You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Agents can load packaged instruction skills and complete workflows that need several Orka-controlled tools. Microsoft Agent Framework exposes load_skill for SKILL.md files packaged in the image. The hosted Foundry model loop uses the same catalog and supports up to 16 sequential tool calls per user turn.
Each operational call returns to Orka. Hosted continuations validate response/call pairs, cache retries, preserve completed rounds across restarts, and retain bounded dialogue for session follow-ups. Hosted /responses acknowledges streaming requests before model work, both with and without brokered tools, and keeps the same response ID and creation time through completion or failure. A client disconnect cancels the active model request, allowing Orka to settle an acknowledged response through authenticated session stop and idle checks.
Brokered model failures preserve known error codes and the upstream HTTP status across continuations. Model and tool-validation messages come from a fixed list; malformed tool names, argument keys, and argument paths are omitted. Provider messages, headers, URLs, identifiers, and credentials never enter the diagnostic payload.
Explicit model HTTP 429 rejections can retry twice within the same hosted response. Retry delays honor server headers up to 60 seconds each; disconnecting the stream cancels the wait. Transport failures and other HTTP statuses are not retried. Streaming events carry sequence numbers for the acknowledgement and terminal result.
Validation:
Common Python suite: 868 passed, including early acknowledgement with and without tools, native model HTTP socket closure on disconnect, initial and continuation cancellation, stable response IDs and creation times, persisted replay after restart, exact message-size boundaries, safe model and tool-validation diagnostics, and bounded 429 retries. Real HTTP/ASGI tests verify disconnect during a 60-second backoff. Six retry regressions and twenty validation-error regressions fail against their preceding implementations.
Microsoft Agent Framework suite: 128 passed for the skill integration.
go test ./... and make lint passed.
Cross-repository tests with the real hosted AgentKit server and Foundry broker passed sequential tools, tool-error recovery, authorization denial, stripped-proof rejection, cancellation, and lease expiry. Dropping the acknowledgement keeps the request uncertain and prevents replay or deletion.
The cross-repository tests use model and Azure lifecycle fixtures. Orka external v2 human tool approvals remain unsupported.
result.usage only covers model calls since the latest resume, while state.initial_usage contains all earlier rounds and the final response reports their cumulative total. From the second operational tool onward, this intermediate function_call therefore underreports usage. Compute the combined usage once and use it for both the payload and the next state.
A resumed round that requests another tool reports only that resume's token usage, even though following.initial_usage and the final response accumulate all prior rounds. This undercounts the intermediate function_call response and makes usage jump inconsistently at completion. Pass the combined usage here as well.
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
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.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
Agents can load packaged instruction skills and complete workflows that need several Orka-controlled tools. Microsoft Agent Framework exposes
load_skillforSKILL.mdfiles packaged in the image. The hosted Foundry model loop uses the same catalog and supports up to 16 sequential tool calls per user turn.Each operational call returns to Orka. Hosted continuations validate response/call pairs, cache retries, preserve completed rounds across restarts, and retain bounded dialogue for session follow-ups. Hosted
/responsesacknowledges streaming requests before model work, both with and without brokered tools, and keeps the same response ID and creation time through completion or failure. A client disconnect cancels the active model request, allowing Orka to settle an acknowledged response through authenticated session stop and idle checks.Brokered model failures preserve known error codes and the upstream HTTP status across continuations. Model and tool-validation messages come from a fixed list; malformed tool names, argument keys, and argument paths are omitted. Provider messages, headers, URLs, identifiers, and credentials never enter the diagnostic payload.
Explicit model HTTP 429 rejections can retry twice within the same hosted response. Retry delays honor server headers up to 60 seconds each; disconnecting the stream cancels the wait. Transport failures and other HTTP statuses are not retried. Streaming events carry sequence numbers for the acknowledgement and terminal result.
Validation:
go test ./...andmake lintpassed.The cross-repository tests use model and Azure lifecycle fixtures. Orka external v2 human tool approvals remain unsupported.
Companion broker change: orka-agents/agent-runtime-foundry#3