production-data-patch

name	production-data-patch
description	Safely update data on the production Your Project database via Fly SSH. Use when fixing tool metadata, pushing autopsy data, or running DB queries on prod.
metadata	{"version":"2.0.0","author":"Master Agent","category":"operations","updated":"2026-03-31T00:00:00.000Z"}

Production Data Patch — Backend Department

You are responsible for safely modifying production data on Your Project's SQLite database via Fly SSH.

Before Starting

Check:

• Is Fly SSH tunnel available? ~/.fly/bin/fly ssh console -a your-project -C 'echo ok'
• What exactly needs changing? (Get specific slugs, values, columns from Master)
• Is this a single fix or a bulk operation?
• Has Master/S&QA approved this change?

How This Skill Works

Mode 1: Single Tool Fix

Update one or a few specific tools — install_command, description, name, etc.

Mode 2: Bulk Data Push

Push autopsy data (migration_paths, verified_combos) from local to production.

Mode 3: Query & Report

Read-only queries to check analytics, find targets, or audit data quality.

Access Pattern

~/.fly/bin/fly ssh console -a your-project -C 'python3 -c "
import sqlite3
conn = sqlite3.connect(\"/data/your-project.db\")
conn.execute(\"PRAGMA journal_mode=WAL\")
# ... queries here ...
conn.commit()
conn.close()
"'

CRITICAL: aiosqlite Row Access

ALWAYS use column name access, NEVER integer indexing:

# BAD — causes silent bugs
row[0], row[1], row[2]

# GOOD — explicit column names
row["slug"], row["name"], row["count"]

This has caused production bugs TWICE. Always use SELECT ... as alias and row["alias"].

Common Operations

Update a tool

UPDATE tools SET install_command = 'npm install X' WHERE slug = 'tool-slug';

Verify after update

SELECT slug, install_command FROM tools WHERE slug = 'tool-slug';

Bulk push (autopsy data)

# Generate SQL locally, pipe via SSH
cat /tmp/data.sql | ~/.fly/bin/fly ssh console -a your-project -C 'python3 -c "
import sys, sqlite3
conn = sqlite3.connect(\"/data/your-project.db\")
conn.execute(\"PRAGMA journal_mode=WAL\")
sql = sys.stdin.read()
stmts = [s.strip() for s in sql.split(\";\") if s.strip()]
for s in stmts:
    try: conn.execute(s)
    except: pass
conn.commit()
conn.close()
"'

Find unclaimed high-star tools

SELECT slug, name, github_stars FROM tools
WHERE status='approved' AND maker_id IS NULL AND github_stars > 1000
ORDER BY github_stars DESC LIMIT 20;

Proactive Triggers

• Master says "fix install command" → Always verify the CORRECT command first (Cal.com had mailhog!)
• Bulk push requested → Use INSERT OR IGNORE for idempotency
• SSH tunnel down → Report to Master, retry in 5-10 minutes, don't block on it
• Em dash or curly quotes in data → Use straight quotes and -- in inline Python (avoids SyntaxError)

Output Artifacts

Request	Deliverable
"Fix tool X"	Confirmation with before/after values
"Push autopsy data"	Row counts: N migrations, N combos pushed
"Find outreach targets"	slug, name, stars, email list
"Check analytics"	Claim token counts, page views, signup counts

Gotchas (2026-03-30)

• Cal.com had mailhog install command — ALWAYS verify correctness before updating
• SuperTokens had "docker pull stats" — nonsense command that slipped through auto-generation
• Shell quoting is painful for inline Python via SSH — for complex queries, pipe a script file
• Fly SSH tunnel can go down for 15+ minutes — site stays up, just can't SSH
• PRAGMA journal_mode=WAL should be set before bulk writes