Downstream jobs read needs.activation.outputs.target_repo, but activation does not declare it in most workflows #67175

Summary

Jobs downstream of activation (agent, conclusion, safe_outputs) read needs.activation.outputs.target_repo and target_repo_name, for example:

- name: Derive GitHub App owner
  id: github-mcp-app-token-owner
  env:
    GH_AW_TARGET_REPOSITORY: ${{ needs.activation.outputs.target_repo }}

In most compiled workflows the activation job does not declare a target_repo output at all. Its outputs: list is artifact_prefix, aw_context, body, comment_id, comment_repo, daily_ai_credits_*, engine_id, lockdown_check_failed, model, oauth_token_check_failed, secret_verification_result, setup-*, text, title. The reference therefore evaluates to an empty string, and the step fails:

[gh-aw] Unable to derive GitHub App owner from repository:
[gh-aw] Normalized repository candidate:
##[error]Process completed with exit code 1.

Evidence (compiler v0.91.6, unpatched, whole fleet compiled)

Two earlier fix attempts that added these fallbacks, #39742 (June) and #66678 (October, "Avoid GHES masking of cross-job workflow outputs", linked to #38345), were closed without merging, so the compiler output has not changed.

Expected

Either every reference carries || github.repository (and target_repo_name carries || github.event.repository.name), or activation always declares the output.

Workaround today

A post-compile rewrite adds the two fallbacks to every reference in both fleets.

Related

Same area as #38345 (cross-job outputs on GHES), which is still open.