team-settings.json

team-settings.json (v1.16+) is an optional file that lets a tech lead set org-wide policy for skill and tier usage. When present in .claude/, Tierward enforces it at runtime via the PreToolUse hook, with no model discretion involved.

.claude/team-settings.json

Schema

{
  "minTier": "m",
  "required": ["security-audit", "arch-audit"],
  "blocked": ["context-review"],
  "prReviewSeverity": "major"
}

Field	Type	Purpose
minTier	"0" | "s" | "m" | "l"	Minimum tier allowed in this project. Claude cannot work below this tier.
required	string[]	Skills that must run before a task is marked done.
blocked	string[]	Skills that are not allowed to run in this project.
prReviewSeverity	"critical" | "major" | "minor"	Minimum severity threshold for /pr-review to block a PR.

All fields are optional. An empty object {} is valid and has no effect.

Enforcement

The PreToolUse hook reads team-settings.json before every tool call. If a blocked skill is invoked, the hook returns decision: block before the skill executes. If minTier is set and the current project tier is lower, the hook blocks the session and explains the mismatch.

This enforcement runs in the hook layer, outside Claude's context, so it cannot be reasoned around or overridden by model output.

MCP access

When the Tierward MCP server is running, team-settings.json is readable via tierward_team_settings:

{
  "mcpServers": {
    "tierward": { "command": "tierward-mcp" }
  }
}

This lets any MCP-aware client (Claude Desktop, Cursor, VS Code) read the team policy without running the Tierward CLI.

Example: enforcing security audit before deploy

{
  "minTier": "s",
  "required": ["security-audit"],
  "prReviewSeverity": "critical"
}

With this config:

• Tier 0 is not allowed in this project
• /security-audit must run before Claude marks any task done
• /pr-review only blocks on Critical findings (Major and Minor are surfaced but do not block)