| name | grade-post |
| description | Score any draft post against the grading rubric (hook 50%, voice 15%, length 10%, CTA 10%, platform-native 10%, anti-AI 5%). Returns score + top 3 fixes. Pass threshold 80/100. Read-only, never writes the draft. |
| when_to_use | ["/grade-post","grade this post","score my draft","is this any good","grade post"] |
/grade-post — v0.1.0
Score a draft against the rubric. Returns 0-100 + top 3 fixes. Read-only — never writes the draft.
Be harsh but fair. An 80 is good. A 90 is strong. A 100 doesn't exist. False positives waste more time than honest feedback.
Inputs
- Required: post content — EITHER a file path (e.g.,
outputs/social-media-content/2026-05-15-linkedin-foo.md) OR pasted text
- Read on every run (lazy-load):
cowork-social/skills/grade-post/templates/grading-rubric.md (the rubric — load on every run)about-me/writing-rules.md (anti-AI banned-phrase list)projects/social-media-content/platform-voice.md (load only the section matching the post's platform)projects/social-media-content/brand-brief.md (voice_signature for voice-match check)_shared/hook-patterns.md (referenced by the rubric's hook sub-rubric)
Logic
Step 1 — Resolve the input
- If the user passed a file path → read it.
- If pasted text → use directly.
- Identify the platform:
- From filename pattern (
-linkedin-, -twitter-, -threads-, -bluesky-, -instagram-, -facebook-, -tiktok-)
- From content cues (length, hashtag pattern, line breaks)
- If ambiguous → ASK the user which platform.
Step 2 — Apply the 6-dimension rubric
Walk every dimension of templates/grading-rubric.md verbatim. Do NOT drift the weights.
| # | Dimension | Weight | Source check |
|---|
| 1 | Hook quality | 50% | First 1-2 sentences against _shared/hook-patterns.md sub-rubric (pattern match, specificity, tension, first-3-words test, standalone-able) |
| 2 | Voice match | 15% | Compare against platform-voice.md (platform section) + brand-brief.md voice_signature |
| 3 | Length fit | 10% | Platform sweet spot from rubric table |
| 4 | CTA clarity | 10% | One verb-driven CTA matched to platform's reward metric |
| 5 | Platform-native feel | 10% | Hashtag count + link rules + format + algorithm-fit |
| 6 | Anti-AI compliance | 5% | writing-rules.md banned-phrase list + rubric's universal checks |
Score each dimension per the rubric. Sum to 0-100.
Step 3 — Top 3 fixes (always exactly 3)
Pick the 3 highest-ROI fixes — the changes that gain the most points per word changed. Ranked by point-gain potential.
Hard rule: Top 3 fixes only. Not 5. Not 7. Forcing prioritization is part of the value.
For each fix:
- What's wrong — quote the specific line
- Why it hurts — what's the cost (attention, voice, algorithm penalty, etc.)
- Specific fix — exact rewrite (not "make the hook better" — give the replacement line)
- Estimated lift — points gained if applied
Step 4 — Verdict
| Score | Verdict |
|---|
| 80-100 | PASS — ship it |
| 60-79 | REWORK — apply top 3 fixes + re-grade |
| 0-59 | REGENERATE — too many dimensions failing; regenerate from scratch |
Step 5 — Display the scorecard
Use the rubric's "Scorecard output format" section verbatim:
## Post Grade: XX/100 — [PASS / REWORK / REGENERATE]
### Score Breakdown
| Dimension | Weight | Score | Note |
|---|---|---|---|
| Hook quality | 50% | XX/50 | [1-line note if under 40] |
| Voice match | 15% | XX/15 | [...] |
| Length fit | 10% | XX/10 | [...] |
| CTA clarity | 10% | XX/10 | [...] |
| Platform-native feel | 10% | XX/10 | [...] |
| Anti-AI compliance | 5% | XX/5 | [...] |
### Anti-AI Audit
| Check | Pass/Fail | Violation |
|---|---|---|
| Em dashes | ... | ... |
| Contractions | ... | ... |
| Numbers as digits | ... | ... |
| Active voice | ... | ... |
| Filler words | ... | ... |
| Filler openers | ... | ... |
### Top 3 Fixes (ranked by impact)
**1. [Issue title] [+X pts]**
- Current: "[exact quote]"
- Why it hurts: [...]
