feat(swarm): wire-validator — rejection rules 1-13 (#86)

[Dewinator]

Summary

• New pure function validateWireRecord(record, kind, opts) in mcp-server/src/services/wire-validator.ts returning { ok: true } | { ok: false, rule, reason }.
• Implements rejection rules 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 12, 13 from SWARM_SPEC §5; skips 10 (caller's DB-read), 14 (local trust), 15 (transport-layer body cap) per the issue.
• 15 new tests in mcp-server/src/__tests__/wire-validator.test.ts: one valid example per kind + 12 negative tests (one per implemented rule).

Research summary

• Read docs/SWARM_SPEC.md §3 (wire types) + §5 (rejection rules) end-to-end. Rules 14 (local trust) and 15 (transport body cap) are explicitly out-of-scope per the issue; rule 10 (duplicate detection) is also caller-side because it needs a DB read against ingested records.
• Reused existing primitives instead of re-implementing:
  • verify(record, sig, pubkey) from services/signature.ts (feat(swarm): signature service — Ed25519 sign/verify over JCS (#77) #82) for rule 5 — strips the on-wire signature field internally before JCS-recomputing.
  • computeNodeId(pubkeyRaw) from services/node-identity.ts (feat(swarm): node identity bootstrap + node_identity_get tool (#76) #81) for rule 6 — already exported, no re-export needed.
• Followed the Phase 3a (wire-types.ts, Swarm Phase 3a: src/services/wire-types.ts — TypeScript types + JCS canonicalize helper #85) interface shapes verbatim so the validator's required-field tables track the spec table directly.
• Validator is async-shaped (Promise<ValidationResult>) per the issue contract even though the body is synchronous; matches the pattern other v1-locked-but-v2-extensible services in this repo took.

What this does NOT do

• No rule 10 (duplicate id+signed_at with conflicting signature — needs DB read; caller territory).
• No rule 14 (local trust weight = 0 — local policy, not on the wire).
• No rule 15 (body size cap — transport layer, not record content).
• No HTTP handler, no DB write path, no peer storage. Phase 3c (Swarm Phase 3c: GET /.well-known/mycelium-node — serve self-signed NodeAdvertisement #87) and Phase 3d (Swarm Phase 3d: migration 071 — peer + signed-record storage (file only, do NOT run) #88) own those.
• No memoization or caching — issue hard constraint to keep this pure for testability.
• Does not modify wire-types.ts or signature.ts; consumes them read-only.

Constitution affirmation

This change touches Pillar 6 — Cyber security: it is the single chokepoint that decides whether a received wire record influences local state, so every later ingest path becomes a one-line validate(...) call. It also reinforces Pillar 1 — Decentralized, networked AI by making the rejection contract identical for every node, and Pillar 3 — Swarm intelligence by enforcing the §3.1 generalization rule (rule 11) on the wire so raw episodes cannot leak as lessons. No pillar is weakened. Pillars 2 (Reproduction), 4 (Microtransactions), and 5 (Experts) are not touched.

Test plan

• cd mcp-server && npm test → 349 tests pass (334 prior + 15 new for the validator).
• Inspect mcp-server/src/tests/wire-validator.test.ts: each negative test starts from a fully-valid signed record and mutates exactly one invariant, then re-signs (so rule 5 doesn't piggy-back on the assertion) — except the rule-5 test itself, which mutates content after signing.
• For rule 6 specifically, the test substitutes a foreign node_id and re-signs with the original key, so rule 5 would also pass; the assertion guarantees rule 6 is reached first in the validator order.

🤖 Generated with Claude Code