draft-score

Lightweight ContentShake AI self-check the /draft stage can call before saving. Returns just SEO + Quality scores (no full optimization) so the writer knows whether the draft is in winning territory b

SKILL.md

GitHub

Draft Score Skill (ContentShake self-check)

Run a lightweight ContentShake AI score on the current draft so the writer / orchestrator knows whether the draft is already close to winning territory. Cheaper than the full /optimize-content pass — uses the score action of contentshake_optimize.py which doesn't return the term/topic recommendation lists.

This skill is optional in the pipeline. If the draft is already strong, it lets the orchestrator skip directly to /quality-check → /verify-claims and save a full /optimize-content pass for a future revision. If the draft scores poorly, it's an early warning that /optimize-content will need its full iteration budget.

Input

/draft-score {slug}

Reads:

  • content-pipeline/5-drafts/{slug}.md (or 6-drafts-cited/{slug}.md if already cited)
  • content-pipeline/1-research/{slug}.md (target keyword)

Process

Step 1 — Preflight

  1. Confirm the draft exists. If not, exit with a clear error.
  2. Read the target keyword from the research dossier's frontmatter (or fall back to the slug with hyphens replaced by spaces).
  3. Check SEMRUSH_API_KEY_CONTENTSHAKE then SEMRUSH_API_KEY_BLOG_AGENT in env.

Step 2 — Soft-fail when key is missing

If neither key is set:

  1. Print to stdout:
    draft-score: SEMRUSH_API_KEY not set in env. Skipping ContentShake self-check.
                 Load via `doppler run -- claude` or export the key. Pipeline continues.
    
  2. Write a stub at content-pipeline/optimization/{slug}-draft-score.json:
    {
      "skipped": true,
      "reason": "SEMRUSH_API_KEY not set",
      "_meta": {"slug": "{slug}", "action": "score"}
    }
    
  3. Exit 0. Same convention /research uses for missing OpenRouter — never block the pipeline on a missing optional key.

Step 3 — Call the script with --action=score

doppler run -- python .claude/skills/optimize-content/scripts/contentshake_optimize.py \
  --slug {slug} --keyword "{keyword}" --action score

The score action returns just seo_score and quality_score (plus _meta) — no term lists, no missing-topic analysis, no competitor coverage. Cheaper than optimize and counts as 1 unit against the same BLOG_AGENT_CONTENTSHAKE_MONTHLY_CAP budget tracked by /optimize-content.

Step 4 — Persist the result

Save the script's stdout to content-pipeline/optimization/{slug}-draft-score.json. Increment the budget counter at content-pipeline/optimization/api-budget.md (shared with /optimize-content).

Step 5 — Print the verdict

draft-score: SEO 7.2  Quality 8.1   verdict=NEEDS_OPTIMIZE
  → /optimize-content {slug} likely needed before publish

Verdict heuristic:

Condition Verdict Implication for orchestrator
seo_score ≥ 8 AND quality_score ≥ 8 WIN_LIKELY /optimize-content may be skippable — quality-check first, then decide
seo_score ≥ 7 AND quality_score ≥ 7 BORDERLINE Run /optimize-content but expect 1–2 iterations to clear
Otherwise NEEDS_OPTIMIZE /optimize-content should run with full iteration budget

Output

  • content-pipeline/optimization/{slug}-draft-score.json — the raw scoring response (or stub if skipped)
  • One-line verdict to stdout

Quality checklist

  • Draft file exists and was readable
  • Target keyword resolved from research dossier or slug
  • Soft-fail path runs cleanly when SEMRUSH_API_KEY is unset (exits 0, writes stub)
  • Result JSON persisted to optimization/{slug}-draft-score.json
  • Verdict printed with the recommended next action

When to call this skill

This skill is designed to be called from the /draft stage after the draft is written but before /quality-check — a "is this close to ready?" signal. It's also fine to call standalone:

  • During iterative revisions, to see if a manual edit moved the score
  • Before deciding whether to schedule /optimize-content (saves a full optimize pass when the draft is already strong)
  • After /update-draft to see if the refresh moved scores in the right direction

When NOT to call this skill

  • When /optimize-content is about to run anyway — the full optimize pass produces scores AND recommendations, so the lightweight self-check is redundant and wastes a budget unit
  • When the budget is at 100% — the script will exit cleanly but the call still gets recorded as attempted

Why not just call /optimize-content

/optimize-content is a 5-iteration loop that mutates the draft. This skill is a read-only scoreback. Call this when you want to know "where do I stand?" without committing to the full optimization. Two distinct jobs.