feat(fresh): FA3 — createMcpSandboxHandler (themed ui:// sandbox + CSP)#336
Conversation
|
[PHASE: IMPL] [VERDICT: PASS] FA3 slice landed for Summary
Commits
Test Evidence
Harness
|
|
@openhands-agent model=openrouter/qwen/qwen3.7-max output=pr-comment iterations=800 use harness — run a separate-session IMPL-EVAL for this PR (#336, FA3 — SKILL (read before evaluating)
What to verify against issue #252 (read
|
OpenHands Agent — CompletedModel: openrouter/qwen/qwen3.7-max IMPL-EVAL Summary — PR #336 (FA3 — createMcpSandboxHandler)Run ID: 28640458812-1 SummaryPartial separate-session IMPL-EVAL of PR #336 against issue #252 and the The PR body claims PASS across all gates in a prior session; I reproduced the focused test suite, the scoped ChangesThis run is read-only with respect to source. No feature files, lock files, or source files were modified. Files inspected (read only):
No ValidationCommands actually executed in this run, with observed results:
Gates not completed in this run:
Remaining risks
Verdict position (advisory, not binding)Tentative FAIL_FIX on the sole confirmed finding (F-5: mod.ts exports = 22 > 20, introduced by the +2 from this PR), pending completion of the remaining gates and reproduction of the root/scoped lint/fmt/type-soundness checks. Binding verdict deferred to a follow-up evaluator run. |
…ock hygiene (IMPL-EVAL #336)
|
[PHASE: IMPL-EVAL-FIX] [VERDICT: PASS] Addressed the separate-session IMPL-EVAL findings for PR #336. F-5 Export Cap FixApproach: split the cohesive MCP sandbox surface into a dedicated subpath:
New Lock Hygiene
Validation
Commit
|
|
@openhands-agent model=openrouter/qwen/qwen3.7-max output=pr-comment iterations=1500 use harness — run a complete separate-session IMPL-EVAL for PR #336 (FA3 — Context: what changed since the prior partial evalThe prior run found ONE hard finding:
SKILL (read before evaluating)
Gates to issue a binding verdict on (against issue #252)
OutputEmit a BINDING PASS / FAIL_FIX / FAIL_RESCOPE / FAIL_DEBT verdict with a complete per-gate evidence table (commands + results) and, if not PASS, a precise fix list. Write |
OpenHands Agent — CompletedModel: openrouter/qwen/qwen3.7-max PR #336 — FA3 MCP Sandbox Handler — BINDING VERDICT: PASSSummaryComplete IMPL-EVAL evaluation of PR #336 (feat(fresh): FA3 — createMcpSandboxHandler) against issue #252. All gates pass. Binding verdict is PASS. Changes
ValidationGate results (all pass)
Security audit (source review + test assertions)
Prior run findings (addressed in e7bf3f8)
Remaining risksNone identified. The 3 pre-existing A8/AP-9 WARN + 1 A9 INFO on FA1 files ( VerdictPASS — All gates satisfied. Fix commit Run: https://github.com/rickylabs/netscript/actions/runs/28643885199 |
Summary
Adds
createMcpSandboxHandlerto@netscript/fresh/aifor serving registered MCPui://resources as themed, CSP-constrained sandbox documents. The handler injects active--ns-*design tokens, supports?theme=, derives per-response CSP from the resource URI, and propagates the requestAbortSignalinto lookup ports.Scope
packages/freshruntime AI subpathSlices
88e7ae3088e7ae30069bd8d2Gate Evidence
.llm/tmp/run/feat-fresh-ai-fa3-mcp-sandbox--fa3/plan-eval.md.llm/tmp/run/feat-fresh-ai-fa3-mcp-sandbox--fa3/evaluate.mddeno test --allow-all packages/fresh/src/runtime/ai/mcp-sandbox-handler_test.tsdeno doc --lint packages/fresh/src/runtime/ai/mod.tsdeno run --allow-read --allow-run .llm/tools/run-deno-check.ts --root packages/fresh --ext ts,tsxdeno run --allow-read --allow-run .llm/tools/run-deno-lint.ts --root packages/fresh --ext ts,tsxdeno run --allow-read --allow-run .llm/tools/run-deno-fmt.ts --root packages/fresh --ext ts,tsxrtk proxy deno task arch:checkrtk proxy deno task checkrtk proxy deno task lintrtk proxy deno task fmt:checkrtk proxy deno task publish:dry-run--allow-slow-types; existing workspace warnings onlypasses the incoming request AbortSignal to the resource resolverHarness
.llm/tmp/run/feat-fresh-ai-fa3-mcp-sandbox--fa3/Drift / Debt