publish-skill— agent skill

You are an autonomous skill publisher. Your job is to take a locally installed skill, add it to the skills-hub-registry GitHub repo, and sync it to the production platform.

Do NOT ask the user questions. Proceed through all phases automatically.

SKILL TO PUBLISH: $ARGUMENTS

=== PRE-FLIGHT ===

Before starting, verify:

• [ ] $ARGUMENTS is provided — this is the skill slug (e.g., "skillify", "my-skill")
• [ ] The skill exists locally at ~/.claude/skills/$ARGUMENTS/SKILL.md
• [ ] ~/git/skills-hub-registry/ exists and is a git repo
• [ ] ~/git/skills-hub/ exists
• [ ] AWS credentials are available: AWS_PROFILE=recipeai AWS_DEFAULT_REGION=us-east-1 aws sts get-caller-identity

If $ARGUMENTS is empty: stop and tell the user "Usage: /publish-skill <skill-slug>" If the skill file doesn't exist: list ~/.claude/skills/ to show available skills, then stop. If AWS creds fail: tell the user to run aws sso login --profile recipeai first.

VALIDATION: All checks must pass before proceeding. FALLBACK: Report exactly which check failed and what to do to fix it.

=== PHASE 1: DETERMINE REGISTRY CATEGORY ===

Read ~/.claude/skills/$ARGUMENTS/SKILL.md and extract the category from frontmatter.

Map the category to the registry directory:

If no match, use meta/ as default.

Target path: ~/git/skills-hub-registry/<category>/$ARGUMENTS/

VALIDATION: Category extracted from frontmatter and target path determined.

=== PHASE 2: COPY TO REGISTRY ===

1. Create the target directory: mkdir -p ~/git/skills-hub-registry/<category>/$ARGUMENTS/
2. Copy all skill files:

• ~/.claude/skills/$ARGUMENTS/SKILL.md → registry (required)
• ~/.claude/skills/$ARGUMENTS/references/ → registry (if exists)
• ~/.claude/skills/$ARGUMENTS/scripts/ → registry (if exists)
• ~/.claude/skills/$ARGUMENTS/assets/ → registry (if exists)
• Do NOT copy LEARNINGS.md — that's local usage history, not part of the published skill

1. Update the SKILL.md frontmatter in the registry copy:

• Ensure platforms: [CLAUDE_CODE] is present (add if missing)
• Ensure version is quoted (e.g., "1.0.0" not 1.0.0)
• Leave all other fields unchanged

VALIDATION: ~/git/skills-hub-registry/<category>/$ARGUMENTS/SKILL.md exists and contains platforms. FALLBACK: If copy fails, check file permissions and retry.

=== PHASE 3: COMMIT AND PUSH TO REGISTRY ===

cd ~/git/skills-hub-registry
git add <category>/$ARGUMENTS/
git commit -m "feat(<category>): add $ARGUMENTS skill v<version>"
git push origin main

VALIDATION: git push exits with code 0. FALLBACK: If push fails, run git pull --rebase origin main then retry push.

=== PHASE 4: SYNC TO PRODUCTION PLATFORM ===

This syncs the registry into the live skills-hub database.

Run the following (all in one shell context so env vars persist):

# 1. Fetch prod DB credentials
DB_PASS=$(AWS_PROFILE=recipeai AWS_DEFAULT_REGION=us-east-1 aws secretsmanager \
  get-secret-value --secret-id "skills-hub/prod/db-password" \
  --query SecretString --output text)

# 2. URL-encode the password safely (avoids shell $ expansion mangling)
DB_PASS_ENCODED=$(RAW_PASS="$DB_PASS" python3 -c \
  "import urllib.parse, os; print(urllib.parse.quote(os.environ['RAW_PASS'], safe=''))")

# 3. Write a temp script so env vars survive subprocess boundaries
cat > /tmp/run-skills-sync.sh << SCRIPT
#!/bin/bash
export DATABASE_URL="postgresql://skillshub_admin:${DB_PASS_ENCODED}@skills-hub-prod.c6tii4e8ymqj.us-east-1.rds.amazonaws.com:5432/skillshub?schema=public"
export NODE_ENV=production
cd ~/git/skills-hub/apps/api
# Pull latest registry so the sync picks up the new skill
cd ~/git/skills-hub && git pull --ff-only 2>/dev/null || true
cd ~/git/skills-hub/apps/api
npx tsx prisma/seed-external.ts --sync
SCRIPT
chmod +x /tmp/run-skills-sync.sh
bash /tmp/run-skills-sync.sh

The sync will print one line per external source showing created/updated/skipped counts. Look for the tinh2 source line — that's the registry with your skills.

VALIDATION: Sync exits 0 and prints summary lines without "failed > 0". FALLBACK: If sync fails with ECONNREFUSED, NODE_ENV wasn't set — re-run the temp script approach. If sync fails with P1010 (access denied), SSL wasn't enabled — confirm NODE_ENV=production is exported. If sync fails with AWS error, run aws sso login --profile recipeai and retry.

=== SELF-REVIEW ===

Score the result (1–5):

• Complete: Did the skill land in registry AND sync to production?
• Robust: Were all pre-flight checks run? Were fallbacks needed?
• Clean: Is the registry SKILL.md properly formatted with platforms field?

If any score < 4: identify the gap, fix it, re-score.

=== OUTPUT SUMMARY ===

## Published: /<skill-name>

**Registry path:** ~/git/skills-hub-registry/<category>/<skill>/SKILL.md
**GitHub commit:** <short SHA>
**Production sync:** <created N / updated N from tinh2 source line>

The skill is now available on skills-hub.ai and installable via:
  /install <skill-name>

STRICT RULES

• Never copy LEARNINGS.md to the registry — it's local history, not public content.
• Always use the temp-script pattern for the sync step (PHASE 4). Direct export before a background command loses the env — the temp script avoids this.
• The sync re-processes ALL skills in the registry, not just the new one. That's expected and harmless.
• If the skill already exists in the registry, overwrite it — this is an update, not a conflict.