Index of every page on this site: llms.txt
All skills
Run it yourself

Verify a social post after publishing it: duplicate guard, live re-read against the source caption, media and link checks, public-visibility check, and a lag re-read before any second delivery. Use after every publish or schedule, and whenever a publish result is uncertain, before retrying — a failed or stuck publish can still land later, and retrying blind creates duplicates.

  • 7 files
  • 898.2 KB
  • Updated 5 days ago
  • GitHub
Use this Skill: https://skilld.dev/gh/jokerhtml/needcraft-skills/post-publish-qa

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈100 tokens for metadata: the name and description. ≈2.5k when used: this file. ≈1.4k more on demand in 3 files.

Description uses 5.0% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈397 characters, or 5.0%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Post-Publish QA

What this skill is

A verification method for social posts, in three parts: check before you publish (right account, no duplicate, latest caption, good media), verify after you publish (live post re-read against the source caption, media, links, visibility), and handle failures without creating duplicates. It works with any publishing tool — a platform's own app or scheduler, an API, or a third-party tool — because every check is done against the live surface, never by trusting the tool's success message.

Who it is for: anyone who publishes or schedules social posts and wants proof the post is live, correct, and not duplicated — instead of assuming the publish call worked.

Prerequisites

  • One caption file per post, as the single source of truth. Fix text there, never by editing live text.
  • Access to read the live account: the platform's app/website, an API read method, or a scheduling tool that shows recent posts. Any read method works; this skill names none, so it stays tool-independent.
  • A way to view the post as a non-owner (see Confirm public visibility).

Check before you publish

  • Open the correct account first. Confirm the account name (and Page name, where the platform uses Pages) before you touch any publish or schedule control. Never publish from a duplicate or personal account when the post belongs to a brand account.
  • Confirm the session is signed in as the correct account before you schedule. Stop and report a blocker when it is signed out.
  • Read the recent posts on that exact account before you publish or schedule anything.
  • Compare the slot caption and media against those recent posts and stop when you find the same caption or media already live.
  • Read the caption file again at publish time and confirm it is the latest fixed version, not an earlier draft.
  • Confirm the media file named in the batch matches the caption and opens without corruption.
  • Write one pre-publish record for the slot (template in examples/pre-publish-record.md) naming the account checked and the duplicate-check result.

Verify after you publish

  • Open the live URL returned by the publish call. If no URL was returned, find the post in the account's recent posts list.
  • Compare the live text with the caption file byte for byte. Fail the post on any hashtag or wording mismatch.
  • Check that images load at full size and that video plays for the full source duration.
  • Check that links appear only where the platform allows them. Never a raw link in a post body where the platform forbids it (for example, a raw URL in a Threads post body — the link belongs in the first reply).
  • Check formatting for literal asterisks, broken unicode, and collapsed line breaks. Fail the post on any corruption.
  • Confirm public visibility through a read that does not rely on the owner session (see below).
  • Check recent posts again for an accidental duplicate. Treat a duplicate as a defect.
  • Verify any product link, self-reply, or first comment on product posts: the exact product URL, in the correct place.
  • Write one verification record per post (template in examples/verification-record.md) with date, platform, status, URL, and problem (or "none").

Confirm public visibility through a non-owner read

An owner-session view can show a post that nobody else can see (draft, scheduled-but-unpublished, restricted, or failed visibility). Confirm visibility one of these ways, in order of preference:

  1. Logged-out view: open the post's public URL in a private/incognito browser window while logged out. If the post loads with its full text and media, it is publicly visible.
  2. Second account: open the post from a different account (a personal account, a colleague's) that does not follow the publishing account. If it loads, it is visible to non-followers.
  3. Ask a second person: send the post URL to someone who does not manage the account and ask them to confirm it loads.
  4. Platform share preview: use the platform's share/copy-link flow and open the copied link logged out (same as method 1).

If none of these is possible for a platform, record "visibility check not possible — owner view only" in the verification record and treat the post as unverified, not as confirmed live.

