Issue #17) — 2 hours

Issue #17: CSP hardening — experiment first, then document [M]

File: src/app.rs:58

Current CSP:

default-src 'self'; script-src 'self' 'unsafe-inline' 'unsafe-eval'; style-src 'self' 'unsafe-inline'; font-src 'self'; img-src 'self' data:; connect-src 'self'

Step 1: Test without unsafe-eval

Build and test the site with unsafe-eval removed from script-src. Codex flagged that this may not actually be required by Leptos 0.8 — WASM instantiation doesn't necessarily need eval. If the site works (hydration, islands, theme toggle, search, pipeline demo), remove it permanently.

Step 2: If unsafe-eval is required

Document why in a comment in app.rs and CLAUDE.md.

Step 3: Evaluate unsafe-inline reduction

  • The 3 inline scripts in app.rs could use hash-based CSP: compute SHA-256 of each script's content, add to script-src
  • Leptos hydration scripts are dynamically generated — test whether their content is deterministic between builds
  • If deterministic: use hash-based CSP for all inline scripts
  • If not: document unsafe-inline as a Leptos framework limitation

Step 4: unsafe-inline for styles

Tailwind CSS utility classes are injected via <style> tags and style attributes. Removing unsafe-inline from style-src would break styling. Document as a Tailwind limitation.

Phase 6: Testing Improvements (Issues #20, #18) — 2-4 hours

Issue #20: Add internal link validation test [S]

File: tests/content_check.rs

Add a test that:

  1. Reads all markdown files in content/docs/
  2. Extracts all [text](/docs/slug) and [text](slug) style internal links via regex
  3. Verifies each linked slug corresponds to an actual .md file

Issue #18: Expand content validation tests [S]

File: tests/content_check.rs

Add tests:

  1. order_values_are_unique_within_section() — no two docs in same section share an order value
  2. description_length_reasonable() — descriptions are 10-200 characters (catches empty or paste-error descriptions)
  3. frontmatter_delimiter_on_own_line() — verify closing --- is on its own line (complements fix for #30)

Phase 7: Code Organization (Issue #21) — 2-3 hours

Issue #21: Split landing.rs into section modules [M]

File: src/pages/landing.rs (415 lines)

Split into:

  • src/pages/landing/mod.rs — Landing component, re-exports, CodeBlockActivator island render
  • src/pages/landing/hero.rs — HeroSection
  • src/pages/landing/problem.rs — ProblemSection
  • src/pages/landing/pipeline.rs — PipelineSection
  • src/pages/landing/differentiators.rs — DifferentiatorsSection, FeatureCard
  • src/pages/landing/comparison.rs — ComparisonSection, ComparisonFeature
  • src/pages/landing/install.rs — InstallSection
  • src/pages/landing/docs_transition.rs — DocsTransitionSection, DocLink

Each section file is self-contained. Best done when no other changes are pending in landing.rs.

Critical Files for Implementation

File Issues Phase
.github/workflows/ci.yml (new) #28, #1, #2, #3 0
.github/workflows/deploy.yml #4 0
src/components/theme_toggle.rs #6 1
src/app.rs #6b, #17, #29 1, 5
src/components/doc_search.rs #7, #31 1, 3
build.rs #10, #11, #29, #30, #34 1, 2
src/bin/prerender.rs #13, #36 2, 4
tests/content_check.rs #18, #20, #30 2, 6
src/components/mobile_menu.rs #32 3
style/tailwind.css #33 3
public/ #15, #16 4
src/pages/landing.rs #21 7

Audit Sources

Auditor Model Mode Scope
Claude Opus 4.6 claude-opus-4-6 Full codebase read + plan creation All .rs, build.rs, CSS, CI, content, assets
Codex gpt-5.4 (pass 1) gpt-5.4 xhigh Independent review of TMP.md plan Verified each issue against codebase, assessed fixes
Codex gpt-5.4 (pass 2) gpt-5.4 xhigh Full independent codebase audit All .rs, build.rs, CSS, CI, tests — 16 findings