arch(level1-merged): integrate ws-1 + ws-2 + ws-3 + ws-3.5

[henrypark133]

Context

Integration branch for the first parallel layer above WS0. It combines alpha/beta/gamma strategy traits and the loop-family registry so downstream planner/default-strategy work has one coherent base.

Master spec: docs/reborn/agent-loop-skeleton.md
Workstream brief: docs/reborn/agent-loop-skeleton.md
Stack base: arch/ws-0

Latest stack maintenance on 2026-05-14:

• Rebased this branch onto its current stack base after the WS0 prompt-authority fix and the follow-up WS8/WS14-parent/WS16/WS17 conflict resolutions.
• Pushed the updated branch with force-with-lease where the remote already existed, or published it as a new branch where it did not.
• Verified the final ancestry chain from origin/reborn-integration through WS17 before publishing the PR descriptions.

What landed

• Merged arch/ws-1, arch/ws-2, arch/ws-3, and arch/ws-3.5 on top of the current WS0.
• Resolved shared strategy-module and family-registry conflicts produced by the current WS0 prompt-authority fix.
• Produced a single downstream base for WS4 and WS5.
• Kept integration-only scope: no additional product/runtime behavior beyond making the workstreams compose.

Reviewer focus

• Cross-branch strategy exports and module visibility.
• Registry/family naming consistency with the skeleton spec.
• Whether this branch contains only integration conflict resolution and no unrelated refactor.

Non-goals / deferred work

• Planner facade behavior lands in WS4.
• Default strategy implementations land in WS5.
• Executor/runtime behavior lands later.

Validation

• cargo check -p ironclaw_agent_loop -p ironclaw_reborn
• Final ancestry check verified arch/ws-0 -> arch/level1-merged.

Stack position

[#3550 ws-0] state/checkpoint foundation -> reborn-integration
   |-- #3551 ws-1 strategy alpha -> ws-0
   |-- #3552 ws-2 strategy beta -> ws-0
   |-- #3553 ws-3 strategy gamma -> ws-0
   |-- #3643 ws-3.5 loop family registry -> ws-0
   '-- #3554 level1-merged -> ws-0
         |-- #3555 ws-4 planner facade -> level1
         |-- #3556 ws-5 default strategies -> level1
         '-- #3557 level2-merged -> level1
               '-- #3596 ws-6a canonical executor -> level2
                     '-- #3597 ws-7 PlannedDriver adapter -> ws-6a
                           '-- #3598 ws-8 integration/test support -> ws-7
                                 |-- #3644 ws-9 capability host wiring -> ws-8
                                 |-- #3645 ws-10 checkpoint load/resume -> ws-8
                                 |-- #3646 ws-11 input port -> ws-8
                                 |-- #3647 ws-12 progress port -> ws-8
                                 |-- #3648 ws-13 cancellation accessor -> ws-8
                                 |-- #3649 ws-15 prompt/identity context -> ws-8
                                 '-- #3650 ws-14-parent integrated host ports -> ws-8
                                       '-- #3651 ws-14 planned default registration -> ws-14-parent
                                             '-- #3652 ws-16 live runtime wiring -> ws-14
                                                   '-- #3653 ws-17 product live cutover -> ws-16

[gemini-code-assist]

Code Review

This pull request introduces the strategy trait contracts for the agent-loop framework, defining traits and associated types for batch policy, budget, capability filtering, context planning, input draining, gate handling, recovery, and stop conditions. The review feedback highlights a naming contradiction in the UnlimitedBudget implementation, a misplaced test in drain.rs, and an inconsistency in the serialization tagging of the StopOutcome enum relative to other strategy outcomes.

[serrrfirat]

Summary

Reviewed Level1 integration PR #3554 only.
Base 3db2d97bf1718554c8936e2ad4346794f02b7ec3 → head 22a8e1d4df8022962a2a9911100d47e5c397c816.

Merge stance: no unique Level1 integration finding beyond WS3.5 issues already posted on #3643.

