test-roundtrip

name	test-roundtrip
description	Validate a publish flow end-to-end. Two modes — dry-run (stop before final Publish) or roundtrip (real publish + delete + verify).

Common rules: docs/OPERATING_RULES.md. Default mode for the /test-roundtrip command is dry-run.

Input

• platforms: list of platform keys (facebook, instagram, x, threads, youtube, tiktok).
• mode: dry-run (default) | roundtrip.
• One of:
  • draft_id: existing draft from reports/drafts/<brand>/<draft_id>.json.
  • plan: { iso_week: "2026-W19", slot_id: "..." } — pull from a weekly plan.
  • url: external URL — caller is responsible for running fetch-reference first; pass the resulting reference and a draft id of <YYYYMMDD-HHMMSS>-test.

Output

For mode: dry-run:

{
  "platforms": [
    { "platform": "facebook", "status": "dry_run_ok" },
    { "platform": "x", "status": "dry_run_failed", "step_failed": "...", "error": "..." }
  ]
}

For mode: roundtrip:

{
  "platforms": [
    {
      "platform": "facebook",
      "status": "success",
      "post_url": "https://...",
      "deleted": true,
      "verified_gone": true
    },
    {
      "platform": "youtube",
      "status": "publish_failed",
      "error": "channel verification required"
    }
  ]
}

Steps

1. Resolve inputs. If draft_id is given, load the draft. If plan is given, read reports/plans/<brand>/<iso_week>/plan.json, find the slot, materialize the same fields a draft would carry. If url is given, expect caller to have done the fetch-reference.

2. For each platform sequentially (the orchestrator owns sequencing; this skill describes the per-platform contract):

   mode: dry-run

   • Run publish-<platform> per its SKILL.md ## Dry-run subsection.
   • Write a marker file: reports/posts/<brand>/<draft_id>-<platform>.dry-run.md:

     # <platform> dry-run — <draft_id>
     - status: dry_run_ok | dry_run_failed
     - step_failed: ...        # only on failure
     - error: ...              # only on failure
     - run_id: <run_id>
     - tested_at: <ISO>
     

   • mcp__playwright__browser_close between platforms.

   mode: roundtrip

   • Run the full publish-<platform> (real publish).
   • Capture post_url. If needs_verify (URL extraction failed) → record and skip the delete step (we can't delete what we can't address). Marker file written; deleted: null, verified_gone: null.
   • If publish failed → record publish_failed, skip delete.
   • On real success:
     • Pass post_url to delete-<platform> as target_url.
     • Capture deleted, verified_gone from the delete skill.
   • Write reports/posts/<brand>/<draft_id>-<platform>.roundtrip.md:

     # <platform> roundtrip — <draft_id>
     - status: success | needs_verify | publish_failed | delete_failed | verify_failed
     - post_url: ...
     - deleted: true | false | null
     - verified_gone: true | false | null
     - run_id: <run_id>
     - tested_at: <ISO>
     - notes: ...
     

   • mcp__playwright__browser_close between platforms.

3. Return the aggregate JSON to the caller.

URL passing

The orchestrator (/test-roundtrip command) holds post_url and feeds it to delete-<platform> — this skill does not call other skills. Skill-to-skill calls are forbidden.

Don'ts

• Do not call publish-* and delete-* from inside this skill — the orchestrator chains them.
• Do not skip mcp__playwright__browser_close between platforms — the next platform will hit a profile lock.
• Do not loop the delete step. One try; on failure surface to the user.
• Do not run mode: roundtrip for a platform whose delete-<platform> skill has not been verified at least once — the orchestrator gates this.
• Do not return success if verified_gone: false in roundtrip mode — that indicates the post is still live; mark verify_failed instead.