You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
id: design: shared threat-list feed protocol — versioned, signed, subscribable feeds with local override lists #6006
Part of the worlds design program: #5955. Wave-2 follow-up from the prior-art research (Lane B item 25); parent topic issue: #5973. Research source: workspace/scratch/worlds-capabilities-brief-2026-11-04/prior-art-research.md.
Scope
The live half of moderation sharing (sibling pack-format follow-up under #5955): adversarial data — scam domains, content hashes, known-raider principals — whose value is freshness. Precedent: cross-server network blacklists maintained by third-party bots (Paladin: on by default, opt-out per server; https://www.shapedimage.com/paladin-docs.html).
Expectation: s — verdict with evidence, fix path scoped
Links: follow-up task 10543
Verdict: CONFIRMED / still open — protocol undesigned and unimplemented.
Evidence at current main (2026-11-05):
Code search finds no threat-list feed protocol in this repo: threat hits only threat-model prose (e.g. pkgs/mcpx/docs/proposals.md, design docs); moderation feed finds no protocol; blacklist finds only lib/notifications.py (a notification filter) and Nix kernel-module lists — no feed format, publisher, subscriber, signing, versioning, or expiry machinery for scam domains / content hashes / known-raider principals.
No comments on this issue dispute the scope; the only comment is the Kaneo tracker link.
Fix path scoped: follow-up task 10543 (small_fix, pilot queue) to write the feed-protocol spec #6006 asks for — entries/versioning/signing, update cadence/expiry, subscription semantics with local override/exemption lists always winning, and stated behavior for stale feed, malicious feed, and feed-vs-feed conflicts, sufficient to implement a publisher and a subscriber. Implementing the publisher/subscriber itself is explicitly out of that follow-up's scope.
Expectation: m — spec PR opened and merged under the standing criteria; spec covers entries/versioning/signing, cadence/expiry, subscription semantics with local overrides always winning, and stale/malicious/conflicting-feed behavior.
Spec covers every Scope bullet in issue #6006: feed format (signed FeedHead/FeedDelta/FeedSnapshot objects on the pkgs/id substrate) and entries (canonicalized domain / content-hash / principal subjects, actions, closed reason vocabulary); versioning (hash-chained monotonic seq, set_hash convergence oracle); signing composed with writer-set authority (#5545) — a feed counts only where a world's moderation pack (#5973) pins it, with signed rotation/revocation; update cadence and two-clock expiry (entries re-added, never silently extended); subscription semantics with local override/exemption lists evaluated first and always winning; and stated behavior for stale feeds (fresh → stale → retired; silencing a feed never unblocks a subject before its own expiry), malicious/corrupt feeds (reject-whole, fork freeze for a pack decision, volume caps), and feed-vs-feed conflicts (deterministic rank → strength → higher-rank signed clear → earliest → entry-id order). Implementation slices are named in the spec §10 as separate future issues, per the task packet's scope guard. No Codex findings were posted on PR #6232 (checked all review rounds before merge); no CI check runs exist for this docs-only PR.
Activity
developing-today commentedon Nov 5, 2026
CODE-1672
dezren39 commentedon Nov 5, 2026
🤖 Agent note · done (triaged)
Verdict: CONFIRMED / still open — protocol undesigned and unimplemented.
Evidence at current main (2026-11-05):
threathits only threat-model prose (e.g.pkgs/mcpx/docs/proposals.md, design docs);moderation feedfinds no protocol;blacklistfinds onlylib/notifications.py(a notification filter) and Nix kernel-module lists — no feed format, publisher, subscriber, signing, versioning, or expiry machinery for scam domains / content hashes / known-raider principals.Fix path scoped: follow-up task 10543 (small_fix, pilot queue) to write the feed-protocol spec #6006 asks for — entries/versioning/signing, update cadence/expiry, subscription semantics with local override/exemption lists always winning, and stated behavior for stale feed, malicious feed, and feed-vs-feed conflicts, sufficient to implement a publisher and a subscriber. Implementing the publisher/subscriber itself is explicitly out of that follow-up's scope.
5 remaining items
developing-today commentedon Nov 5, 2026
🤖 Agent note · picked up
Merge pull request #6232 from developing-today/queue/10543-6006-threa…
developing-today commentedon Nov 5, 2026
🤖 Agent note · done
doc/2026-11-05T13-36-21Z_design_threat_list_feed_protocol/on mainSpec covers every Scope bullet in issue #6006: feed format (signed FeedHead/FeedDelta/FeedSnapshot objects on the pkgs/id substrate) and entries (canonicalized domain / content-hash / principal subjects, actions, closed reason vocabulary); versioning (hash-chained monotonic seq, set_hash convergence oracle); signing composed with writer-set authority (#5545) — a feed counts only where a world's moderation pack (#5973) pins it, with signed rotation/revocation; update cadence and two-clock expiry (entries re-added, never silently extended); subscription semantics with local override/exemption lists evaluated first and always winning; and stated behavior for stale feeds (fresh → stale → retired; silencing a feed never unblocks a subject before its own expiry), malicious/corrupt feeds (reject-whole, fork freeze for a pack decision, volume caps), and feed-vs-feed conflicts (deterministic rank → strength → higher-rank signed clear → earliest → entry-id order). Implementation slices are named in the spec §10 as separate future issues, per the task packet's scope guard. No Codex findings were posted on PR #6232 (checked all review rounds before merge); no CI check runs exist for this docs-only PR.