Known Issues

Known Issues & Intentional Design Notes

These notes were captured during the Goal 1–3 consolidation so that future cleanup doesn't re-introduce removed bloat or mistake intentional structure for redundancy.

Source: docs/KNOWN_ISSUES.md

---

Intentional duplication — do NOT remove as "bloat"

Some per-skill vendoring is deliberate self-containment, verified during the refactor. Git stores identical blobs once, so the on-disk repetition is not a storage cost:

What	Scope	Why it's intentional
canvas-fonts	vendored per skill (~4 skills × 54 files)	Keeps each canvas/design skill runnable standalone, without a shared font dependency.
OOXML / XSD schemas	vendored per skill (~5 skills × 39 files)	Keeps document-generation skills self-contained.

Warning

Don't "deduplicate" these. They are preserved on purpose; removing them breaks the skills' standalone runnability.

---

Resolved bugs

• infrastructure/mcp-bridges/bridge_context_mode.ps1 previously launched Node via the full path C:\Program Files\nodejs\node.exe, which could space-split the -Command argument under mcp-proxy. Fixed (2026-06-17): the invocation now uses the 8.3 short path C:\PROGRA~1\nodejs\node.exe.

---

Dependency vulnerabilities (Dependabot)

skills/onefilellm/references/ pins onefilellm's upstream dependency set.

Resolved 2026-06-17:

Package	Bump	Advisory
yt-dlp	2026.3.17 → 2026.6.9	GHSA-f7j3-774f-rfhj (cookie leak)
bleach	6.3.0 → 6.4.0	GHSA-8rfp-98v4-mmr6 (URI sanitization) + GHSA-g75f-g53v-794x (linkify CPU exhaustion, ==6.3.0 only)

Accepted risk (no upstream fix available):

• nltk==3.9.4 — GHSA-p4gq-832x-fm9v (URL-encoded path traversal in nltk.data.load()), severity High. 3.9.4 is the latest release; no patched version exists. onefilellm uses NLTK only for stopword removal via the hardcoded stopwords corpus name — it never passes untrusted input to nltk.data.load(), so the vulnerable path is not reachable in this usage. The Dependabot alert was dismissed as tolerable risk; it will be bumped as soon as upstream ships a fix.

---

Related pages

• Security — reporting and secret handling
• Testing & CI — the gates that protect against regressions
• Skills Corpus — why some vendoring is intentional