| name | stage-terraform-provider-repro |
| description | Stage durable Terraform-side repros for bridged-provider issues. Use when triage selects Terraform behavior as the decisive discriminator, an upstream acceptance-style repro is needed, or the user asks for a Terraform repro. |
Stage Terraform Provider Repro
Stage the sharpest Terraform-side discriminator for the issue.
Prefer durable upstream or repo-native artifacts over disposable local
experiments. When a temporary config is useful as a quick check, treat it as a
probe, not the final repro artifact.
Stay in the repro lane. Do not drift back into broad routing or bridge-theory
work unless the Terraform result directly changes the question.
Prefer triage-provider-issue first unless a prior pass already established
that a Terraform repro is the next best action or the user explicitly invoked
this skill.
Read references/repro-shape.md before editing.
Workflow
- State the exact question Terraform needs to answer.
Examples: Does Terraform reproduce the same read failure? Does the issue
disappear with
-refresh=false? Is this accepted upstream behavior?
- Choose the strongest durable artifact available.
Prefer upstream acceptance tests or the closest durable Terraform repro in
the relevant repo.
- Preserve the exact discriminator.
If refresh semantics matter, stage that explicitly.
- Keep the config minimal, but not so minimal that it erases the behavior.
- If the relevant repro surface lives in another repository, continue there only
when the environment can preserve correct repository and artifact ownership.
Otherwise leave a repository-scoped handoff with the exact artifact required.
- Stage the artifact locally.
- Run only what is safe and available.
- If credentials or approvals are missing, stop with a ready-to-run artifact
and exact commands.
Operating Rules
- Do not substitute a loose temp-dir config when the real next step is an
upstream acceptance-style repro.
- Preserve semantic variants intentionally, including flags like
-refresh=false, import, or read-after-update behavior.
- Treat "Terraform did not reproduce" as a meaningful result only when the
repro actually matches the question being asked.
- If the upstream repo has a testing harness, use it instead of inventing a
new convention.
Deliverable
Leave behind:
- current state
- confidence
- what is settled
- what is not settled
- next best action
- the staged Terraform repro artifact
- the exact command matrix
- the discriminator being tested
- what blocked execution, if anything
- workaround status
