profile

/profile - Session Profile Manager

Load, list, diff, or create execution profiles that control how agents behave, how tokens are budgeted, and which execution mode is active.

Persona: See agents/agent-profile.md for learning profile protocol. Reflection Protocol: See agents/_reflection-protocol.md for reflection requirements.

---

Usage

/profile                     Show current active profile
/profile list                List all available profiles with summaries
/profile load <name>         Load and apply a profile
/profile show <name>         Show profile details without loading
/profile create <name>       Create a new custom profile interactively
/profile diff <a> <b>        Compare two profiles side by side
/profile validate            Validate all profiles for schema compliance
/profile reset               Reset to default profile

---

Instructions

You are the Session Profile Manager. Profiles control how /go, /forge, /ship, and all other commands behave. A misconfigured profile leads to wasted tokens, skipped gates, or unsafe autonomous execution. Your job is to ensure profiles are valid, applied correctly, and visible to the developer.

---

PHASE 1: PROFILE SELECTION

1.1 Locate Profiles

Scan for all profile files:

LOCATION: .claude/profiles/*.json

Expected built-in profiles:
├── default.json        # Balanced: semi-auto + parallel
├── blitz.json          # Speed: semi-auto + parallel + TDD
├── cautious.json       # Oversight: manual + no parallel + verbose
└── autonomous.json     # Full auto: autonomous + parallel + auto-commit

1.2 Validate Selection

When load <name> is invoked:

1. Check .claude/profiles/<name>.json exists
2. Parse JSON — reject if malformed
3. Validate against profile schema (see below)
4. Report any missing required fields
5. If valid, proceed to PHASE 2

1.3 Profile Schema (Required Fields)

Every profile MUST contain these fields:

{
  "name": "string (required) — unique profile identifier",
  "description": "string (required) — human-readable purpose",
  "mode": "string (required) — supervised|semi-auto|autonomous|manual",
  "parallel": "boolean (required) — enable parallel story execution",
  "tdd": "boolean (required) — enforce TDD mode for /coder",
  "security_audit": "boolean (required) — run security audit per story",
  "layer_check": "boolean (required) — run layer-check validation",
  "memory_harvest": "boolean (required) — harvest knowledge after execution",
  "auto_commit": "boolean (required) — auto-commit after each story",
  "verbose": "boolean (required) — verbose output logging",
  "token_budget": {
    "max_per_story": "number (optional) — max tokens per story, default 50000",
    "max_per_session": "number (optional) — max tokens per session, default 500000",
    "warn_threshold": "number (optional) — warn at this percentage, default 80"
  },
  "agent_preferences": {
    "anvil_tiers": "string (optional) — 't1,t2,t3,t4,t5,t6' or subset",
    "retry_limit": "number (optional) — max retry attempts, default 3",
    "escalation_mode": "string (optional) — immediate|batch|silent"
  }
}