CLI: integrations, automations, and evals
These commands connect Codifix to humans, agents, built-in loop automations, MCP clients, and evaluation suites.
| Command | Writes | Purpose |
|---|---|---|
codifix tui | Mostly read | Open the terminal UI. |
codifix auth | Mixed | Inspect identity and administer scoped service access. |
codifix agent | Yes | Coordinate concurrent agent branches and worktrees. |
codifix watcher | Yes | Manage project watchers and watcher daemons. |
codifix loops | Mixed | Inspect and operate built-in automation loops. |
codifix mcp | Mixed | Run and inspect the built-in Codifix MCP server. |
codifix eval | Yes | Run automated Memory quality evaluations. |
codifix tui
Open the main terminal interface.
codifix tui
codifix tui --project memoryThe TUI gives humans a dense view of memories, query evidence, replacement proposals, agents, watchers, activities, embeddings, errors, and project status.
codifix auth
Use auth to verify the credential seen by the backend and to administer scoped service tokens. For browser identity, project membership, and Authentik setup, see Authentication.
codifix auth whoami
codifix auth token listcodifix agent
Record which local agents are working on which branches or worktrees.
codifix agent start --project memory --task "Graph UI follow-up"
codifix agent status --project memory
codifix agent status --project memory --include-finished --json
codifix agent finish --project memory --summary "Pushed graph UI follow-up"
codifix agent finish --project memory --abandoned --summary "Superseded by main"start reads local Git metadata and records the repository root, worktree path,
branch, optional task label, head/base commits, dirty files, agent CLI, session
id, writer id, and service endpoint. status warns about same-branch work,
same-worktree work, dirty-file overlap, dirty active workspaces, and stale
heartbeats. finish marks one workspace completed or abandoned and records the
final head, dirty files, pushed state, optional summary, and optional merge
commit.
codifix watcher
Watchers attach to active agent sessions and record useful runtime state.
codifix watcher manager enable --dry-run
codifix watcher manager status
codifix watcher run --project memory
codifix watcher status --project memory
codifix status --project memoryThe watcher manager is the preferred local automation path for Codex-linked sessions. Legacy per-project watcher services and foreground watcher runs remain compatibility paths.
codifix loops
Operate the built-in loop automation control plane.
codifix loops list
codifix loops show memory_hygiene --project memory
codifix loops enable memory_hygiene --project memory --mode suggest_only --explicit-user-approval
codifix loops run memory_hygiene --project memory --dry-run --reason "inspect cleanup candidates"
codifix loops runs --project memory
codifix loops inspect <run-id>
codifix loops approvals --project memory
codifix loops memory-proposals --project memory --status pendingUse codifix loops for the Automations tab workflows from a terminal: definitions, settings, trigger routing, manual runs, context packs, approvals, and loop-produced memory proposals. See Automations for the built-in loop catalog and safety model.
codifix watcher flush
Flush pending watcher-collected work into captures when waiting for the idle threshold is not useful. This writes state, so start with a dry run.
codifix watcher flush --project memory --dry-run
codifix watcher flush --project memory --curateMigration note: older documentation used memory automation flush. The current command is codifix watcher flush. Use codifix loops for the built-in automation control plane.
codifix mcp
Run the local stdio MCP server or inspect service-mounted HTTP MCP.
codifix mcp run --project memory
codifix mcp status --project memory
codifix mcp status --project memory --jsonStdio mode is the normal local Codex/Claude path and must write only MCP JSON-RPC messages to stdout. HTTP MCP is local-first and should require token-bearing clients.
codifix eval
Run repeatable memory-quality suites.
codifix eval doctor --suite evals/examples/memory-smoke
codifix eval run --suite evals/suites/memory-improvement-v1 --condition full-memory --dry-run
codifix eval compare --baseline target/memory-evals/no-memory.json --candidate target/memory-evals/full-memory.json --text
codifix eval gate --comparison target/memory-evals/comparison.json --policy evals/gates/research-v1.tomlUse dry-runs before shell-enabled suites. Evaluation artifacts are evidence for bounded claims about the suite, model, commit, configuration, retrieval settings, token cost, and latency.