v7.28.3: IntegrityCheck opens red on a stock install: 239-ref payload debt (#1541 regrown), unshipped replay corpus, BOM/CRLF-blind frontmatter probes

[umair-a11y]

Summary

On a fresh v7.28.3 Windows install, LIFEOS/TOOLS/IntegrityCheck.ts reports 281 blocking findings. After separating out local/derived-config classes, three vendor-side classes remain that a stock install can never clear. This is the same shipped-payload debt class as the closed #1541 (223 missing refs at the time) and #1504; the count has grown to 244.

Environment

LifeOS	v7.28.3 (release tarball)
OS	Windows 11 Home 10.0.26200
Harness	Claude Code v2.1.220
bun	1.3.14

Finding 1: 244 missing references in the shipped payload

references check: 2648 refs across 1619 files, 244 missing. Top clusters:

Referencing file	Missing refs
LIFEOS/ALGORITHM/changelog.md	42
LIFEOS/DOCUMENTATION/LifeosSystemArchitecture.md	16
LIFEOS/DOCUMENTATION/Work/WorkSystem.md	14
LIFEOS/DOCUMENTATION/Testing/TestingDoctrine.md	8
LIFEOS/DOCUMENTATION/Observability/ObservabilitySystem.md	8
LIFEOS/DOCUMENTATION/Memory/MemorySystem.md	8
skills/Remotion/Workflows/GeneratedContentVideo.md	8
LIFEOS/DOCUMENTATION/Pulse/PulseSystem.md	7
LIFEOS/DOCUMENTATION/Ledger/LedgerSystem.md	7
LIFEOS/DOCUMENTATION/Tools/Cli.md	6

Docs also reference hooks that do not ship, e.g. hooks/ULWorkSync.hook.ts (6 refs). Additionally, the checker scans runtime artifacts under LIFEOS/MEMORY/ (e.g. OBSERVABILITY/reviewer-runs/<ts>/prompt.user.md picked up 7 "missing refs" from generated content); excluding runtime dirs from the reference scan would keep the signal clean.

Finding 2: replay-corpus check demands a corpus the tarball does not ship

The replay-corpus check runs bun test <configRoot>/test/regression. The public release contains no test/regression, so the check fails on every stock install with Test filter ... had no matches. Suggested: skip-with-note when the corpus directory is absent (the retirement-registry and gate_counts checks already model this: "skipped — release tooling not installed").

Finding 3: frontmatter probes are CRLF- and BOM-intolerant (Windows false positives)

The frontmatter checks match /^---\n([\s\S]*?)\n---/ against raw file bytes (IntegrityCheck.ts:658 for commands/agents, :742 for skills). Two byte-level realities of Windows-authored files defeat the anchor:

• CRLF line endings: the file starts ---\r\n, the regex fails (reproduced with a Windows-authored skill; converting to LF cleared the finding with zero content change).
• UTF-8 BOM: EF BB BF before the opening --- breaks the ^ anchor even with LF endings (reproduced with two command files; stripping the BOM cleared them).

Both report as no frontmatter block on files whose frontmatter is complete. Suggested fix: strip a leading BOM and use /^---\r?\n([\s\S]*?)\r?\n---/ (or normalize before matching). Same Windows-probe family as the detection notes in #1730.

Additional notes

• hook_registration counts only directly-registered hooks, while hook_wiring credits dispatcher-invoked hooks; dispatcher-called hooks therefore inflate the registered-vs-on-disk delta on installs that are actually correctly wired.
• CarrierProbe.ts on Windows always returns INCONCLUSIVE with sidechain: [] (it does not find the subagent transcript in the session dir), so the carrier fact can never be established on this platform.

Impact

A fresh install's integrity verdict opens at ~250+ blocking findings the user cannot fix, which buries the findings they can. Once local classes were repaired here, the vendor classes above are the entire remaining verdict.

Workarounds used locally

None available for Findings 1-2 (vendor payload). Finding 3: converted affected files to LF.

[danielmiessler]

Thanks @umair-a11y — part of this landed: the BOM/CRLF frontmatter tolerance and a settings-parity gate are fixed in source. The 239-reference payload-debt half needs a payload-aware reference gate, which isn't built yet, so I'm keeping this open to track it.