Content Write

/content-write — The Content Director

Write a piece of content using the most recent brief from /content-brief. Loads the appropriate framework, persona, and learned patterns. Automatically runs the quality gate on completion.

Preamble

source "$(dirname "$0")/../lib/preamble.sh"

The Skill

Step 1: Find the Most Recent Brief

Look for the most recent .json file in ~/.kai-marketing/briefs/. Read it and parse the JSON.

If no brief exists, tell the user: "No brief found. Run /content-brief {format} {site} \"{keyword}\" first."

If the brief is older than 7 days, warn: "Brief is {N} days old. Consider re-running /content-brief for fresh data."

Display a summary of the brief being used:

Using brief: {filename}
  Format:  {format}
  Site:    {site}
  Keyword: {keyword}
  Persona: {persona}
  Angle:   {angle}

Step 2: Load Framework Context

Based on the brief's format field, read the corresponding framework files from the knowledge base:

Also read:

• The persona file from knowledge/personas/{persona-slug}.md
• The skill contract from harness/skill-contracts/{format-slug}.yaml
• Learned defaults from ~/.kai-marketing/marketing-defaults.md (if exists)
• Voice profile from ~/.kai-marketing/voice.md (if exists)

Step 3: Write the Content

Using all the loaded context (framework rules, persona hooks, brief fields, learned patterns, voice profile, skill contract constraints), write the content piece.

Follow these rules strictly:

• Word count: Match the skill contract's word_count field (typically 1200-1800 for blog, shorter for ads/email)
• Algorithmic Authorship rules (for SEO content): conditions after main clause, instructions start with verbs, sentences under 20 words, bold the answer not the query
• Persona voice: Use the hooks, pain points, and language patterns from the persona file
• Proof points: Incorporate the brief's proof_available field as evidence
• Internal links: Include the brief's internal_links naturally
• CTA: End with the brief's cta

Step 4: Save the Draft

Save the written content to ~/.kai-marketing/drafts/{date}-{slug}.md using the Write tool.

Step 5: Auto-Run Quality Gate

Run the quality gate on the draft:

kai-gate score ~/.kai-marketing/drafts/{date}-{slug}.md --format json

Display the scorecard to the user:

QUALITY GATE — {format}
═══════════════════════
Score: {score}/100 ({grade})

  Algorithmic Authorship: {aa_score}%
  GEO/AEO Signals:       {geo_score}%
  Content Structure:      {cs_score}%
  Four U's:              {four_us}/16

Violations: {count}
{top 3 violations with fix suggestions}

If the gate passes (score >= threshold per skill contract):

• Tell user: "Draft passed quality gate. Run /content-gate for full gate proposal, or publish directly."

If the gate fails:

• Show the specific failures
• Offer to revise: "Gate failed on {N} rules. Want me to fix the specific violations? (Only failing rules will be revised — passing sections are protected.)"
• If user says yes, revise ONLY the failing dimensions. Max 2 revision attempts.

Step 6: Log to Content Chain

After gate pass, log to the CANONICAL content log (data/content_log.json) — this is the log the learning loop (performance_check, pattern_extract, weekly_report) actually reads. Do NOT append to ~/.kai-marketing/content-log.jsonl (legacy; fold old entries in with python3 -m scripts.content.migrate_legacy_log) and do NOT write ~/.kai-marketing/pending/*.json (orphaned — nothing reads them; canonical pending checks live in data/pending_checks/ and are created automatically).

The draft is not live yet, so log it WITHOUT a url (status becomes approved_unpublished; no 30-day check is scheduled until a real URL exists):

python3 - <<'PY'
from scripts.content.content_log import log_entry, compute_content_hash

body = open("{draft_path}").read()
entry = log_entry(
    url=None,                      # never fabricate a URL
    keyword="{keyword}",
    site="{site}",
    format="{format}",
    title="{title}",
    four_us={score},
    notes="hook_type={hook_type}; draft={draft_path}",
    content_hash=compute_content_hash(body),
    campaign_id={campaign_id_or_None},   # cmp-YYYYMMDD-<slug> if part of a campaign
)
print(entry["id"])
PY

When the piece actually goes live, backfill the REAL url — this flips status to published and schedules the 30-day check in data/pending_checks/ automatically:

python3 -c "from scripts.content.content_log import mark_published; mark_published('{entry_id}', '{real_url}')"

Error Handling

• No brief found: Direct to /content-brief
• Framework file missing: Warn and write without that framework (log the gap)
• LLM failure: Show error, suggest retry
• Gate failure after 2 retries: Escalate — show all remaining failures, suggest human review

Chain State

Reads from: ~/.kai-marketing/briefs/{date}-{slug}.json Writes to: ~/.kai-marketing/drafts/{date}-{slug}.md (scratch), data/content_log.json (canonical log; pending checks auto-created in data/pending_checks/ once a real URL is set) Read by: /content-gate (if user wants full gate proposal), /content-report, the 30-day performance_check cron