agentleFS
priyanshu-aryapriyanshu-arya/LinkedIn_Automation/.claude/skills/critique-draft/SKILL.md

critique-draft

Skill0 starsChanged 2 months agoMITView on GitHub

Use when the user asks to review/critique a draft's quality, check a draft for accuracy or duplication, compute its viral potential score, or explicitly invokes /critique-draft. Runs the Quality/Critic Agent on a status:draft Draft Note, auto-revises it once if it scores low, and hands it off to the approval workflow by moving it to status:in_review.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the averaged score is below 6, make exactly one revision pass targeting the specific weak factors (e.g.

    line 72
  2. Very specific:

    Add a ## Critic Notes section to the body: the per-factor scores, any duplicate/fatigue findings from step 2, and (if revised) what changed.

    line 86
  3. Very specific:

    After a run that moves one or more drafts to in_review, send a single batched PushNotification (status: proactive) — e.g.

    line 102
  4. Distinctive:

    Never invent a "historical performance" trend when no analytics data exists — score that factor neutral and say why.

    line 109
  5. Distinctive:

    Never revise more than once per critique run.

    line 111
  6. Very specific:

    Never move a draft straight to approved or scheduled — only in_review.

    line 112
  7. Very specific:

    Never advance a draft flagged as a genuine duplicate (2a) to in_review — a known duplicate must never reach human review as if it were …

    line 114
  8. Distinctive:

    Never soften an accuracy fix by leaving the unsupported claim in place "just flagged" — fix it or remove it.

    line 116

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Critique Draft (Quality/Critic Agent — Phase 5)

Checks a draft for accuracy, originality, and overall strength, computes a Viral Potential Score, auto-revises once if the score is weak, and moves the draft to status: in_review — ready for human approval (Phase 7). This skill never approves, schedules, or publishes anything itself.

Arguments

  • draft-id (optional) — critique one specific draft. If omitted, run against every Draft Note currently status: draft (i.e. not yet critiqued).

Process (per draft)

1. Re-verify accuracy

Re-read the draft's sources[] research notes. Confirm every factual claim in the post text still traces to something in those notes' Summary/Key Findings. If drift is found (a claim not actually supported), fix it now — soften or remove the claim, don't just flag it and move on.

2. Originality / duplicate check (expanded, Phase 13)

First pass: scan Content-Learnings/content-index.md (fast, compact) across seven dimensions — topic, hook, storytelling structure, examples, conclusion type, image concept, and category/format — for anything close to this draft. For anything the index flags, read the actual flagged note in full (Published-Posts/, other Drafts/, Scheduled/) to confirm before acting on it. Also check Content-Learnings/playbook.md's "Topic Fatigue Watch" section. If content-index.md doesn't exist yet or is too thin to trust, fall back to reading the folders directly, as before.

Draw a hard line between two severities:

  • Genuine duplicate — the same underlying story/news item, example, hook, or conclusion as an existing note from the same or a recent week. This is not a scoring matter — see step 3a below, it skips normal scoring entirely.
  • Merely similar but substantively distinct — same general topic or style, different angle/argument/example. Note it in the critique and let it pull the Originality sub-score down in step 3, same as before; still goes to human review, the human approver makes the final call.

2a. If a genuine duplicate was found, stop here — don't score or revise

A genuine duplicate isn't fixable by rewording — it needs a different underlying angle entirely. Skip steps 3-4 (scoring, auto-revise) for this draft. Keep status: draft (do not advance to in_review, since advancing would surface a known duplicate to human review, which defeats the point of catching it here). Append {action: duplicate_flagged, date, note: "<what it duplicates and how>"} to history, and say so plainly and directly in the report — this draft's idea needs to be replaced with a genuinely different angle before it can be redrafted, not merely revised.

3. Score Viral Potential (0-10)

Average these factors, each 0-10:

FactorHow to assess it
Hook strengthSpecific and attention-grabbing vs. generic
Topic freshnessInherited from the source research note's freshness_score
Audience relevanceFit with the AI/dev/career LinkedIn audience
Educational valueInherited from the source research note's educational_value, adjusted for how well the draft actually delivers it
Shareability / discussion potentialWould someone repost this or comment with a real opinion?
OriginalityInverse of step 2's findings
CredibilityInherited from the source research note's authority_score
Historical performance of similar postsCheck Content-Learnings/playbook.md for a rule matching this draft's category/format/hook_style with real evidence attached — use it if present. Otherwise score neutral (5) and say plainly that no data exists yet. Do not fabricate a trend either way.

4. Auto-revise once if the score is weak

If the averaged score is below 6, make exactly one revision pass targeting the specific weak factors (e.g. if hook strength is low, rewrite only the hook; if originality is low, shift the angle) — re-reading Content-Learnings/voice-guide.md so the revision doesn't drift from voice. Recompute the score after revising. Do not loop more than once — if it's still below 6 after one revision, say so plainly in the report rather than repeatedly regenerating.

5. Update the Draft Note

(Skip this step entirely if 2a fired — that path already updated the note and stopped.)

  • Set viral_score to the final computed value.
  • Append a history entry: {action: regenerated, date, note} if step 4 fired, otherwise {action: critiqued, date, note}.
  • Add a ## Critic Notes section to the body: the per-factor scores, any duplicate/fatigue findings from step 2, and (if revised) what changed.
  • Set status: in_review.

Report back

Per draft: either "HARD DUPLICATE — needs a new angle" (if 2a fired, name what it duplicates) or the final viral_score, per-factor breakdown, whether it was auto-revised, and any similar-but-distinct flags. Make clear a draft that reached in_review is now waiting for human approval — nothing has been approved, scheduled, or published. A draft left at status: draft with a duplicate flag is not waiting for anything — it needs a replacement idea before this skill can be run on it again.

Notifications (REQUIREMENTS.md §24)

After a run that moves one or more drafts to in_review, send a single batched PushNotification (status: proactive) — e.g. "3 LinkedIn drafts ready for your review." Batch across the whole run; don't send one per draft.

Hard rules

  • Never invent a "historical performance" trend when no analytics data exists — score that factor neutral and say why.
  • Never revise more than once per critique run.
  • Never move a draft straight to approved or scheduled — only in_review. Approval is a human action (Phase 7).
  • Never advance a draft flagged as a genuine duplicate (2a) to in_review — a known duplicate must never reach human review as if it were novel.
  • Never soften an accuracy fix by leaving the unsupported claim in place "just flagged" — fix it or remove it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.