feat: add composable target runtime config contract#1625
Merged
Conversation
Deploying agentv with
|
| Latest commit: |
10f7162
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://afc79db1.agentv.pages.dev |
| Branch Preview URL: | https://av-y7eq-integration.agentv.pages.dev |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
AgentV now has one composable config graph for project config, global config, and eval-local overrides, with explicit target runtime contracts for coding agents. Evals can resolve
.agentv/config.yaml,$AGENTV_HOME/config.yaml, and*.eval.yamlas the same authored surface, including split field files viafile://,defaults.grader, andexecution.max_concurrency.This also makes target execution failures first-class run output. Providers now return structured target execution envelopes for subprocess, timeout, malformed output, sandbox, and coding-agent runtime failures, and run artifacts expose those envelopes to CLI summaries and the Dashboard.
Design Notes
.agentv/config.yaml, optional local overlays,$AGENTV_HOME/config.yaml, and eval YAML share the same config contract. Field-levelfile://refs load bare YAML values, not wrapped objects.--max-concurrency/commandLineOptions.maxConcurrency/evaluateOptions.maxConcurrency; this PR keeps Promptfoo compatibility reads but uses AgentV-authoredexecution.max_concurrencyfor the new YAML contract.codex,claude, andcopilotprovider names are rejected as ambiguous. Runtime-specific providers use explicit names such ascodex-cli,codex-app-server,codex-sdk,claude-cli,claude-sdk,copilot-cli,copilot-log, andcopilot-sdk.pi-cli,pi-rpc, andpi-sdkare explicit runtime choices.pi-rpcuses a stdio process boundary instead of being folded into CLI semantics.Brand casing work is separate: #1624 updates user-facing AGENTV/AgentV copy without changing identifiers.
Validation
git diff --checkbun run lintbun run validate:examplesbun test packages/core/test/evaluation/config.test.ts packages/core/test/evaluation/loaders/config-loader.test.ts packages/core/test/evaluation/providers/cli.test.ts packages/core/test/evaluation/providers/target-execution-artifacts.test.ts packages/core/test/evaluation/providers/sdk-child-provider.test.ts packages/core/test/evaluation/providers/sdk-provider-registry.test.ts packages/core/test/evaluation/providers/targets.test.ts packages/core/test/evaluation/providers/sandbox-runtime.test.ts packages/core/test/evaluation/providers/pi-runtime.test.ts packages/core/test/evaluation/providers/claude-provider-aliases.test.ts packages/core/test/evaluation/providers/copilot-cli.test.ts packages/core/test/evaluation/providers/normalize-tool-call.test.ts packages/core/test/evaluation/validation/targets-validator.test.ts packages/core/test/evaluation/validation/eval-validator.test.ts— 410 pass, 0 failbun --filter @agentv/core typecheckbun --filter agentv typecheckbun --filter @agentv/web build— exits 0; Pagefind still reports existing docs pages without outer<html>elementsDogfood Gaps
Live provider dogfood is still the main reviewer gate before merge. Deterministic Docker sandbox dogfood passed in the sandbox worker, but live OpenAI sandbox dogfood hit unavailable/invalid local credentials, and the Codex/Claude/Copilot/Pi live runtime checks were not completed because they would start real provider sessions plus grader runs from worker branches. This PR is pushed specifically for dogfooding those provider paths.
Related
Related: av-y7eq.1, av-y7eq.2, av-y7eq.3, av-y7eq.4, av-y7eq.5, av-y7eq.6, av-y7eq.7, av-y7eq.8