CLI commands

The aicr CLI (built from packages/cli) is the entry point for serving, one-shot reviews, eval, replay, memory inspection, and template linting. Build it first with pnpm build, then invoke via Node:

node packages/cli/dist/index.js <command> [options]

On Linux/CI where pnpm runs directly, the same binary is invoked the same way. Show help at any time with --help / -h.

Commands

Command	Purpose
serve	Start the webhook server
review	Run a code review (prompt prep or full dry-run)
eval	Run evaluation benchmarks against the configured LLM
replay	Replay a stored review run scaffold
memory	Inspect or clear workspace memory
migrate	Inspect or apply the schema migration ledger
lint	Validate templates or config scaffold
doctor	Print environment diagnostics
help	Show the help message

Global options

Flag	Description
--config <path>	Path to config YAML file
--workspace <id>	Workspace ID
--help, -h	Show the help message
--version, -v	Show version

serve

Start the HTTP server that receives webhooks and trigger events.

node packages/cli/dist/index.js serve \
  --config example/config.yaml \
  --port 8080

Flag	Description
--port <number>	HTTP listen port (default: server.port from config, else 8080)
--base-prompt <path>	Base system prompt file (default prompts/system/code-reviewer.system.md)

The server exposes /healthz, /readyz, /metrics, /dashboard, /api/admin/*, /webhooks/*, and /triggers/*. See Authentication & secrets for how each route is protected.

On a catchable SIGTERM or SIGINT, serve stops ingress and claims, waits for accepted requests, reviews, publication and final persistence, then closes its stores and queue connections. Wait for AICR server drained and closed. and process exit before migrating. A failed drain exits the command with code 1 and does not authorize migration while the process is still alive. Forced termination (including Windows process termination) does not provide graceful-drain evidence.

review

Run a single review without the long-running server. With --dry-run it prepares and runs the review but skips all output channels.

node packages/cli/dist/index.js review \
  --config example/config.yaml \
  --repo "my-org/my-repo" \
  --provider gitea \
  --source-root . \
  --dry-run

Flag	Description
--repo <ref>	Repository reference (owner/repo)
--provider <name>	Trigger provider kind from the config schema
--trigger <name>	Trigger name
--reason <text>	Review reason
--source-root <path>	Source root directory
--base-prompt <path>	Path to base system prompt template
--changed-file <path>	Changed file (repeatable)
--base-sha <sha>	Base revision SHA
--head-sha <sha>	Head revision SHA
--url <url>	PR / MR / commit URL
--author-username <u>	Author username
--author-email <e>	Author email
--author-display-name <n>	Author display name
--operator-override <kv>	Operator override (repeatable)
--memory-hint <text>	Memory hint passed to the review (repeatable)
--task-context <text>	Extra task context
--dry-run	Run without publishing to output channels
--max-prompt-tokens <n>	Maximum prompt token budget

eval

Run evaluation fixtures. Without secrets, --validate-only checks fixture shape and expected-problem contracts only — this is what CI runs.

# Validate fixtures only (no LLM, no config secrets needed)
node packages/cli/dist/index.js eval --validate-only

# Full benchmark run (loads config + LLM; needs AICR_LLM_API_KEY etc.)
node packages/cli/dist/index.js eval --eval-dir eval/

Flag	Description
--eval-dir <path>	Directory containing eval JSON fixtures
--validate-only	Validate fixtures without loading config or LLM
--base-prompt <path>	Path to base system prompt template

Fixtures live under eval/*.json. The root CI pipeline runs pnpm eval:validate (= eval --validate-only) on every change.

replay

Replay a stored review run scaffold — useful for reproducing a past run from its captured inputs without re-fetching from VCS.

node packages/cli/dist/index.js replay \
  --config example/config.yaml \
  --run-id <id>

Flag	Description
--run-id <id>	Run ID to replay
--workspace <id>	Workspace the run belongs to (default default)
--source-root <path>	Source root directory

memory

Inspect or clear workspace reflection/memory scaffolds. Memory is scoped per workspace; clearing does not cross workspace boundaries.

# Show memory for a workspace
node packages/cli/dist/index.js memory --workspace <id>

# Include full file contents
node packages/cli/dist/index.js memory --workspace <id> --all

# Clear a specific scope (e.g. false-positives)
node packages/cli/dist/index.js memory clear --workspace <id> --scope false-positives

Flag	Description
--workspace <id>	Workspace ID
--scope <scope>	Memory clear scope (false-positives, recurring-issues, etc.)
--all	Include full file contents in memory show

memory subcommands: show (default), clear.

migrate

Inspect or apply the schema migration ledger of the deployment database (storage.database) without starting the server. The command runs the same migration runner the server uses at startup; storage.database.migrate controls the startup behavior (auto / verify).

node packages/cli/dist/index.js migrate --check   --config example/config.yaml

Flag	Description
--status	Read-only ledger report (JSON). Exit 0.
--check	Read-only gate. Exit 0 when clean, 1 when migrations are pending, 2 on checksum drift or a newer unknown schema.
--apply	Apply pending migrations. Exit 0 on success; 2 when the ledger is unsafe (drift / newer schema).

Exactly one of --status, --check, --apply is required. --status and --check never create or modify the database file. Checksum drift and unknown newer schemas are never auto-repaired (exit 2). Both sqlite and postgres report and upgrade the config and store namespaces. A clean config ledger alone is insufficient. SQLite relative paths resolve from the command working directory; --apply creates missing parent directories. PostgreSQL uses postgres.url_env, with postgres.url as a fallback. Connection failures also exit 2. Status/check do not create schemas or tables.

lint

Validate templates or the config scaffold. Render a single template against a sample context to catch template errors before deploying.

node packages/cli/dist/index.js lint \
  --template path/to/template.hbs \
  --template-kind summary

Flag	Description
--template <path>	Template file to render and validate
--template-kind <kind>	Template kind: summary or problem
--channel-kind <kind>	Output channel kind for lint sample context

doctor

Print environment diagnostics as JSON: the current working directory, the Node version, and the resolved config path. Useful first step when troubleshooting a deployment.

node packages/cli/dist/index.js doctor --config example/config.yaml

doctor takes only the global --config flag.