[self-care:open-source-failures] Fix recurring safe-output processing failures in gh-aw-cao workflows

githubnext/gh-aw-cao has at least 9 bounded-snapshot failures across `self-care-data-acquisition-audit`, `software-development-practices-github-well-architected`, `self-care-dashboard-performance`, and `pr-sous-chef`, all reaching the `safe_outputs` job and then failing in `Process Safe Outputs`. The recurring post-agent failure is actionable, but the bounded snapshot does not expose the underlying processor error, so the immediate need is to inspect the safe-output handoff path for these workflows and restore successful write processing.

**Action:** CAO maintainers should inspect the `safe_outputs` job implementation and payload contract for the affected workflows, then fix the shared failure path; accept when a rerun of one affected workflow completes the `Process Safe Outputs` step successfully.

Agent prompt
Investigate recurring `Process Safe Outputs` failures in `githubnext/gh-aw-cao` for `self-care-data-acquisition-audit`, `software-development-practices-github-well-architected`, `self-care-dashboard-performance`, and `pr-sous-chef`. Trace the shared `safe_outputs` job and any common output-processing code or workflow imports those workflows use. Identify the smallest safe fix that restores safe-output processing without weakening validation or fail-closed behavior. If the issue is contract drift between workers and the processor, update the shared contract and the affected workflow sources, then regenerate derived artifacts instead of editing lock files directly.

Critical findings:
- Severity: P1
- Normalized signature: `Process Safe Outputs`
- Affected repository: `githubnext/gh-aw-cao`
- Affected workflows: `self-care-data-acquisition-audit`, `software-development-practices-github-well-architected`, `self-care-dashboard-performance`, `pr-sous-chef`
- Representative run: https://github.com/githubnext/gh-aw-cao/actions/runs/33963457581
- Additional corroborating runs: https://github.com/githubnext/gh-aw-cao/actions/runs/33963403527, https://github.com/githubnext/gh-aw-cao/actions/runs/33963453396, https://github.com/githubnext/gh-aw-cao/actions/runs/33963505601
- Evidence-supported cause: a shared safe-output processing path is failing after agent completion; the bounded snapshot does not reveal the precise processor error.
- Confidence: medium

Bounded evidence

Observed run counts in the snapshot:
- `.github/workflows/self-care-data-acquisition-audit.lock.yml`: 4 enriched failures with `Process Safe Outputs`
- `.github/workflows/software-development-practices-github-well-architected.lock.yml`: 5 enriched failures with `Process Safe Outputs`
- `.github/workflows/self-care-dashboard-performance.lock.yml`: 1 enriched failure with `Process Safe Outputs`
- `.github/workflows/pr-sous-chef.lock.yml`: 1 enriched failure with `Process Safe Outputs`

The evidence window is the fixed seven-day snapshot ending at 2026-09-05T11:59:05.013Z. The snapshot is truncated to the newest 100 failed runs, so additional occurrences may exist outside the bounded sample.

### Control Plane
- Correlation ID: `33965965540-220`
- Central repository: `githubnext/gh-aw-cao`
- Control plane run URL: `https://github.com/githubnext/gh-aw-cao/actions/runs/33965965540`

> Generated by [SelfCare / Open Source Failures](https://github.com/githubnext/gh-aw-cao/actions/runs/33966125093) · pi · gpt54 · 24.7 AIC · ⌖ 8.8 AIC · ⊞ 8.3K · [◷](https://github.com/search?q=repo%3Agithubnext%2Fgh-aw-cao+is%3Aissue+%22gh-aw-workflow-call-id%3A+githubnext%2Fgh-aw-cao%2Fself-care-open-source-failures%22&type=issues)
> - [x] expires on Sep 19, 2026, 12:33 PM UTC

Start with the `safe_outputs` job and `Process Safe Outputs` step in the four named `.github/workflows/*.lock.yml` files, then trace their shared output-processing path and payload contract. Compare the affected workflow sources and generated artifacts without editing lock files directly. Done means a rerun of one affected workflow completes `Process Safe Outputs` successfully while preserving validation and fail-closed behavior.

Written by the indexing model from the issue text.