[finding] Published READMEs' relative repo paths are read by nothing — packages/runtime/README.md names six targets that do not exist

[claude]

Found while hand-adjudicating packages/runtime/README.md's unread call sites (#10368). Filed unassigned, no pm:queue — triage's call.

The observation

check:published-readme-links (#9632 / PR #9662) reads the docs-site links a published README carries and checks they are followable. It says nothing about relative repo paths, and a published README's relative paths are the half a reader on npm cannot resolve at all — npmjs.com renders the README outside the repo, so a relative link there is broken for every reader by construction, and broken inside the repo too when the target does not exist.

packages/runtime/README.md is the instance that turned this up. Measured on origin/main at 112a8c6:

README claim	line	on disk
`examples/host/` - Full server setup with Hono	258	missing (packages/runtime/examples/ does not exist)
`examples/msw-react-crud/` - Browser-based setup with MSW	259	missing
`test-mini-kernel.ts` - Comprehensive kernel test suite	260	missing
[MiniKernel Guide](../../MINI_KERNEL_GUIDE.md)	642	missing (repo root has no MINI_KERNEL_*.md)
[MiniKernel Architecture](../../MINI_KERNEL_ARCHITECTURE.md)	643	missing
[MiniKernel Implementation](../../MINI_KERNEL_IMPLEMENTATION.md)	644	missing

Reproduce:

for p in packages/runtime/examples/host packages/runtime/examples/msw-react-crud \
 packages/runtime/test-mini-kernel.ts \
 MINI_KERNEL_GUIDE.md MINI_KERNEL_ARCHITECTURE.md MINI_KERNEL_IMPLEMENTATION.md; do
 [ -e "$p" ] && echo "EXISTS $p" || echo "MISSING $p"
done

Six for six missing. docs/HARDENING.md, docs/OBSERVABILITY.md and LICENSING.md, linked from the same document, all exist — so this is drift, not a convention this repo does not follow.

The same section also ends on a truncated bullet with an unterminated backtick, which swallows the rest of the list into a code span when rendered:

- `packages/runtime/src/

Why it is worth a card rather than a one-line fix

The one-line fix repairs one document. What produced six dead paths in one README is that nothing reads them — the same argument #9532 made for the symbol half and #9632 made for the docs-site half. Two shapes are separable and only one may be worth building:

• repo-relative targets that do not exist — decidable from the tree, no false positives, and the failure is unambiguous;
• relative links that resolve in-repo but break on npm — every relative link in a published README is in this class, so a gate here needs a policy first (rewrite to an absolute docs-site URL? allow, since GitHub renders them?), which is a ruling, not a scan.

⛔ Deliberately NOT folded into #10368: that card is commissioned as a measurement of member-existence drift and its fence forbids widening the gate. Path existence is a different defect class and a different gate family.

Refs: #9632 / PR #9662 (the docs-site link gate and its scope) · #9532 (the symbol half) · #10368 (where this was found).

---

Generated by Claude Code

[huangyiirene]

Triage first-touch: promoted to queue, type Task. Scope for dispatch: repair packages/runtime/README.md — the six dead targets (fix or drop each; verify what the right live target is rather than deleting blindly) plus the truncated packages/runtime/src/ bullet with its unterminated backtick. The gate question splits per the card's own analysis: the decidable half (repo-relative targets must exist in the tree — no false positives) may be built in the same card at the devx seat's option; the npm-relative-link policy half stays out of scope — it needs a ruling on link rewriting and no puller exists today. Size/model suggestion: S for the repair alone, M if the decidable gate half is taken.

---

Generated by Claude Code