protect: validate delivered rule bundles; bound patterns; harden the telemetry origin

[patchstackdave]

Defense-in-depth follow-ups from an external review. Delivered rules are policy fetched over the network and executed on every request, so an upstream problem (schema drift, a corpus mistake, a compromised or unsafe endpoint) could hand the engine unbounded work — or leave rules that look loaded but protect nothing.

Rule-bundle validation (new rules/validate.js)

Wired into normalizeBundle, the single chokepoint every rule path already funnels through (live fetch, last-known-good cache, bundled fallback). It bounds what the engine will walk and requires a sane shape:

Bound	Limit
rules / whitelists per bundle	5000 / 2000
conditions per rule	250
condition nesting depth	12
regex pattern length	1000
match-value length	8192

plus a known phase (request/response/egress), a known action, a non-empty rule_v2, and a positive max_bytes.

Rejection is explicit, not silent: a failing rule is dropped with a reason, reported via onRuleRejected (or a single warning line). An unenforceable rule must never appear enforced. Whitelists are validated too — a malformed whitelist suppresses real rules, so it's a protection risk, not just a detection one.

Pattern-length backstop

safeRegExp now refuses a pattern over 1000 chars, covering caller-supplied bundles that never passed the validator. Being honest about the limit: a complete static ReDoS analysis isn't possible — a bounded/off-loop matcher remains the real fix and stays on the backlog.

Telemetry origin

PATCHSTACK_API_BASE is the origin the site api_key is exchanged against, so an injected value (CI/env) was a credential-exfiltration path. It must now be https (localhost allowed for local testing); anything else is refused with a warning and falls back to the default origin.

Test hygiene

egress-dns tests bind a loopback listener, which some sandboxed/CI environments refuse with EPERM — they now skip cleanly instead of failing and timing out. The suite is green both in a binding-capable environment and in a restricted one (724 passing either way); typecheck + build clean.

Notably, validation rejected nothing in the existing corpus/demo rule sets — the caps sit far above real rules.

[coderbuds]

Comprehensive rule validation and origin hardening greatly improves security robustness.

🎯 Quality: 100% Elite · 📦 Size: Large — consider splitting if possible

📈 This month: Your 53rd PR — above team average · Averaging Excellent

_{See how your team is trending →}

[patchstackdave]

/review