AI readiness toolkit

Two deterministic-core wiki bugs caught by a cross-tool review

CodeRabbit reviewed v1.11.0/v1.12.0 assessment artifacts on a downstream PR and flagged 12 inline issues; 10 confirmed existing items in #47, two were genuinely new and live in the wiki output rather than the run-context or stats sidecar. Both manifest as cross-file contradictions - index.md and assess-report.md describing the same files differently, or log.md having same-day headings that break GitHub anchors.

Bug 1: graduated rows in index.md showed Latest CCN | 0 | Latest LOC | 0 (#53, closes #52)

• Graduation just means a file fell off top_hotspots[:10], not that its LOC or CCN dropped to zero. The actual metrics are still in complexity-stats.json's wider top_complex and top_large lists. Zero in the wiki contradicted assess-report.md (which described the same file as "still a 1,096-LOC ccn-172 problem") and misled reviewers into thinking the file was emptied.
• assess_core now merges the three top-N lists per-field before building graduated rows. Per-field because top_complex only carries ccn and top_large only carries loc - only top_hotspots carries both. When a metric genuinely isn't in any list, HotspotEntry.ccn/.loc stay None and the wiki renders - rather than misleading zeros. Real zeros still render as 0 so "unknown" stays distinguishable from "actually empty."

Bug 2: log.md duplicate ## YYYY-MM-DD headings on same-day re-runs

• GitHub auto-anchors collided (#2026-05-29, #2026-05-29-1, #2026-05-29-2), markdownlint MD024 fired in repos that lint markdown, and skim-reading the log was harder.
• LogEntry now carries plugin_version. Headings are always ## YYYY-MM-DD (vX.Y.Z) when the version is known, so the log doubles as a version history and two same-day runs across versions distinguish themselves naturally. If the same date+version heading already exists in the file, a HH:MM stamp is appended inside the parens.

Tracked as follow-up (not in this release): adding a "consistency lint" pass at the end of assess_core that catches cross-file contradictions structurally - "does index.md agree with complexity-stats.json on every file's current metrics?" / "does any date in log.md appear more than once?". The two named bugs are the immediate fix; the lint is structural prevention.

Both the Claude Code plugin and the standalone Desktop/Cowork ZIPs are updated (ZIPs on the standalone-skills-latest rolling release).

What's Changed

Fixes

• fix(assess): wiki output bugs surfaced by external review by @bjcoombs in https://github.com/bjcoombs/ai-native-toolkit/pull/53

Full Changelog: https://github.com/bjcoombs/ai-native-toolkit/compare/v1.13.0...v1.13.1