arch(level2-merged): integrate ws-4 + ws-5 - #3557
henrypark133 wants to merge 11 commits into
Conversation
There was a problem hiding this comment.
Code Review
This pull request implements the Planner facade and the reference baseline implementations for the nine agent-loop strategies, including Context, Capability, Model, Batch, Gate, Recovery, Stop, Drain, and Budget. It introduces the DefaultPlanner as a composition layer and the PlannerId type for stable identification in checkpoints. The reviewer feedback correctly identifies that while the production-ready Default*Strategy implementations are now available, the DefaultPlanner is still using internal PlaceholderStrategy instances. Suggestions were provided to update the imports and the Default implementation of the planner to use these new production strategies.
8c3b67b to
3359dfb
Compare
0b43ead to
2072ba5
Compare
# Conflicts: # crates/ironclaw_agent_loop/Cargo.toml # crates/ironclaw_agent_loop/src/strategies/budget.rs # crates/ironclaw_agent_loop/src/strategies/context.rs # crates/ironclaw_agent_loop/src/strategies/drain.rs # crates/ironclaw_agent_loop/src/strategies/stop.rs
3359dfb to
22a8e1d
Compare
2072ba5 to
255c8c0
Compare
SummaryReviewed Level2 integration PR #3557 only. Merge stance: no unique Level2 merge-only finding, but this PR inherits blocking issues from #3555 and #3556. Tests run:
Findings
Security/data-flow notes
Correctness/invariant notes
Missing tests
Suggested fixes
|
Review notes — Level2 integrationClean mechanical join with good encapsulation discipline. Two observations worth tracking. Silent defaults promotionWS5 defaults are now implicitly the production defaults for This is the same class of risk as the No end-to-end testTitle says "integration" but the verification is mechanical join + already-present unit tests. There is no test that simultaneously drives multiple strategies through a shared Minor
|
# Conflicts: # crates/ironclaw_agent_loop/src/families/mod.rs # crates/ironclaw_agent_loop/src/family.rs # crates/ironclaw_agent_loop/src/strategies/batch.rs # crates/ironclaw_agent_loop/src/strategies/budget.rs # crates/ironclaw_agent_loop/src/strategies/capability.rs # crates/ironclaw_agent_loop/src/strategies/context.rs # crates/ironclaw_agent_loop/src/strategies/drain.rs # crates/ironclaw_agent_loop/src/strategies/gate.rs # crates/ironclaw_agent_loop/src/strategies/mod.rs # crates/ironclaw_agent_loop/src/strategies/model.rs # crates/ironclaw_agent_loop/src/strategies/recovery.rs
|
Addressed the Level2 review feedback in What changed:
Local validation:
|
|
Follow-up from the final review pass landed in The default family digest now has an explicit Additional validation after this follow-up:
|
# Conflicts: # crates/ironclaw_agent_loop/Cargo.toml # crates/ironclaw_agent_loop/src/default_planner.rs # crates/ironclaw_agent_loop/src/families/mod.rs # crates/ironclaw_agent_loop/src/strategies/budget.rs # crates/ironclaw_agent_loop/src/strategies/context.rs # crates/ironclaw_agent_loop/src/strategies/drain.rs # crates/ironclaw_agent_loop/src/strategies/stop.rs
Context
Integration branch for the planner/default-strategy layer. It produces the base used by the canonical executor workstream.
Master spec:
docs/reborn/agent-loop-skeleton.mdWorkstream brief:
docs/reborn/agent-loop-skeleton.mdStack base:
arch/level1-mergedLatest stack maintenance on 2026-05-14:
origin/reborn-integrationthrough WS17 before publishing the PR descriptions.What landed
arch/ws-4andarch/ws-5onto the current level1 integration branch.Reviewer focus
Non-goals / deferred work
Validation
cargo check -p ironclaw_agent_loop -p ironclaw_rebornarch/level1-merged -> arch/level2-merged.Stack position