Baseline
5a306f8956cb1eeae69f9709de0e4d61b44e11e7 (v1.3.1)
Reproduction
- Check out the baseline commit.
- Compare the deployment workflow in
skills/google-agents-cli-deploy/SKILL.md with the Agent Runtime order in skills/google-agents-cli-observability/SKILL.md.
- Also inspect the README and Starter Pack migration entry points for an authoritative ordering rule.
Actual behavior
The deploy skill says not to run infra single-project before deployment and describes it as optional. The observability skill says Agent Runtime users must run it before the first deploy. README and the migration guide do not resolve the target/ownership distinction. A user following different entry points can therefore receive opposite instructions.
Issue #2 reports a related downstream symptom: applying Terraform after an imperative Agent Runtime deployment can create or target a second Runtime. This issue tracks the separate documentation root cause across the four user entry points.
Expected behavior
All four entry points should use one decision table:
- Basic Agent Runtime / Cloud Run: deploy directly.
- GKE: deploy performs the required targeted Terraform.
- Terraform-managed observability, any target: run
agents-cli infra single-project --apply before deploy.
- Existing imperative deployment: do not apply Terraform afterward without importing or deleting before switching; otherwise keep imperative ownership and configure observability manually.
Minimal fix
Add the same compact decision table to README, the migration guide, deploy skill, and observability skill; remove the conflicting statements; add a text contract test that requires the table and rejects the old contradictory wording.
Tested fix:
Verification
- Documentation contract tests: 7 passed
- Full test suite: 99 passed
ruff check src tests: passedty check src: passeduv build: passed- Built wheel installed normally;
agents-cli --version returned 1.3.1 on Python 3.11 and 3.13
I am not opening a PR because the upstream repository currently does not accept PRs.
Baseline
5a306f8956cb1eeae69f9709de0e4d61b44e11e7(v1.3.1)Reproduction
skills/google-agents-cli-deploy/SKILL.mdwith the Agent Runtime order inskills/google-agents-cli-observability/SKILL.md.Actual behavior
The deploy skill says not to run
infra single-projectbefore deployment and describes it as optional. The observability skill says Agent Runtime users must run it before the first deploy. README and the migration guide do not resolve the target/ownership distinction. A user following different entry points can therefore receive opposite instructions.Issue #2 reports a related downstream symptom: applying Terraform after an imperative Agent Runtime deployment can create or target a second Runtime. This issue tracks the separate documentation root cause across the four user entry points.
Expected behavior
All four entry points should use one decision table:
agents-cli infra single-project --applybefore deploy.Minimal fix
Add the same compact decision table to README, the migration guide, deploy skill, and observability skill; remove the conflicting statements; add a text contract test that requires the table and rejects the old contradictory wording.
Tested fix:
Verification
ruff check src tests: passedty check src: passeduv build: passedagents-cli --versionreturned1.3.1on Python 3.11 and 3.13I am not opening a PR because the upstream repository currently does not accept PRs.