Repository navigation
clef-proxy: a JSON array body passes the #2816 'oracle' guard and fabricates an upstream Clef call — typeof [] is 'object', so [] and [1,2] run with the default model and empty state/questions instead of a 400 #2938
Description
Summary
bin/clef-proxy.mjs is held up in #2816 as one of the two proxies that
already handle non-object JSON bodies correctly — the oracle the other
seven are to copy. Its guard does not actually reject arrays:
// bin/clef-proxy.mjs:211 (main 16bae8ea)
if (!body || typeof body !== 'object') { /* 400 */ }typeof [] === 'object', so a JSON array body passes the guard. The
handler then reads body.model / body.state / body.questions off the
array (all undefined) and fabricates an upstream Clef call with the
default model and the payload { state: {}, questions: {} } — the same
fabrication class #2821 filed for antigravity-direct, in the one proxy
#2816's fix is most likely to skip because its body text calls clef
already-correct (its acceptance bullet 4 even says "no behaviour change
for ... the guarded clef / opencode-route proxies").
The sibling oracle gets this right: opencode-route-proxy.mjs:659-660
defines isPlainObject as
typeof v === "object" && v !== null && !Array.isArray(v). Clef is the
only bin/*-proxy.mjs handler using the bare typeof guard (grep over
bin/*.mjs, this run).
Evidence (verified this run, main 16bae8ea)
Probe on the laptop (~/git/fleet/probe-2026-10-30-laneB-clefarray/,
real clef-proxy.mjs + proxy-port.mjs, md5-identical to main —
clef-proxy 5e55c6637574d1c69c143bd4430bc030), proxy spawned on a
scratch port with a dummy CLOUDFLARE_API_TOKEN, POST /v1/systemone:
| body | result |
|---|---|
[] |
HTTP 401 from the upstream call — i.e. NOT a 400: the proxy fabricated {state:{}, questions:{}}, called the Cloudflare gateway with the default model, and relayed its auth error |
[1,2] |
same 401 upstream relay |
{} (valid control) |
same 401 upstream relay — the array responses are indistinguishable from a valid empty-object request |
"hello" |
HTTP 400 "Invalid JSON body: expected an object" |
null |
HTTP 400 (the #2816 case — this part of the guard works) |
42 |
HTTP 400 |
Process stayed alive throughout (this is a fabrication bug, not the
#2816 crash).
Why it matters
- A client bug or stray probe that POSTs an array gets a real upstream
Clef execution attempted with a defaulted model and empty
state/questions, and the response it gets back (an upstream
status/error) points at Cloudflare, not at its own malformed request. - Process-of-record risk: bin proxies: a JSON null body crashes seven chat proxies — parse succeeds, first deref throws, process exits 1 (clef and opencode-route already 400) #2816's evidence names clef as the pattern to
copy. If its fix lands by copying clef's guard shape — or by skipping
clef per its "no behaviour change" bullet — the array hole is
cemented into whichever proxies adopt the baretypeofcheck, and
bin proxies: a JSON null body crashes seven chat proxies — parse succeeds, first deref throws, process exits 1 (clef and opencode-route already 400) #2816 closes with this still live. This issue exists so clef is in
somebody's evidence set; if the bin proxies: a JSON null body crashes seven chat proxies — parse succeeds, first deref throws, process exits 1 (clef and opencode-route already 400) #2816 PR fixes clef too, close this
as completed by that PR and say so.
Acceptance criteria
-
POST /v1/systemone(and/systemone,/) with a JSON array
body returns the proxy's existing 400 "expected an object" error
before any upstream call is attempted;null, string, and
number bodies still 400; object bodies are unchanged. - The guard matches the
opencode-route-proxy.mjsisPlainObject
semantics (plain object only), or an equivalent shared check —
triage pins whether clef imports/copies it. - A regression test in
bin/clef-proxy.test.mjs(existing harness
style): array body → 400 and no upstream fetch attempted;
object-body control still reaches the upstream path. - Cross-check bin proxies: a JSON null body crashes seven chat proxies — parse succeeds, first deref throws, process exits 1 (clef and opencode-route already 400) #2816's PR when it lands: any proxy that adopted a
baretypeof body !== 'object'guard gets the same array test.
Dedup checked this run: open/closed search for "clef array",
"clef-proxy body" — no hits. #2816 covers the seven crashing proxies
and names clef as a correct oracle (its secondary note mentions array
bodies only for design-code's dispatch); #2821 is antigravity-direct,
a different file outside the bin proxy set.
Activity
dezren39 commentedon Oct 31, 2026
🤖 Agent note · filed
dezren39 commentedon Oct 31, 2026
🤖 Design-architect pass 26 (2026-10-31) — agent note: this issue is adopted as part of Phase 1 of design #3068 ("parsed is not validated"). Triage: do NOT re-file, re-scope or split it — record the phase structure instead. The design records the rule this phase lands under: validate the shape you deref at the boundary, before any side effect; "object" means the exact plain-object vocabulary (opencode-route-proxy.mjs:659 idiom); failure is the surface's JSON 400 with the process alive; one shared probe matrix (null, array, scalar, {}, each relied-on field wrong-typed). Phase-specific decisions are on #3068 and in files/designs/code-parsed-is-not-validated.md (goal workspace).
84 remaining items
dezren39 commentedon Nov 2, 2026
🤖 Agent note · resumed
queue/5483-2938-clef-array-guard) already implements the pinned fix exactly —|| Array.isArray(body)on thebin/clef-proxy.mjssystemone guard — plus the clef-proxy: a JSON array body passes the #2816 'oracle' guard and fabricates an upstream Clef call — typeof [] is 'object', so [] and [1,2] run with the default model and empty state/questions instead of a 400 #2938 regression test (array bodies[]/[1,2]-> 400, zero stub-upstream hits). Codex review on PR fix(clef-proxy): reject array bodies in the systemone guard (#2938) #4449 completed with no findings. Laptop SSH is unreachable this shift (connection closed, 3 attempts), so the test file could not be re-run locally; merging on diff-vs-pinned-spec verification plus the passing PR check, per standing merge authority.Merge pull request #4449 from developing-today/queue/5483-2938-clef-a…
dezren39 commentedon Nov 2, 2026
🤖 Agent note · done
74feb21f1fd29d2c8f7573d0f71247490496dbdd).bin/clef-proxy.mjson main shows the guard at line 235:if (!body || typeof body !== 'object' || Array.isArray(body)), with the clef-proxy: a JSON array body passes the #2816 'oracle' guard and fabricates an upstream Clef call — typeof [] is 'object', so [] and [1,2] run with the default model and empty state/questions instead of a 400 #2938 regression test inbin/clef-proxy.test.mjs. Codex review on PR fix(clef-proxy): reject array bodies in the systemone guard (#2938) #4449: no findings. Laptop SSH was unreachable this shift, so the test file was not re-run locally by this lane.