firestore-rules-creation skill: have your agent write and attack-test Firestore security rules

• Delegates to firestore-rules-author when that subagent is available

  If subagent delegation and the firestore-rules-author subagent both exist, the skill tells the agent not to write the rules itself. Otherwise the agent follows the workflow directly.

  Read · Subagent hand-off

• Scans your code for collections, queries, data models and auth

  Collects every Firestore path, every where, orderBy and limit clause, field types, constraints and CRUD operations. The rules must allow the queries your app really runs.

  Try: “Read this codebase and write Firestore security rules for it.”

  Files · Phase 1: codebase analysis

• Default deny, validator functions, no public reads on first draft

  Starts from deny-all and adds helper functions such as isAuthenticated and isOwner. Every create and update must call a validator, and the first file must contain no allow read: true.

  Write · Phase 2: rules generation

• Keeps personal data out of readable profile documents

  Documents with email or phone can be read only by their owner. For public profiles the skill asks for denormalized public fields or a separate public collection.

  Write · User data separation

• Blocks users from creating or raising their own roles

  Rules must stop self-assigned admin roles and role changes. The skill gives a bad and a good example.

  Write · Role guidelines

• Enums, state transitions, path scoping, safe counters

  Covers allowed values, legal status changes, scoped paths and counter updates verified with getAfter().

  Write · Business-logic checks

• 21 attack checks run on the agent's own rules

  The agent tries things like ownership hijacking, schema pollution, 1MB strings and mixed-content leaks, then fixes the rules and repeats until no attack works.

  Read · Phase 3: devil's advocate

• Checks rules with firebase deploy --only firestore:rules --dry-run

  SKILL.md says to validate with that command or a similar tool before the final file is output. Then the agent creates or updates firestore.rules.

  Runs code · Phase 4: syntax validation

Watch out for

• Generated rules are a first draft

  SKILL.md tells the agent to call them a prototype and to tell users to review and harden them before launching to a large audience. A rule gap can expose user data.

• It writes your security rules file

  The skill creates or overwrites firestore.rules. Use a git branch and read the diff before you deploy anything.

• It asks for a Firebase CLI command

  The syntax check uses firebase deploy with --dry-run, or a similar tool. Check which project the CLI points at before approving it.

• The subagent definition can run commands

  agents/firestore-rules-author.md in the repo lists run_command among its tools and sets commandExecutionPolicy: auto. The skills CLI route installs the skill; check whether your install route also adds this subagent. github.com

• Plugin routes add a Firebase MCP server

  The README plugin routes install more than this skill. The repo .mcp.json registers an MCP server that runs npx -y firebase-tools@15.32.1 mcp. github.com