Bernstein operator commands

Full operator command surface. The README keeps the short list; this page is the long form.

For session monitoring commands (live, dashboard, status, ps, cost, doctor, recap, trace, etc.) see bernstein --help.

Core operator commands

Command	What it does
bernstein pr	Auto-creates a GitHub PR from a completed session; body carries the janitor's gate results and token/USD cost breakdown.
bernstein from-ticket <url>	Imports a Linear / GitHub Issues / Jira ticket as a Bernstein task. Label-based role + scope inference. Supports --dry-run and --run.
bernstein ticket import <url>	Alias / group form of from-ticket for scripting.
bernstein backlog claim --role reviewer	Atomically claims one eligible row from .sdd/runtime/task-backlog.json for external workers sharing a same-host JSON backlog. Supports --backlog, --agent-id, --project, --capability, --done, --max-attempts, and --json.
bernstein hooks	Lifecycle hooks for pre_task, post_task, pre_merge, post_merge, pre_spawn, post_spawn; shell scripts or pluggy @hookimpls. hooks list, hooks run <event>, hooks check.
bernstein chat serve --platform=telegram|discord|slack	Drive runs from chat with /run, /status, /approve, /reject, /switch, /stop.
bernstein workflow run <name>	Run a YAML workflow manifest. Also workflow list, workflow init, workflow validate.
bernstein approve-tool / bernstein reject-tool	Interactive mid-run tool-call approval. --latest, --id, --always. Flag form: approve --tool <id> / reject --tool <id>.
bernstein autofix	Daemon that monitors open Bernstein PRs; spawns a fixer agent when CI fails and pushes the repair automatically.
bernstein preview start	Sandboxed dev server for the current branch with a shareable public tunnel URL.
bernstein remote	SSH sandbox backend. remote test <host>, remote run <host> <path>, remote forget <host>. ControlMaster socket reuse for fast repeat calls.
bernstein tunnel start <port> [--provider auto|cloudflared|ngrok|bore|tailscale]	One wrapper around four tunnel providers. Also tunnel list, tunnel stop <name>|--all.
bernstein daemon install [--user|--system] [--command="..."] [--env KEY=VAL]...	Installs a systemd (Linux) or launchd (macOS) unit for auto-start. Also daemon start/stop/restart/status/uninstall.
bernstein connect <provider> / bernstein creds	Stores and rotates API credentials in the OS keychain. Agents inherit scoped keys per-run.
bernstein sandbox web-test <task-id> --url <url> --scenarios <yaml>	Drives a Playwright self-test against the dev server. See docs/sandbox/playwright-self-test.md.
bernstein agents-md	Generates a canonical AAIF AGENTS.md for the repo and rewrites it into each CLI's native shape (CLAUDE.md, .cursor/rules/*.mdc, CONVENTIONS.md, .goosehints). generate, write, sync, verify, diff.
bernstein scaffold "<prompt>"	Bootstraps a project skeleton from a single goal prompt. --template auto|python-cli|..., --output <dir>, --force.
bernstein wiki build	Renders WIKI.md for the current repo from the AST symbol graph. Local, no LLM call, no cloud round-trip.
bernstein simulate <plan.yaml>	Digital-twin dry-run: predicts cost band (p50/p90), wall-clock, abandonment probability, per-task blast-radius, and bottlenecks against historical .sdd/traces/ + .sdd/metrics/ without spawning a real agent or hitting the network. See docs/operations/simulate.md.
bernstein compare <spec> --adapters claude,codex[,...]	Side-by-side adapter A/B in isolated per-adapter worktrees. Up to four adapters, deterministic seed, unified diff against baseline. See docs/operations/compare.md.
bernstein recipes list / show / run	First-class workflow library. Parameterised recipes live in templates/recipes/*.yaml. See docs/operations/recipes.md.
bernstein recipes register / fire / history / repair-lineage	Content-addressed registered runs. Each fire writes a receipt to the audit chain; fire needs a reachable, authenticated task server (exit 2 when it cannot submit); repair-lineage resolves a forked definition lineage additively. See docs/operations/recipes.md.
bernstein resume <task-id>	Pick up a task from its last checkpoint.json instead of restarting. See docs/operations/resume.md.
bernstein fork --run <id> --from-step N	Rewind a run to journal step N and branch a new run from its content-addressed worktree snapshot. The snapshot sha is recorded in the event journal, so a tampered snapshot ref is detected. See docs/operations/fork-from-step.md.
bernstein worktrees list / gc	Inspect and reap orphan worktrees. Four-state classifier (active / orphan / stale / corrupt). See docs/operations/worktrees.md.
bernstein telemetry on / off / status / export	Opt-in operator telemetry. Default off; honours DO_NOT_TRACK and BERNSTEIN_TELEMETRY=0. See docs/telemetry.md.
bernstein doctor extended	Extended pre-flight on top of bernstein doctor: adapter conformance, network reachability, and CI integration probes. See docs/operations/doctor.md.
bernstein adapters check / list-status	Conformance plus capability matrix for installed adapters. See docs/operations/adapters.md.
bernstein decisions tail / search	Inspect .sdd/runtime/decisions.jsonl: every routing / criterion-profile / gate-fire decision. See docs/operations/decision-log.md.
bernstein abandonments list / stats	Read-side of the agent-abandon ledger at .sdd/runtime/abandonments.jsonl. See docs/operations/abandonments.md.
bernstein criterion-profile list / show	Inspect per-task criterion profile (correctness / cost / latency / reversibility). See docs/operations/criterion-profiles.md.
bernstein eval calibration report	Brier score + ECE + reliability buckets over .sdd/metrics/calibration.jsonl. See docs/operations/calibration.md.
bernstein lineage v2 show / verify / export	Opt-in two-layer lineage store. See docs/operations/lineage-v2.md.
bernstein run --retry-budget SPEC	Criterion-aware retry budget. See docs/operations/retry-budget.md.
bernstein identity show / decode / verify / disable	Operator-side helpers for the install-rev fingerprint embedded in shared yaml/trace/role-prompt artefacts.
bernstein security role-adapter-policy	Inspects and edits the per-role adapter allow-list (deny-list enforcement at spawn time).
bernstein run-lookup NAME	Resolve a memorable run name back to its run UUID; exits non-zero when the name is malformed or unknown. Example: bernstein run-lookup brave-otter-1234.
bernstein stop [--force]	Graceful drain (soft stop) by default. --force/--hard SIGKILLs everything immediately, then reaps whole process groups so a re-parented grandchild (disowned heartbeat/curl loops) dies with its leader; the summary counts only PIDs confirmed terminated.
