fix(discovery): harden upstream-OpenAPI discovery (size cap, cache, SSRF, pricing, ERC-8004)#779
Merged
Conversation
… pricing, ERC-8004) #777's upstream-OpenAPI discovery feature had six issues: - getJSONMap accepted up to 8MiB and re-marshaled it into the shared "obol-skill-md" ConfigMap; anything over k8s's ~1MiB ConfigMap limit bricked static-site publishing for every offer. Cap fetch + re-marshal at 200KiB and fall back to buildOfferScopedOpenAPI when exceeded. - tryUpstreamOpenAPI ran inline (up to 3x3s GETs, uncached) inside buildOfferBundles, itself called under staticSiteMu on every offer's reconcile. A slow/flapping upstream serialized reconciles and flip-flopped the content hash, rolling the shared discovery pod on every flap. Added upstreamOpenAPICache, keyed by offer UID+generation: refresh() runs from each offer's own reconcile outside staticSiteMu, at most once per generation; buildOfferBundles now takes a lookup func and only ever reads the cache. - fetchUpstreamOpenAPI followed redirects and let Upstream.Namespace (offer-author-controlled) pick an arbitrary namespace, with none of the ReferenceGrant checks the Gateway API data path requires for cross-namespace targets. Disabled redirect following, pinned the probe to the offer's own namespace, and rejected non-simple openapiPath overrides (traversal / embedded scheme). - buildOfferWellKnownX402FromOpenAPI priced every paid op at the offer's default, ignoring spec.routes[] price overrides the non-upstream path already honors. Added routePriceOverride, matching the verifier's own pattern semantics. - buildOfferAgentRegistration never set registrations[], which ERC-8004 requires once an offer is on-chain; buyers using --expected-agent-id failed closed. Populated from status.AgentID + the resolved payment network's registry. - Services[].Endpoint origin scoping used HasPrefix(ep, origin), which admits an evil-suffix host like origin+".evil.tld". Fixed to exact match or origin+"/". - Deleted an unreachable duplicate security-empty check in openAPIOpIsPaid. Every fix has a test that fails without it (verified by reverting each change in isolation and re-running its test). Claude-Session: https://claude.ai/code/session_014YjPMViNrZ7zBVgUQzwEKk
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Blockers #2 & #3 + discovery mediums from the PR #756 review. Caps the accepted upstream OpenAPI (~200KiB, falls back to the offer-scoped doc) so one offer can't brick the shared static-site ConfigMap for all offers; moves the fetch into a per-offer generation-keyed cache out of the shared rebuild path so a slow/flapping upstream can't convoy reconciles or roll the discovery pod; disables redirect-following and constrains the fetch to the offer's own namespace (confused-deputy SSRF); honors per-route price overrides in the upstream-derived x402.json; populates registrations[] so
obol buy --expected-agent-idworks on dedicated origins; tightens same-origin scoping; drops a dead check. 9 new tests.https://claude.ai/code/session_014YjPMViNrZ7zBVgUQzwEKk