[henrypark133]

Context

Integration and test-support workstream. It proves the skeleton pieces compose through caller-level surfaces and provides feature-gated support used by later host-port branches.

Master spec: docs/reborn/agent-loop-skeleton.md
Workstream brief: docs/reborn/agent-loop-briefs/e2e-integration-tests.md
Stack base: arch/ws-7

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

• ironclaw_agent_loop/test-support and Reborn/turns support helpers for planned-driver tests.
• Integration tests around planned-driver execution, prompt assembly, checkpoint/state refs, and loop-exit evidence surfaces that are available at this layer.
• Contract coverage for serialized host surfaces, including redaction/sentinel behavior.
• Updates to test support after the WS0 prompt-authority and inline-message changes.
• Sanitization fix for ParentLoopOutput::AssistantReply.content so redaction applies to serialized parent-loop output too.

Reviewer focus

• Tests should drive the caller path, not just helper methods.
• Feature-gated test support must not leak into production defaults.
• Redaction and serialized surfaces should never expose sentinel secret values.
• Future-only cases should stay documented as blocked until the relevant host port branch exists.

Non-goals / deferred work

• Real host capability/checkpoint/input/progress/cancellation/identity adapters are sibling branches WS9 through WS15.
• Default planned-driver registration is WS14.
• Product-live cutover is WS17.

Validation

• cargo test -p ironclaw_turns --test agent_loop_host_contract redaction_sentinels_never_leak_through_serialized_surfaces
• cargo check -p ironclaw_agent_loop --features test-support
• git diff --check
• Final ancestry check verified arch/ws-7 -> arch/ws-8.

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]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[serrrfirat]

Summary

Reviewed WS8 PR #3598 only.
Base 7b0097184af74860356838b1a288e80459c00c05 → head cbdd70890afb200cdfe47076b23a885f0c70cd8c.

No new product-path security finding in reviewed scope, but integration suite currently inherits/fails on executor test compile issue from #3596.

Validation:

• cargo check -p ironclaw_agent_loop ✅
• cargo test -p ironclaw_agent_loop --lib ❌: executor.rs:1136 missing LoopPromptBundle.instruction_fingerprint in test fixture.

Findings

#	Sev	Category	File:Line	Issue	Fix suggestion
inherited	Medium	Tests / Regression coverage	#3596	WS8 integration branch inherits non-compiling ironclaw_agent_loop lib tests from #3596 (LoopPromptBundle fixture missing instruction_fingerprint).	Fix #3596 test fixture and rebase WS8.

Security/data-flow notes

• New test-support exposure appears gated behind test/feature cfg.
• No test-only bypass found leaking into default prod build.

Correctness/invariant notes

• WS8 coverage cannot be trusted until inherited test compile failure is fixed.

Missing tests

• Fix inherited executor test compile failure, then rerun suite.

[zmanian]

Review notes — WS8 integration suite

Genuinely caller-driven (executor + PlannedDriver through public APIs), not glue-code happy paths. The redaction-sentinel test paired with the ParentLoopOutput::AssistantReply.content sanitization fix is exactly the regression it would have caught — that pairing is the kind of thing that proves the suite has teeth.

Deferred coverage tracking

Three gaps explicitly deferred to WS9–WS15, but the deferral is asymmetric: there are no negative tests pinning current behavior so the later workstreams could silently change semantics:

1. No cancellation negative test. WS13 brief defers cancellation, but nothing here asserts the executor does NOT short-circuit on a cancel signal today. WS13 could regress the contract invisibly. Suggest a #[ignore]'d placeholder test or a compile-time "not yet wired" assertion.
2. No checkpoint-resume integration test. from_checkpoint_payload is unit-tested, but no integration test drives LoopExecutionState::initial_for_run → serialize → reload → resume against the executor. Highest-value resume safety net; would catch schema drift between WS0 and any later state field added by WS9–WS15.
3. No mixed-outcome parallel batch. parallel_batch_runs_in_one_iteration only covers all-success. Mixed success/fail/gate in a single batch is the realistic shape and the likely regression target for stop-on-first-suspension.

Each of these can land in the relevant later WS PR — just track them so they don't get lost.

Minor

• MockHostCall::InvokeCapability { capability_name: String } uses String rather than the CapabilityId newtype — small types.md drift in test-support; low impact.
• lock_or_panic plus panic!("test bundle ref should be valid: {error}") inside the prompt builder will surface as a tokio task panic, not a clean test failure. Explicit expect at the call site would read better.
• Test name parallel_batch_runs_in_one_iteration overpromises — no #[tokio::test(flavor = "multi_thread")] so parallelism isn't actually exercised. The semantic claim ("one iteration") is what's tested; the name should reflect that.
• recovery_budget_exhaustion_uses_single_call_retry asserts call counts but not message ordering after each retry — a regression that drops the retry result on the wire would still pass.

test-support feature gating is right; no flake risk (no sleep, no wall-clock deadlines).

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!