Known Issues
Everything here is real, reproduced, and open against v6.0.0. It is published rather than quietly tracked because most of these fail silently - you get a plausible wrong answer, not an error - and an hour spent debugging one of them is an hour this page could have saved.
The worst kind left: a tool returning something unserializable still reaches the model as a raw memory address. If an agent’s answers look subtly corrupted, read that entry first.
Silent wrong answers
These produce no error. Nothing in the log says anything went wrong.
A tool returning something unserializable reaches the model as a memory address
Return a value JSON cannot carry and it is neither rejected nor flagged - it is coerced to its
repr(), and that string, typically containing a raw memory address, enters both the event log
and the prompt. result_sha256/result_size are computed over that same repr, so two identical
non-serializable results are recorded as two different ones.
This was filed as #251 and closed by folding its done-when items into #250, on the reasoning that both are “a tool result mishandled in the same translation function”. #250’s fix shipped only the raise half; this half - the warning and the hashing fix - was never implemented, and no open issue tracks it.
@tool
def get_user(user_id: str) -> object:
return object() # not JSON-serializable# the model receives, and the event log records byte-for-byte:
<object object at 0x7fb09bf28b30>Until it is fixed: return JSON-compatible values from tools. If you return an object today, check what the model is actually receiving. → #251 , folded into #250 , which did not implement this half
Rough edges
Not silent, not traps - just things that will cost you a few minutes.
| What | Issue |
|---|---|
A binding cannot supply per-request context: DeckGateway.start() takes none, so a run started over Native.http(), AGUI.http() or Terminal.stdio() gets context=None. A deck-level Deck(context=...) is the way to give tools their dependencies; a request cannot vary it | #227 |
run()’s return type makes the documented interrupt idiom fail a type checker | #231 |
| The CLI can send signals but cannot read anything - no inbox, no run view | #256 |
Fixed in v6.0.0
Removed from this page rather than struck through, because a Known Issues page that lists fixed things teaches you to distrust the entries that are still true. Every release’s fixes in full are in the changelog.
| What | Issue |
|---|---|
A @tool exported beside a @workflow became a runnable target and could be started by name | #488 |
A workflow’s own InputError after run.started reached the wire as engine_error with its message stripped | #621 |
A failure late in Deck.__aenter__ left observers open and the process claim held, blocking a second Deck(...) | #572 |
A workflow input mapping that did not match its parameters raised ConfigError rather than InputError | #583 |
usage.usd used to be listed on this page. It isn’t a defect: agentdeck does not price model
calls, on purpose (#177 ) - see
TurnResult for the ruling, not a fix.
What is being done about all this
v6.0.0 replaced the serve surface with bindings, which is why #227 reads the way it does now: the limitation survived the rewrite in a new shape. The reference bindings still to come, and the read-path CLI in #256, are tracked on the v6.x milestone .
Hit something that is not here? Open an issue . A reproduction is worth more than a diagnosis - most of this page came from people who sent one.