Tests run:

• cargo test -p ironclaw_agent_loop --lib ✅

Findings

#	Sev	Category	File:Line	Issue	Fix suggestion
inherited	Medium	Correctness / Versioning	#3643	This integration PR inherits #3643’s zero-digest ComponentIdentity issue.	Fix in #3643/source branch.
inherited	Medium	Boundary / Registry integrity	#3643	This integration PR inherits #3643’s duplicate family-id silent overwrite issue.	Fix in #3643/source branch.

Security/data-flow notes

• Strategy contracts remain crate-internal and sanitized.
• No unique raw input/secret/auth expansion found in Level1 merge glue.

Correctness/invariant notes

• Family registry/versioning concerns are inherited from WS3.5, not newly introduced by merge resolution.

Missing tests

• Same as arch(ws-3.5): add loop family registry #3643: non-zero/content-derived family digest, duplicate family-id rejection.

Suggested fixes

1. Fix inherited arch(ws-3.5): add loop family registry #3643 findings at source, then rebase this integration branch.

[zmanian]

Review notes — Level1 integration

Mechanical join is clean; nothing to propagate back into WS1/WS2/WS3 individually. One ask before promoting out of draft.

Validation bar is too low

The PR checklist runs cargo check -p ironclaw_agent_loop -p ironclaw_reborn. Per repo CLAUDE.md the bar is fmt + clippy + tests; for an integration PR whose job is to prove four trait sets compose, "compiles" understates the verification. Recommend cargo test -p ironclaw_agent_loop -p ironclaw_reborn plus cargo clippy --all --tests before un-drafting.

One cross-trait state-composition smoke test

Every test here is unit-level: object-safety + serde round-trip per strategy file. There is no test that simultaneously drives multiple strategies through a shared LoopExecutionState transition — e.g., GateOutcome::Block carrying GateStrategyState round-trips through state, or that the recovery re-export in strategies/mod.rs actually composes with WS0's state types. Defensible because executor behavior lands in WS6, but worth one smoke test demonstrating the four PRs compose at the data layer, not just the type layer.

Minor

• app_loop_family.rs test asserts the registry binds default but doesn't assert the family's version().digest is the all-zero placeholder (which is what families::default() actually returns). If WS-5 swaps in real digests this test silently keeps passing. Sharper assertion would catch the drift.
• LoopFamilyId::new takes Cow<'static, str> with no validation (empty, length cap, character set), unlike sibling typed IDs (CapabilityId::new returns Result). Inconsistent with the typed-ID pattern.
• pub(crate) use ironclaw_turns::run_profile::ConcurrencyHint; in strategies/mod.rs is an odd re-export — drop it; callers can import directly.

[henrypark133]

Addressed the Level1 integration feedback in fa888339a.

What changed:

• Merged the fixed WS0, WS1, WS2, WS3, and WS3.5 heads into Level1.
• Added a cross-strategy smoke test that applies gate/recovery/stop outcomes into one LoopExecutionState and round-trips the composed state.
• Switched StopOutcome to the same internally tagged outcome wire shape as the other strategy outcomes and updated the stop tests.
• Removed the ConcurrencyHint re-export from strategies/mod.rs.
• Carried the WS3.5 registry fixes into Level1: validated LoopFamilyId, duplicate family IDs return a typed error, and the default family digest is non-zero and asserted.
• Updated Reborn milestone tests to build prompt bundles through the host prompt port before stream_model.
• Tightened model-visible redaction so bare env-var names remain readable while assigned credential values are redacted.

Local validation passed:

• cargo fmt --check
• git diff --check
• cargo test -p ironclaw_turns --lib
• cargo test -p ironclaw_agent_loop --lib
• cargo test -p ironclaw_reborn
• cargo clippy -p ironclaw_turns -p ironclaw_agent_loop -p ironclaw_reborn --all-targets -- -D warnings

GitHub checks are running on the new head.

[henrypark133]

Closing pull request. the changes are already merged in.