lsm / HyperNeo / 30762061474

83%

push

github

feat(space): spike — github connector + generic validator (re-express pr_ready/pr_merged/codex) [#2300] (#2312)

* feat(space): spike — re-express pr_ready/pr_merged/codex over github connector (#2300)

Throwaway, unwired spike behind HYPERNEO_WORKFLOW_CONNECTORS_SPIKE that proves
the L2/L3/L4 abstraction can carry the three coding capabilities with zero
engine special-casing. Not imported by any production path; exercised by tests
only. Adds a Connector interface + github connector (getPr/getPrReadiness/
getReactions), a domain-agnostic predicate language, and a generic
external_state validator parameterized by (connector, op, predicate). The three
capabilities become presets over that validator. Verdict: GO for all three
(codex fits via freshness-as-op-param + poll-until composition).

* fix(space): address #2312 review — 4 blocking fixes in connector spike (#2300)

- Bind `-f cursor=` on paginated review-thread requests (the >100-threads
  branch declared $cursor but never supplied it; GitHub rejected page 2).
- pollUntilAllow now converts only predicate-pending results (tagged
  externalStatePending) to allow, never a connector failure (rate limit /
  outage) — mirroring the production gate, which fails before its timeout
  when the fetch errors.
- getPrReadiness fails closed (parses the canonical prOutcome.url instead
  of fabricating an empty unresolved-thread list) when the input selector
  is noncanonical.
- getPath reads own properties only (hasOwnProperty guard) so
  present:'__proto__' is no longer a tautology.

Adds 6 tests (pagination cursor, fail-closed, canonical fallback, getPath
prototype guard, pending tag, pollUntilAllow-failure passthrough). The 8
non-blocking findings (SSRF allowlist, GraphQL error-body rate limits, codex
freshness/audit metadata, hostHint, shared deadline, prUrl hardcode,
CLOSED+UNKNOWN ordering) are deferred to P1/P2 and acknowledged on #2300.

12282 of 17501 branches covered (70.18%)

Branch coverage included in aggregate %.

564 of 604 new or added lines in 7 files covered. (93.38%)

104131 of 122178 relevant lines covered (85.23%)

297.95 hits per line