fix: treat daily-memory no-op as success instead of hard failure

[chubes4]

Closes #2783

Problem

DailyMemoryTask logged a legitimate "MEMORY.md unchanged" outcome as an ERROR-level hard job failure, generating recurring false-positive noise that polluted error-rate metrics and the wake briefing.

Real log evidence from extrachill.com (and wire.extrachill.com), repeating ~3×/day across multiple agents and every day:

Task failed (job #5151): Daily memory completion policy was not satisfied. MEMORY.md unchanged.

{"job_id":5151,"task_type":"daily_memory_generation","error":"Daily memory completion policy was not satisfied. MEMORY.md unchanged."}

Root cause (investigated, not assumed)

Inspecting live failing jobs confirmed the mechanism:

• The job that "failed" (e.g. job #5151, agent_id 17) runs for ~5s with empty token_usage and the error is the exact fallback string — meaning $response['error'] was empty.
• That fallback only fires when the conversation loop returns completed=false with no genuine error — i.e. the completion policy never returned complete() within the turn budget.
• The affected agent's MEMORY.md is 144 bytes (MAX is 8192), so the file is small and already healthy. The task didn't skip at the size-threshold guard only because there was some activity context that day. The model reviewed the day, found nothing memory-worthy, and never emitted an acceptable ===PERSISTENT=== / ===ARCHIVED=== partition. Nothing is written to disk on this path, so MEMORY.md is genuinely unchanged — a successful no-op, not a fault.

The conversation loop (datamachine_run_conversation) sets completed=false for both genuine faults (provider error, runtime exception, malformed result, budget_exceeded/interrupted/failed status) and this benign "ran out of turns without an acceptable changed split" case. The old code blindly failJob'd both.

Fix

At the completed=false branch in DailyMemoryTask::executeTask(), distinguish the two by explicit error signal:

$genuine_failure = '' !== $response_error
    || ! empty( $response['error_code'] )
    || in_array( (string) ( $response['status'] ?? '' ), array( 'error', 'failed', 'interrupted' ), true );

• Genuine failure (any of the above) → unchanged behavior: log at error and failJob.
• No-op (no error signal; the model simply produced no acceptable changed split) → completeJob with skipped/no_change markers and log at info. Safe because replace_all() happens later in the method, so MEMORY.md is untouched at this point.

How legitimate no-op is distinguished from genuine failure

Outcome	Signal	Behavior
Provider/runtime error	non-empty error / error_code	failJob (error)
Hard loop failure / interruption	status ∈ {error, failed, interrupted}	failJob (error)
Empty model output	later empty($ai_output) guard	failJob (error) — unchanged
Lossy/duplicative split	planMemoryCompaction conservation checks	failJob (error) — unchanged
Policy unsatisfied, file untouched, no error	none of the above	completeJob no-op (info) ✅

All genuine-failure paths downstream (empty response, parse failure, conservation/expansion failures in planMemoryCompaction) are reached only when completed=true and remain loud — they represent a model that produced bad output, which the issue explicitly says must still fail.

Fork decisions / out of scope

• Chose to gate on the loop's existing error signals rather than add a new "no-op" state to the Agents API completion-decision substrate. The substrate decision is intentionally binary (complete()/incomplete()); the no-op semantics are a Data-Machine-task concern (the file being untouched at this call site), so the distinction belongs in the task, not the generic loop. This keeps layer purity intact.
• Did not change the prompt/completion-policy contract to let the model emit an explicit "no change" sentinel. That would be a larger behavioral change; the minimal, evidence-backed fix is to stop treating the existing untouched-file outcome as an error. A future enhancement could add a first-class "decline" signal, but it's not needed to resolve the noise.

Verification

• php -l clean.
• phpcs --standard=WordPress clean (exit 0, no warnings — covers the array-arrow/assignment-alignment gate).
• phpcbf made no changes.

[homeboy-ci]

Homeboy Results — data-machine

Lint

❌ lint — failed

ℹ️ Auto-fix: homeboy lint data-machine --path /home/runner/work/data-machine/data-machine --changed-since 8a413e6 --fix (or homeboy refactor data-machine --path /home/runner/work/data-machine/data-machine --changed-since 8a413e6 --from lint --write)
ℹ️ Some issues may require manual fixes
ℹ️ Full options: homeboy docs commands/lint
Deep dive: homeboy lint data-machine --changed-since 8a413e6

Artifacts and drill-down

• CI results artifact: homeboy-ci-results-data-machine-lint-quality-Linux-node24 contains immediate command JSON for this action invocation.
• Observation artifact: homeboy-observations-data-machine-lint-quality-Linux-node24 contains exported Homeboy run history for deeper queries.
• Drill-down: download the observation artifact, then run homeboy runs import <dir>, homeboy runs list, and homeboy runs findings <run-id>.
• Artifacts are attached to the workflow run: https://github.com/Extra-Chill/data-machine/actions/runs/28076247259

Test

✅ test — passed

ℹ️ No impacted tests found for --changed-since 8a413e6
ℹ️ Run full suite if needed: homeboy test data-machine
Deep dive: homeboy test data-machine --changed-since 8a413e6

Artifacts and drill-down

• CI results artifact: homeboy-ci-results-data-machine-test-quality-Linux-node24 contains immediate command JSON for this action invocation.
• Observation artifact: homeboy-observations-data-machine-test-quality-Linux-node24 contains exported Homeboy run history for deeper queries.
• Drill-down: download the observation artifact, then run homeboy runs import <dir>, homeboy runs list, and homeboy runs findings <run-id>.
• Artifacts are attached to the workflow run: https://github.com/Extra-Chill/data-machine/actions/runs/28076247259

Audit

✅ audit — passed

• audit — 28 finding(s)
• Total: 28 finding(s)

Deep dive: homeboy audit data-machine --changed-since 8a413e6

Artifacts and drill-down

• CI results artifact: homeboy-ci-results-data-machine-audit-quality-Linux-node24 contains immediate command JSON for this action invocation.
• Observation artifact: homeboy-observations-data-machine-audit-quality-Linux-node24 contains exported Homeboy run history for deeper queries.
• Drill-down: download the observation artifact, then run homeboy runs import <dir>, homeboy runs list, and homeboy runs findings <run-id>.
• Artifacts are attached to the workflow run: https://github.com/Extra-Chill/data-machine/actions/runs/28076247259

Tooling versions

• Homeboy CLI: homeboy 0.259.0+b3d82bf59679+451de638
• Extension: wordpress from https://github.com/Extra-Chill/homeboy-extensions
• Extension revision: 94ff2c48
• Action: unknown@unknown