fix(core): escapeDkgRdfLiteral covers only ECHAR; non-ECHAR ASCII controls leak through

[Jurij89]

Summary

The canonical RDF literal escaper at packages/core/src/publisher-extension.ts:252 only covers the seven ECHAR characters: backslash, double-quote, newline (LF), carriage-return, tab, form-feed, backspace. Other ASCII control bytes — NUL (0x00), VT (0x0B), DEL (0x7F), plus 0x01-0x07 and 0x0E-0x1F — are left raw in the output, which produces invalid N-Triples literals at the storage layer.

This was discovered during Codex review on PR #413 (the OpenClaw dkg_share addition). The OpenClaw handler now wraps the call with a defensive UCHAR-encoding post-pass, but every other consumer of escapeDkgRdfLiteral and normalizeDkgPublisherObject has the same gap. Fixing it in core eliminates the per-call-site workaround.

Proposed fix

Extend the helper to UCHAR-encode every ASCII control byte that isn't already covered by an ECHAR shortcut. Sketch:

escapeDkgRdfLiteral(value):
  // Step 1: escape backslash and double-quote first (ECHARs introducing the escape char).
  // Step 2: replace each remaining control byte 0x00-0x1F or 0x7F:
  //   - 0x08 -> \b
  //   - 0x09 -> \t
  //   - 0x0A -> \n
  //   - 0x0C -> \f
  //   - 0x0D -> \r
  //   - any other -> \uXXXX (uppercase hex, 4 digits)

Order matters: the backslash/quote replacements run first so the UCHAR escape sequences (which start with backslash) don't get themselves escaped a second time.

Acceptance criteria

• Input containing a NUL byte (0x00) produces output with the 6-char sequence: backslash, u, 0, 0, 0, 0.
• Input containing VT (0x0B) produces backslash, u, 0, 0, 0, B.
• Input containing DEL (0x7F) produces backslash, u, 0, 0, 7, F.
• Existing ECHAR coverage unchanged (\b, \t, \n, \f, \r, \\, \" still produce the canonical short forms).
• Tests added in packages/core/test/publisher-extension.test.ts (or wherever the helper is exercised) covering at least the three non-ECHAR cases above plus a mixed-content round-trip.
• No existing call site that relies on the helper changes behavior — only previously-broken inputs (raw control bytes) now produce well-formed output.

Out of scope

• Removing the per-call-site UCHAR workaround in packages/adapter-openclaw/src/DkgNodePlugin.ts handleShare once this lands. Should be done as a small follow-up after this fix merges so the workaround can lean on the canonical helper without redundant work.

Related

• feat(adapter-openclaw): add dkg_share tool for direct SWM writes (#382, #408) #413 — OpenClaw dkg_share (current consumer carrying the per-call-site workaround this issue would let us remove).
• bug(adapter-hermes): dkg_share has constant-subject upsert and unquoted-literal bugs #414 — Hermes dkg_share parallel bugs (Hermes carries a Python _quote_literal with the same gap).

[Bojan131]

Automated liveness test added in PR #1129 (docs/testing/ISSUE_LIVENESS_TESTS.md). The test asserts the correct behaviour and currently fails-as-expected (bug confirmed live on main as of 2026-06-11). When this is fixed the test flips RED — remove its it.fails wrapper and close this issue.