bernstein self check-update / update / pin / rollback	Provenance-verified update lifecycle. Verifies a signed release feed against a configured trust root before naming a candidate, seals a chain-anchored advisory that --verify recomputes offline, refuses to update while a run is active, and re-hashes the wheel before install. Offline-first and opt-in; the air-gap profile disables the remote path. See updates.md.
bernstein cluster status / bernstein cluster nodes	Render the node registry as a table (id, adapter, heartbeat age, claimed tasks). status adds the topology summary line. Both take --json-output and --server-url. See cluster-mode.md.
bernstein serve	Runs the task server in the foreground (blocks until SIGINT/SIGTERM) instead of detaching it, so a container's PID 1 stays alive and can host a long-lived central/coordinator node whose /health endpoint stays reachable. It is the published image's default CMD. Set BERNSTEIN_BIND_HOST=0.0.0.0 and BERNSTEIN_CLUSTER_ENABLED=1 to bind all interfaces and expose cluster endpoints.
bernstein schedule add|list|run	Manage operator-registered recurring schedules; schedule audit walks persisted fire receipts to prove the sequence is replayable.
bernstein schedule show <id> --at <time>|schedule verify	Treats a recurring fire as a pure projection of (schedule, fire_time, state) onto a canonical task graph: show --at prints the deterministic graph hash a fire would dispatch without firing (no journal, no receipt, no side effects), and verify replays a recorded fire and confirms the graph hash reproduces byte-identically. RFC-5545 RRULE and cron are both accepted; a webhook / file-change trigger binds its event as an input hash.
bernstein sla add|list|show|verify|report	Attach a per-goal SLA contract (a content-addressed document) to a single recurring goal, task family, or spend envelope, declaring run-duration, start-lateness, fire-frequency, artifact-freshness, and spend-rate axes. The supervisor evaluates contracts against chain evidence on each tick (read-only, never dispatches); a breach becomes a signed violation receipt embedding the contract hash, the chain evidence of the miss, and a deterministic, budget-gated remediation. sla verify <receipt.json> re-derives the verdict from the embedded evidence and checks the Ed25519 signature offline (flip any byte and it fails); sla report projects a per-contract error budget over the work-ledger segment so two parties derive identical numbers from identical history.
bernstein templates compress <role>|--all	Operator-gated, one-time LLM compression of role prompt templates: mechanically validated (fenced blocks, headings, URLs, placeholders, completion contract stay byte-equal), originals backed up out of tree by content hash, receipt chained to the audit log. bernstein templates restore <role> reverses it byte-identically; savings appear in bernstein cost --by role.
bernstein identity keydir	Prints the install-identity key directory (JWKS) - the Ed25519 public keys that verify the RFC 9421 HTTP Message Signatures Bernstein places on its outbound agent-facing requests (also served at /.well-known/http-message-signatures-directory). Set BERNSTEIN_HTTP_SIGNING_REQUIRED=1 to refuse unsigned outbound paths.
bernstein security-review [<task>]	Pattern-scans a unified diff for hardcoded secrets, unsafe eval/exec, shell injection, weak crypto, path traversal, and SQL injection without calling a model. Scans the working tree against --base by default, one agent's task diff when given a task id, or a piped diff with --diff-file -. Exits 1 on any critical/high finding so it can gate a pre-commit hook or a CI step, 2 when there is no diff to scan, and --as-json emits the findings for a pipeline.
bernstein delegation verify <run>	Reconstructs the principal -> orchestrator -> sub-agent delegation chain for a run from HMAC-chained per-hop receipts and confirms it is intact offline; exits non-zero on any tamper or deleted hop.
bernstein artifact list|show <task> [<key>]	Renders a task's agent-posted artifacts (markdown reports, tables, preview links, and normalized SARIF findings) with per-version verification state. Finding addresses bind the rule, normalized artifact URI, source snippet, relative region shape, scanner version, pinned ruleset/feed, invocation, and target. They stay stable when blank lines move an unchanged snippet, when the same source is checked out with different line endings, and when a path arrives in Windows separator or decomposed-Unicode form. Each artifact is content-addressed, spine-anchored, and journal-chained; a version whose stored blob or recorded finding address does not recompute renders as tampered rather than as content. bernstein audit verify walks every artifact, so a mismatch fails naming the key and journal position.
bernstein skills package install|update|verify|status|conformance	Installs the bundled cross-vendor bernstein-run agent skill into a host's skill directory (--host claude|codex|copilot|cursor|gemini, or --dest) and anchors a content-addressed install receipt in the lineage spine and audit chain; verify re-hashes the installed tree and proves it against the receipt; update supersedes a prior install with a receipt binding the prior content address to the new one; status verifies every host install at once; conformance installs into several hosts against one install, replays the skill's self-check contract per host, and seals the pass/fail table into a chain-anchored conformance receipt. --record-only anchors a plugin checkout the host installed itself. See agent sessions.
bernstein datasource register|query|verify	Read-only SQL datasources whose results become content-addressed query receipts: query canonicalises the exact result set an agent saw and binds its SHA-256 into a signed lineage entry (.jws sidecar); verify proves the signature, chain anchor and stored copy offline, and verify --re-execute re-runs the query to report MATCH or DRIFT. DML/DDL is refused with a typed error and connection secrets never reach a receipt. See datasources.
bernstein bench run <suite>|bench verify <bundle>	Runs a content-addressed benchmark suite and emits a signed submission bundle in which every task carries the replay receipt its score was derived from. bench verify replays those receipts offline with no access to the submitter's machine, recomputes each score, and reports MATCH or names the exact task whose replay diverged. A flipped verdict or a corrupted receipt fails verification, so a published number is only worth the bundle a third party can re-verify. See bench.

