Repository navigation
fix: fix the CLI bugs an insights report surfaced - #1545
Merged
Merged
Conversation
…reply The supervisor moves a run: true agent to the next free host port when the configured one is taken, but readiness, liveness and every A2A request kept using the configured port. Another service on that port answered /health, so the agent looked ready while requests got a 404. Requests are now routed to the port the container was given, and the configured URL stays the agent's identity for the model, the task tracker and auth. - A readiness timeout reports the container's last log lines, or that it exited, before --rm deletes the container. - A2A_SubmitTask returns an agent's direct Message reply as the answer instead of failing with "no task ID received". - A JSON-RPC method-not-found error tells the model the agent speaks another protocol version and that retrying will not help. - A missing .env file is no longer reported on every agent start. - The container network is ensured right before docker run, since another session may remove it while an image is pulled.
… run Every session removes the shared network on shutdown when no container is attached. A session that ensured it before a long image pull could then fail with "network infer-network not found". EnsureNetwork now checks on every call, and MCP servers and the gateway ensure it after pulling.
fang title-cases the first word of an error, which turned flags such as --serve and --model into --Serve and --Model.
…tions - On the ripgrep path head_limit became --max-count, a per-file cap, while the description promises "| head -N". It now caps the total like the Go fallback does. - Grep says its regex syntax is RE2, without lookaround or backreferences. - SendSubagentInput no longer claims submit waits for the subagent, which led the model to pass submit=false to headless subagents.
The log section listed every warn-level line as a failure the sessions never recorded, although tool failure lines repeat the failed tool calls already counted. The table and digest now carry each line's level, the section is named "Log warnings and errors", and the analysis prompt is told not to count tool lines twice. A dropped live preview chunk of bash output is logged at debug. The full output still reaches the tool result.
4 tasks
Contributor
|
🎉 This PR is included in version 0.226.0 🎉 The release is available on: Your semantic-release bot 📦🚀 |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Summary
Fixes the real CLI bugs behind the failures in an
infer insightsreport. Each item was checked against the code first. The rest of the report was environment issues, expected behaviour or the model misusing a tool, and is not touched here.run: trueagent to the next free host port when the configured one is taken. Readiness, liveness and every A2A request kept using the configured port. A gateway on that port answered/health, so the agent looked ready while/a2areturned 404. Requests are now routed to the port the container was given, and the configured URL stays the agent's identity for the model, the task tracker and auth. The route table is per process: an agent started detached withinfer agents starton a reassigned port is still reached on its configured port from another process.--rmdeletes it.Messagereply as the answer instead of failing with "no task ID received"..envis no longer reported on every agent start.docker runfailed withnetwork infer-network not found.EnsureNetworkchecks on every call, and agents, MCP servers and the gateway ensure it right before they run.--serveinto--Serve.head_limitcaps the total on the ripgrep path like| head -N, not matches per file. The description says the syntax is RE2.submitwaits for the subagent.Test plan
task precommit:rungo test -race ./...--serve takes no task, wheremainprints--Serverun: trueagent and delegate a task to it