Complete CLI reference for csa. All commands support --format json for
machine-readable output.
Run a prompt against an AI tool with session management and resource checks.
csa run --sa-mode false [OPTIONS] [PROMPT]| Flag | Description |
|---|---|
--sa-mode <BOOL> |
Root callers must pass true or false; internal recursive calls default to false |
--tool <TOOL> |
Tool selection: auto (default), any-available, or specific name |
--auto-route <INTENT> |
Resolve routing intent through [tier_mapping] or a tier selector while keeping tool choice automatic |
--hint-difficulty <LABEL> |
Resolve a difficulty label through [tier_mapping] when no explicit --tier/--model-spec is set |
--skill <NAME> |
Run a named skill as a sub-agent |
--session <ID> |
Resume existing session (ULID or prefix) |
--last |
Resume the most recent session |
--description <TEXT> |
Description for a new session |
--ephemeral |
Ephemeral session (no project files, auto-cleanup) |
--model-spec <SPEC> |
Full model spec: tool/provider/model/thinking |
--model <MODEL> |
Override tool default model |
--thinking <LEVEL> |
Thinking budget: low, medium, high, xhigh |
--force |
Bypass tier whitelist enforcement |
--no-failover |
Disable automatic 429 failover |
--wait |
Block-wait for a free slot instead of failing |
--idle-timeout <SECS> |
Kill when no output for N seconds |
--no-idle-timeout |
Disable idle-timeout killing |
--stream-stdout |
Force stdout streaming to stderr |
--no-stream-stdout |
Suppress real-time streaming |
--cd <DIR> |
Working directory |
If PROMPT is omitted, reads from stdin.
Examples:
csa run --sa-mode false "fix the login bug"
csa run --sa-mode false --tool codex --thinking high "refactor error handling"
csa run --sa-mode false --tool claude --hint-difficulty quick_question "answer briefly"
csa run --sa-mode false --auto-route analysis "trace the auth flow"
csa run --sa-mode false --model-spec "codex/openai/gpt-5.3-codex/xhigh" "complex task"
csa run --sa-mode false --last "continue where I left off"
echo "analyze this" | csa run --sa-mode false --tool gemini-cliReview code changes using a heterogeneous AI model.
csa review --sa-mode false [OPTIONS]| Flag | Description |
|---|---|
--sa-mode <BOOL> |
Root callers must pass true or false; internal recursive calls default to false |
--diff |
Review uncommitted changes (git diff HEAD) |
--range <RANGE> |
Review a commit range (e.g., main...HEAD) |
--commit <SHA> |
Review a specific commit |
--files <PATHSPEC> |
Review specific files |
--branch <BRANCH> |
Compare against branch (default: main) |
--tool <TOOL> |
Override tool selection |
--hint-difficulty <LABEL> |
Resolve a difficulty label through [tier_mapping] when no explicit --tier/--model-spec is set |
--model <MODEL> |
Override model |
--fix |
Review-and-fix mode (apply fixes directly) |
--security-mode <MODE> |
auto, on, or off |
--reviewers <N> |
Number of parallel reviewers (default: 1) |
--consensus <STRATEGY> |
majority, weighted, or unanimous |
--context <FILE> |
Path to context file (e.g., TODO plan) |
--timeout <SECS> |
Absolute wall-clock timeout |
--idle-timeout <SECS> |
Kill on output silence |
--allow-fallback |
Warn instead of error when pattern missing |
--session <ID> |
Resume existing review session |
Examples:
csa review --sa-mode false --diff
csa review --sa-mode false --tool claude-code --hint-difficulty code_review --diff
csa review --sa-mode false --range main...HEAD
csa review --sa-mode false --diff --reviewers 3 --consensus majority
csa review --sa-mode false --diff --fix --security-mode onRun a multi-round debate between heterogeneous AI tools.
