Repository navigation
Conversation
…a multi-provider inner factory
…runner backend seam
…ompatibility facade
…te realm validation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Step 7 of the execution-architecture series tracked in #527: agent steps now execute through the runner-backend seam (
RunnerBackend.run_agent), so an agent — together with its stdio MCP servers, skills, plugins, and session state — can run in a chosen execution realm instead of only in-process. Workflows without realm profiles behave byte-for-byte as before.Execution realms
execution.profileon agent steps selects a realm: local (today's in-process behavior, unchanged), Docker, or ACA. Model provider and execution placement are orthogonal (provider: copilot|openai|claude+ any realm).docker execstreaming bridge to a loopback-bound runner — no published ports, works with a remoteDOCKER_HOST, no bind mounts (skills/plugins arrive via the run bundle with staged path mapping into/workspace)./healthhandshake: a stale image fails at realm startup with a named rebuild instruction, not mid-run.Runner
skill_directories,custom_agents, per-callenv_overlay, andexecution_idon the wire; the runner stops dropping components.POST /interruptwith per-execution targeting: concurrent calls are interrupted individually, unknown ids 404, terminal ids 409, partial results framed honestly; hosts degrade gracefully on images without the feature.Secrets
ACA
aca:block, with lease-owned session identifiers and the transport extracted from the provider.provider: acakeeps working unchanged through a compatibility facade (identical wire requests and identifier semantics, verified by equivalence tests) and now emits a once-per-run deprecation notice pointing at the profile form.conductor.providers.aca_protocolshim is removed; imports migrate toconductor.runner.protocol.Validation & pinning
Tests & CI
.cmdmechanics), and a docker-integration lane that builds the runner image from the PR head SHA (fetchability-checked) with staged-skill, concurrency, cancel-cleanup/sweep, and interrupt end-to-end cases.Docs:
docs/design/agent-realms.md,docs/configuration.md,docs/workflow-syntax.md,docs/providers/aca.md(deprecation + migration guide), runnable examples (examples/docker-agent-realm.yaml,examples/aca-profile.yaml), and a towncrier fragment.Test plan
make test— full suite green (12,998 passed; the only parallel-mode failures are three known Rich-render wrapping flakes that pass serially).make check— ruff + format + ty clean (4 pre-existingunused-ignorewarnings in the untouchedclaude_agent_sdk.py).make validate-examplesgreen;conductor validate examples/docker-agent-realm.yaml --environment …resolves the realm offline.DOCKER_HOSTcase env-gated)..cmdshim mechanics, binary-safe NDJSON pipes, case-insensitive host-path matching for staged mapping).Part of the PR series for #527.