Handle failures

  • Stop the batch on any defect. Do not continue to the next slot until the defect is resolved or the owner explicitly waives it.
  • Check the live surface first when a publish result is uncertain, and never retry a publish before that check.
  • Treat a failed or stuck write as possibly delivered later, and re-check the live surface before you publish a replacement. A publish call that errors or times out can still land minutes later.
  • Ask the owner for approval before you delete any live post. Never delete automatically.
  • Stop after three genuine verification attempts when a live post cannot be reached, and report the blocker. A genuine verification attempt means: wait at least 60 seconds after the publish call, reload the recent-posts list (or re-open the post URL) from scratch — not a cached view — and check both the direct URL and the account's recent posts. Space attempts at least 5 minutes apart. Three attempts that all fail this way = report the blocker, do not keep retrying.

Re-read after a lag before any second delivery

A "zero results" read is not proof of absence when an earlier run may already have delivered the same post without reporting it.

  • Treat a zero-guard result as unverified, not as proof of absence.
  • For threaded replies: read the nested replies in the thread itself, not only the top-level recent-posts list, before you publish a reply.
  • Wait for the threaded surface to settle and re-read the live thread a second time after a short delay (at least 2 minutes) when the first read shows no matching reply.
  • Check the published log and any delivery record for the slot before you make a second delivery attempt.
  • Stop and report the slot as unverified when the second read still cannot confirm absence, instead of publishing a second copy.
  • Record both reads, with their times and results, in the pre-publish record for the slot.

Alternatives for better results

  • Schedule inside each platform's native scheduler and verify the slot in that platform's scheduled list, rather than trusting a schedule command's success message.
  • Keep one caption file per slot as the single source of truth and fix it there rather than editing live text.
  • Track each destination separately, so recovering one failure cannot duplicate a post that already succeeded on another destination.
  • For image-first platforms, check recent pins/posts before you publish and verify the live pin/post after you publish, rather than trusting the publish result.

Worked example — one full pass

Slot: a tutorial carousel for Instagram. Caption file: tutorial-carousel.txt. Media: 5 slides, 1080x1350.

Pre-publish (14:02). Account checked: brand Instagram account (display name confirmed). Recent posts read: last 10 posts, no matching caption or media. Caption file re-read: latest version, 3 hashtags. Media: all 5 slides open, correct order. Pre-publish record written: duplicate check = clear.

Publish (14:05). Scheduled via the platform's native scheduler for 17:00. Slot confirmed in the platform's scheduled list.

Verify after publish (17:04). Post opened from the scheduled-list link. Live text compared byte for byte with tutorial-carousel.txt: match. All 5 slides load at full size. No raw link in the body (link is in the first comment, as intended). No literal asterisks or broken unicode. Public visibility: opened the post URL in a logged-out private window — loads with full text and all slides. Recent posts re-checked: one copy only. First comment verified: exact product URL present. Verification record written: status PASS, problem none.

What would have failed it: if the live text had shown a hashtag missing from the caption file, the record would read FAIL (text mismatch), the batch would stop, and the fix would be made in the caption file — never by editing the live post.

Templates

  • examples/pre-publish-record.md — blank template + one filled example (the carousel above).
  • examples/verification-record.md — blank template + one filled example (the carousel above).

Troubleshooting

Problem What to do
Publish call errors or times out Do not retry. Read the live surface first (recent posts + direct URL). It may have landed.
Live text differs from the caption file FAIL the post. Fix the caption file, then decide with the owner whether to edit or re-post. Never silently pass a mismatch.
Two copies live Report both URLs to the owner and ask which to keep. Never delete without approval.
Post visible to owner but not logged-out Visibility defect. Check the account's visibility/audience settings and the platform's status; report it, do not re-post.
Cannot reach the live post after 3 genuine attempts Stop. Report the blocker with the three attempt times. Do not keep retrying.
A threaded reply may already exist Re-read the nested thread after a 2-minute lag. If still unconfirmed, report as unverified — do not publish a second copy.

Known limits

  • This skill verifies what is visible on the live surface. It cannot verify platform-internal states (shadow restrictions, reduced distribution) — those show up only in the platform's own insights or support channels.
  • A logged-out visibility check is not possible on every platform (some require login to view any post). In that case the record must say so explicitly; the post is unverified, not confirmed.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 90f40b1. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 5 days ago
Other metadata
metadata
{
  "version": "1.0.0",
  "date": "2026-11-16",
  "author": "Muhammad Saad Ranjha / NeedCraft"
}

README badge

README badge for jokerhtml/needcraft-skills/post-publish-qa