- Fix: [specific rewrite]
**2. [Issue title] [+X pts]**
- Current: "[exact quote]"
- Why it hurts: [...]
- Fix: [specific rewrite]
**3. [Issue title] [+X pts]**
- Current: "[exact quote]"
- Why it hurts: [...]
- Fix: [specific rewrite]
Hard rules
- READ-ONLY — NEVER modify the draft being graded. This is a grader, not a writer.
- Use the rubric VERBATIM — don't drift weights. Hook is 50%. Period.
- Top 3 fixes ONLY — not 5, not 7. Forcing prioritization is part of the value.
- Top 3 fixes must be ACTIONABLE — specific replacement text, not "consider strengthening"
- Don't pad scores — if the hook is a 25, say 25. A false 85 wastes more time than an honest 60.
- Don't rewrite the post — fix instructions only.
- 3rd-4th grade reading level in user-facing prompts.
- Plan-then-approve only applies to the Foundation B memory.md append (not to the grade output itself — the grade displays immediately).
Voice
Direct, scored, no flattery. The grader's job is to find the weak link, not encourage. If a hook is a 25, the grader says 25 — and tells the user exactly how to push it to 40+.
Self-improvement close (Foundation B)
See _shared/foundations.md → Foundation B. After delivering this skill's main output + the ⚡ NEXT MOVE block, ask the user:
"What would've made this 10% better?"
Then:
-
Append to projects/social-media-content/memory.md:
<YYYY-MM-DD> | /grade-post | <answer verbatim>
-
Read memory.md and check if any pattern recurs 3+ times for /grade-post. Match by:
- Substring overlap ≥ 60% with prior entries
- Same keyword (e.g., "hook", "voice", "fix", "platform", "weight", "rubric")
-
If recurrence detected → surface: "I've seen this 3+ times. Want me to update /grade-post itself?" → on yes, draft change to projects/social-media-content/skill-improvements.md.
-
If no recurrence → silent.
Actionable close (Foundation C)
See _shared/foundations.md → Foundation C. Skill output MUST end with a ⚡ NEXT MOVE block matching the validation regex:
⚡ NEXT MOVE: .+ .+ .+\n Why: .+
The Next Move is verdict-aware. Pick from these three templates:
If PASS (≥80)
⚡ NEXT MOVE: Schedule this <platform> draft via Blotato within the next hour.
Why: Score is above your usual ceiling — momentum compounds when you ship on a high.
If REWORK (60-79)
⚡ NEXT MOVE: Rewrite the hook using the '<pattern>' pattern from hook-patterns.md before scheduling.
Why: Hook is <X>/50 — pulling it to ≥40 lifts the total score most efficiently.
(Substitute the specific hook pattern that would best fit this post's angle, picked from _shared/hook-patterns.md: Receipts, Reframe, Reverse, Stolen Lesson, Vulnerable Story, Specific Number, Contrarian Take, Question, Pain Point List, Behind-the-Scenes, Parallel Contrast.)
If REGENERATE (<60)
⚡ NEXT MOVE: Re-run /draft-<platform> on this topic with a tighter angle today.
Why: Multiple dimensions failing — faster to regenerate than patch.
Validation
The block MUST match: ⚡ NEXT MOVE: .+ .+ .+\n Why: .+
If it doesn't match, the output is incomplete — regenerate the close.
What NOT to do
- Don't rewrite the post (you're a grader, not a writer)
- Don't grade leniently (a false 85 wastes more time than an honest 60)
- Don't return more than 3 fixes (forcing prioritization is the value)
- Don't skip "why it hurts" (that's where the user learns)
- Don't flag style preferences as errors (grade against the rubric, not personal taste)
- Don't pad scores