Monitoring

bernstein live       # TUI dashboard
bernstein dashboard  # web dashboard
bernstein status     # task summary
bernstein ps         # running agents (PID files + a live process-table cross-check, so it still lists survivors after stop --force removed the PID files)
bernstein cost       # spend by model/task
bernstein doctor     # pre-flight checks
bernstein recap      # post-run summary
bernstein export     # shareable HTML/Markdown report of the latest run
bernstein trace <ID> # agent decision trace
bernstein changelog --hours 48  # changelog from agent-produced diffs
bernstein explain <cmd>  # detailed help with examples
bernstein dry-run    # preview tasks without executing
bernstein impact deps # API breakage + downstream caller impact
bernstein aliases    # show command shortcuts
bernstein config-path    # show config file locations
bernstein init --wizard  # interactive project setup
bernstein debug bundle   # collect logs, config, and state for bug reports
bernstein skills list    # discoverable skill packs (progressive disclosure)
bernstein skills show <name>  # print a skill body with its references

bernstein fingerprint build --corpus-dir ~/oss-corpus  # build local similarity index
bernstein fingerprint check src/foo.py                 # check generated code against the index

Deprecated command names

Five top-level names duplicated a group that already owned the same action (#3138). The group spelling is canonical from 3.x on; the top-level spelling still resolves, prints a deprecation warning on stderr, and is unregistered in v4.0.0. Flags and arguments are unchanged unless noted.

Deprecated	Use instead	Notes
bernstein estimate	bernstein cost estimate	Same flags.
bernstein cost-envelopes show	bernstein cost envelopes show	Same subcommand and flags.
bernstein artifacts list|show	bernstein artifact list|show	Same arguments. artifact list with no task keeps its own meaning (lineage-spine keys).
bernstein skill provenance|verify	bernstein skills provenance|verify	Same arguments.
bernstein debug-bundle	bernstein debug bundle	Not a rename. Different builder: --output becomes --out, --yes is unnecessary (no confirmation prompt), and --extended has no equivalent. See debug bundle.

bernstein pool is not deprecated. It projects the sandbox-pool registry from the audit chain; bernstein limits pool governs admission slot pools in the work ledger. The two address different stores, so folding one into the other would change results rather than just the name.

Four more top-level names were folded into the domain group that already owned the surface (#3140). Same rule: the fold spelling is canonical, the old spelling still resolves with a stderr warning, and it is unregistered in v4.0.0.

Deprecated	Use instead	Notes
bernstein quickstart	bernstein demo --flask-todo	Same demo. The retained alias keeps its own adapter auto-detection; demo --flask-todo runs on the mock adapter unless --real is passed.
bernstein init-wizard	bernstein init --wizard	Same wizard. bernstein i is the shortcut.
bernstein validate	bernstein plan validate	Same argument and flags.
bernstein routine <sub>	bernstein schedule routine <sub>	Same subcommands (export, provision, register, bindings) and flags.

The warning goes to stderr only, so bernstein cost-envelopes show --json | jq and friends keep working while a script is migrated.