csa debate --sa-mode false [OPTIONS] [QUESTION]| Flag | Description |
|---|---|
--sa-mode <BOOL> |
Root callers must pass true or false; internal recursive calls default to false |
--tool <TOOL> |
Override tool selection |
--hint-difficulty <LABEL> |
Resolve a difficulty label through [tier_mapping] when no explicit --tier/--model-spec is set |
--session <ID> |
Resume existing debate session |
--model <MODEL> |
Override model |
--thinking <LEVEL> |
Thinking budget |
--rounds <N> |
Number of debate rounds (default: 3) |
--timeout <SECS> |
Absolute wall-clock timeout |
--idle-timeout <SECS> |
Kill on output silence |
Examples:
csa debate --sa-mode false "Should we use anyhow or thiserror?"
csa debate --sa-mode false --tool claude-code --hint-difficulty architecture_design "Pick the storage boundary"
csa debate --sa-mode false --session 01JK "reconsider with performance data"
csa debate --sa-mode false --rounds 5 "Redis vs Memcached for session storage"csa session list [--tree] [--tool <TOOLS>] [--branch <BRANCH>] [--cd <DIR>]Send tool-specific compression command (/compress or /compact).
csa session compress --session <ID> [--cd <DIR>]csa session delete --session <ID> [--cd <DIR>]Remove sessions not accessed within N days.
csa session clean --days <N> [--dry-run] [--tool <TOOLS>] [--cd <DIR>]Show the last execution result.
csa session result --session <ID> [--json] [--cd <DIR>]csa session logs --session <ID> [--tail <N>] [--cd <DIR>]Check whether a session is still running via filesystem liveness signals.
csa session is-alive --session <ID> [--cd <DIR>]List artifacts in a session's output directory.
csa session artifacts --session <ID> [--cd <DIR>]Show git history for a session.
csa session log --session <ID> [--cd <DIR>]Write a checkpoint note (git notes) for audit trail.
csa session checkpoint --session <ID> [--cd <DIR>]List all checkpoint notes.
csa session checkpoints [--cd <DIR>]Display effective merged configuration.
csa config show [--cd <DIR>]Open project config in $EDITOR.
csa config edit [--cd <DIR>]Validate configuration file syntax and references.
csa config validate [--cd <DIR>]Query a single config value by dotted key path.
csa config get <KEY> [--default <VALUE>] [--project] [--global] [--cd <DIR>]Examples:
csa config get review.tool
csa config get tools.codex.enabled --default true
csa config get review.tool --globalcsa todo create <NAME>csa todo show -t <TIMESTAMP>csa todo diff -t <TIMESTAMP> --from <VER> --to <VER>csa todo dag --format mermaidcsa todo list [--status <STATUS>]csa todo status <TIMESTAMP> <STATUS>Execute compiled weave workflow files.
csa plan run --sa-mode false <FILE> [--var KEY=VALUE...] [--tool <TOOL>] [--dry-run] [--cd <DIR>]csa mcp-hub serve [--background] [--foreground] [--socket <PATH>] [--systemd-activation]csa mcp-hub status [--socket <PATH>]csa mcp-hub stop [--socket <PATH>]Regenerate the mcp-hub routing-guide skill from live tools/list.
csa mcp-hub gen-skill [--socket <PATH>]csa skill install <SOURCE> [--target <TOOL>]csa skill listcsa audit init [--root <PATH>] [--ignore <PATTERN>...] [--mirror-dir <DIR>]csa audit status [--format text|json] [--filter <STATUS>] [--order topo|depth|alpha]csa audit update <FILES...> [--status <STATUS>] [--auditor <NAME>]csa audit approve <FILES...> [--approved-by <NAME>]csa audit reset <FILES...>
csa audit sync| Command | Description |
|---|---|
csa init [--full] [--template] |
Initialize project configuration |
csa doctor |
Check environment and tool availability |
csa gc [--dry-run] [--max-age-days N] [--global] |
Garbage collect expired sessions and locks |
csa tiers list |
List configured tiers with model specs |
csa batch --sa-mode false <FILE> [--dry-run] |
Execute tasks from a batch TOML file |
csa setup claude-code |
Setup MCP integration for Claude Code |
csa setup codex |
Setup MCP integration for Codex |
csa setup opencode |
Setup MCP integration for OpenCode |
csa migrate [--dry-run] [--status] |
Run pending config/state migrations |
csa self-update [--check] |
Update CSA to the latest release |
csa mcp-server |
Run as MCP server (JSON-RPC over stdio) |