cli reference

familiar conjure

compose system instructions for an agent.

familiar conjure <agent> <conjurings...> [--into <path>]

arguments:

argument	description
agent	target agent: codex or claude
conjurings	one or more profile names

options:

option	description
--into	target repository path (default: current directory)
--save-subagent	write composed conjurings as a reusable subagent
--subagent-name	override subagent name (default: joined conjuring names)

behavior:

• combines core profile with specified conjurings
• writes AGENTS.md (codex) or CLAUDE.md (claude) to repo root
• with --save-subagent, also writes a subagent file:
  • claude: .claude/subagents/<name>/AGENT.md
  • codex: .codex/subagents/<name>/AGENT.md

examples:

# rust project with security focus
familiar conjure codex rust sec

# python project, different directory
familiar conjure claude python --into /path/to/repo

# export conjurings as subagents
familiar conjure claude python sec --save-subagent
familiar conjure codex rust infra --save-subagent --subagent-name ship_ops

familiar invoke

run a task prompt through the agent.

familiar invoke <agent> <invocation> [--into <path>] [--headless] [--kv <key=value>...] [args...]

arguments:

argument	description
agent	target agent: codex or claude
invocation	invocation name
args	positional arguments for the invocation

options:

option	description
--into	target repository path (default: current directory)
--headless	run without interactive ui
--auto	skip agent permission prompts (--full-auto for codex, --dangerously-skip-permissions for claude)
--dry-run	print rendered prompt and exit (does not run the agent)
--worktree	run in a separate git worktree to avoid interfering with local changes
--save-skill	write rendered invocation as a reusable skill and exit
--skill-name	override skill name (default: invocation name)
--kv	named arguments as key=value pairs

behavior:

• renders invocation with provided arguments
• with --save-skill, writes a skill file and exits:
  • claude: .claude/skills/<name>/SKILL.md
  • codex: .codex/skills/<name>/SKILL.md
• if --worktree is specified:
  • creates a temporary git worktree from HEAD
  • copies the agent instruction file (CLAUDE.md or AGENTS.md) to the worktree
  • runs the agent in the worktree directory
• runs the agent with the prompt

examples:

# interactive mode
familiar invoke codex bootstrap-rust myapp bin

# headless mode
familiar invoke codex add-tests "parse_config" --headless

# with named arguments
familiar invoke codex implement-feature --kv spec="add caching"

# preview rendered prompt without running the agent
familiar invoke codex bootstrap-python myapp cli --dry-run

# export invocation as a reusable skill
familiar invoke claude code-review --save-skill
familiar invoke codex refactor src/foo.py --save-skill --skill-name cleanup_refactor

familiar list

list available conjurings, invocations, and snippets.

familiar list [kind] [--into <path>] [-v|--verbose]

arguments:

argument	description
kind	what to list: conjurings, invocations, or snippets (default: all)

options:

option	description
--into	target repository path (default: current directory)
-v, --verbose	show first line of each file

behavior:

• lists built-in and local items
• local overrides are marked with (local)
• items are sorted alphabetically

examples:

# list everything
familiar list

# list only conjurings
familiar list conjurings

# list invocations with descriptions
familiar list invocations -v

# list available snippets
familiar list snippets

# list in specific repo
familiar list --into /path/to/repo

familiar lint

lint conjurings and invocations.

familiar lint [--into <path>] [--errors-only]

options:

option	description
--into	target repository path (default: current directory)
--errors-only	show only errors, not warnings

behavior:

• checks conjurings for proper markdown structure
• checks invocations for recommended sections and placeholder documentation
• errors cause non-zero exit code; warnings do not

see linting for detailed rules.

examples:

# lint everything
familiar lint

# lint with only errors (no warnings)
familiar lint --errors-only

# lint specific repo
familiar lint --into /path/to/repo