publish

name	publish
description	Run the full Astral Index weekly publishing pipeline with quality gates. Scrapes sources, generates drafts, evaluates quality, and delivers via Buttondown. Use when publishing a newsletter issue.
disable-model-invocation	true
allowed-tools	["Bash","Read","Write","Grep","Glob","Agent","AskUserQuestion"]

/publish — Intelligent Weekly Publishing

You are running the Astral Index weekly publishing pipeline. You execute the full ingest-author-eval-deliver pipeline with quality gates and editorial review.

Arguments

$ARGUMENTS may contain:

• --since N — lookback window in days (default: 7)
• --dry-run — preview mode, no delivery
• --send — include Buttondown delivery at the end
• --no-expand — skip the expand step
• --strategy X — skip strategy comparison, use this strategy directly

Parse these from $ARGUMENTS. Set defaults: SINCE=7, DRY_RUN=false, SEND=false, TODAY=$(date +%Y-%m-%d).

Step 1: Pre-flight

Check that required environment is set up:

# Check for API keys (existence, not values)
test -f .env && grep -q "ANTHROPIC_API_KEY" .env && echo "ANTHROPIC_API_KEY: set" || echo "ANTHROPIC_API_KEY: MISSING"
test -f .env && grep -q "BRAINTRUST_API_KEY" .env && echo "BRAINTRUST_API_KEY: set" || echo "BRAINTRUST_API_KEY: MISSING"
test -f .env && grep -q "BUTTONDOWN_API_KEY" .env && echo "BUTTONDOWN_API_KEY: set" || echo "BUTTONDOWN_API_KEY: MISSING"

Report which keys are present. If --send was requested and BUTTONDOWN_API_KEY is missing, warn that delivery won't work.

Set output paths (staged in git-tracked issues/ directory):

• ISSUE_DIR=issues/${TODAY}
• DRAFT_MD=${ISSUE_DIR}/draft.md
• DRAFT_JSON=${ISSUE_DIR}/draft.json
• EVAL_OUTPUT=data/eval/${TODAY}_results.json

Step 2: Ingest

Run sequentially:

uv run --package astral-ingest astral-ingest scrape
uv run --package astral-ingest astral-ingest expand --since {SINCE} --js
uv run --package astral-ingest astral-ingest classify --since {SINCE}

If --dry-run, add --dry-run to each command. If --no-expand, skip the expand step.

After scraping, check item count:

uv run --package astral-ingest astral-ingest export --since {SINCE} --format json | head -5

Or count items in data/items/ directories from the lookback window.

Quality gate: If fewer than 20 items were ingested, warn the user and ask whether to:

• Continue anyway
• Widen the window (suggest --since {SINCE * 2})
• Abort

Report: total items, category breakdown if available.

Step 3: Strategy Comparison

If --strategy was specified, skip this step and use that strategy.

Otherwise, compare all strategies:

uv run --package astral-eval astral-eval compare baseline headlines-only wide-coverage recency-biased --since {SINCE}

Parse the comparison output. Present a table with scores per strategy. Recommend the highest-scoring strategy with a brief rationale.

Ask the user to confirm the strategy or pick a different one.

Step 4: Draft

Generate the newsletter (defaults to staging in issues/{TODAY}/):

uv run --package astral-author astral-author draft --since {SINCE} --strategy {STRATEGY}

If --dry-run, add --dry-run.

Confirm the draft files were staged:

ls -la {DRAFT_MD} {DRAFT_JSON}

Step 5: Evaluate

Skip if --dry-run.

mkdir -p data/eval
uv run --package astral-eval astral-eval quality --since {SINCE} --draft-file {DRAFT_JSON} --output {EVAL_OUTPUT}

Read the eval results. Present a score table.

Quality gate: If the average score < 0.5 or any individual scorer < 0.3:

• Flag the problematic scorers
• Recommend re-generating with a different strategy
• Ask whether to continue, re-draft, or abort

Step 5.5: Stage for Review

If not --dry-run, the draft is now in issues/{TODAY}/. Tell the user:

Draft staged at issues/{TODAY}/. You can edit draft.md, commit edits, then continue to delivery.

Step 6: Editorial Review

Read the full markdown draft from {DRAFT_MD}.

Check for and report:

• Empty sections (no items)
• Sections with only 1 item (thin sections)
• Deep-dive sections missing prose/summaries
• Near-duplicate titles across sections
• Items without links
• Very short or very long sections (imbalance)
• Broken markdown formatting

Present a bullet-point editorial critique. If issues are found, suggest whether they warrant re-drafting or are acceptable.

Step 7: Deliver

Only proceed if:

• --send was passed OR user explicitly confirms delivery
• Not in --dry-run mode
• Quality gates passed (or user overrode)

# Push draft to Buttondown (reads from issues/{TODAY}/)
uv run --package astral-serve astral-serve draft {TODAY}

After the draft is pushed, ask for explicit confirmation before sending:

The draft has been pushed to Buttondown. Review it in the dashboard, then confirm: should I send it now?

Only on explicit "yes":

uv run --package astral-serve astral-serve send {TODAY}

Report the final status:

uv run --package astral-serve astral-serve status {TODAY}

Important Notes

• This skill has three hard stops requiring user input: strategy selection, quality gate, and send confirmation. Everything else runs autonomously.
• If any step fails, report the error and ask whether to retry, skip, or abort.
• The serve CLI now accepts a date string (e.g., 2026-03-15) and reads from issues/{date}/. It uses draft.md for the email body (respecting human edits) and draft.json for metadata.
• All uv run commands should be run from the project root directory.