Commandem feature reference

Open Command Center from the palette or return to Overview with no project selected. Home aggregates work needing attention and offers configurable widgets. Selecting a card follows its underlying project, run or message; it is not a second authority or data store.

Open Monitor from the palette for project health, agent runs and signals. Inspect a signal's source and timestamp before acting; filtered or paged lists are bounded views, not full archives.

Replay reconstructs a run's recorded events, gate evidence and known causes. Open a run or receipt first when investigating why work stopped; distinguish recorded causal links from inferred explanations.

Use the Experiments workspace to find suites and recorded runs, then inspect their design, controls, outcomes and artifacts. The workspace indexes evidence; it does not make a result valid merely because a runner completed.

Automations explains triggers and their effects. Inspect the trigger, project scope, configured action and resulting receipt before enabling recurring work. Scheduling preferences and notification schedules do not grant tool authority.

Matrix relates conversation evidence across projects and providers. Use relationships to find candidate context, inspect source evidence and review a handoff before another agent receives it.

Fleet is the standing-capacity and bounded-fan-out workspace, peered with the per-launch Swarm record view: Overview lists registered nodes and surfaces with an active-swarms strip, and Swarms lists dispatched swarms with a read-only detail (workers, phases, verdict and receipt totals) for a selected one. This is a review surface; starting, steering, killing, budgeting or adding/removing workers from here is not yet exposed.

Read recent recorded project updates in the full Roadmap or its full and Glance companion. The expanded news region defaults to roughly half the available space and has a draggable divider. Its committed section height controls a row limit capped at twenty; each project's full Roadmap and Glance save separate resize preferences. Refresh checks the feed; Retry recovers an unavailable reader. After the initial load, Show latest updates applies a waiting snapshot without moving Mission reading position. Receiving updates preserves unsent task drafts in the full Roadmap and new-mission drafts in Glance. View updates opens an updates popover; Open recorded source is an explicit browser action; Done closes recorded details. Failed refreshes retain last good updates. No older paging or full-history claim.

Open Plan in the palette to work across project roadmaps. Campaigns organize outcomes; Missions identify coherent work; tasks describe executable slices and dependencies. Research belongs in a Mission when it has a design and acceptance evidence.

Open Week for a weekly review of planned work and recorded progress. Use it to reconcile priorities and capacity across projects rather than dispatching from every status card.

Open Agent runs or Monitor to inspect queued, planning, awaiting-approval, executing and terminal work as exposed by the runner. Approve the exact planned run once and verify its branch and results independently.

Open Runner control to inspect unattended execution. Bounded turn, budget and progress preferences constrain auto-continuation; they are not an instruction to pursue arbitrary work.

Compose evidence-backed checks and inspect preflight before applying changes. A shell verification command and a browser URL check can be combined; the browser gate requires HTTP success and rejects JavaScript exceptions or console errors.

Open Run receipts or the run detail for provider/account identity, execution results, gate evidence and failure reasons. Keep evidence links and redacted screenshots associated with their run.

Command Center's Broadcast inlet scopes run/queue/goal actions across projects. Announce routes a message to active chats; it is a communication path, not permission to execute the message.

Use Peer review workspace, Tool approvals, Review inbox, Review & merge agent branches or Git across projects from the palette. Inspect requested tool authority, branch contents and verification before approving or landing.

History reads the supported agent's session archive for the project/account. A folder rename can orphan path-keyed archives unless they are migrated. Check identity and working directory before concluding that a conversation was lost.

Native resume continues a source-provider session under its owning account. A cross-provider bridge transfers reviewed evidence into a different conversation; it does not grant the destination ownership of the native transcript.

Inspect which sources were selected, included, excluded or truncated in a context manifest. Source identity, revision, privacy provenance and payload accounting explain what the destination could receive.

Select candidate evidence, inspect the preview, remove irrelevant material and approve a current handoff. The transition contract binds plan, preview, review and current state to the payload and manifest draft.

Human labels judge whether two conversations share useful objectives or evidence, not whether their wording looks similar. Related pairs should help continue actual work; insufficient evidence can be skipped. Cross-provider examples need deliberate coverage.

Replay, context provenance and project memory help reconstruct trigger, evidence, mechanism, change and measured effect. Use source-backed links and state confidence explicitly when recording a lesson.

Profiles collects provider configuration, native account references and project-specific selections. A named Claude, Codex or Gemini CLI account isolates provider-owned authentication roots. An API profile instead points to an endpoint.

Loadouts bundle reusable routing and behavior choices, including account references per role. Project and pane overrides are distinct from the global default. Inspect role bindings before unattended work.

Notification preferences provide per-type delivery/priority behavior and a quiet posture. Select a declared mode where supported; inspect the current effective policy rather than assuming a profile name means Do Not Disturb.

Privacy separates provider data-use posture, local retention and delivery authority. A gate needs to know which account/endpoint will receive data; unknown training posture must stay unknown with a provider-setting link.

Incognito concerns local retention and selected session behavior; it is separate from a provider's training policy. Verify provider support and the applicable local artifacts before sending sensitive information.

Open Provider usage or Home Usage to inspect account-scoped evidence. Reported windows, observed exhaustion, transcript estimates and API costs have different meanings. Quota exhaustion is distinct from model-capacity rejection.

Notes preserve editable project knowledge; comments attach discussion to project, Campaign or Mission scope. Inspect whether the current board is local or synchronized and whether the chosen provider supports the operation.

Board read state feeds Home and Notifications. Marking board messages read preserves their history. A notification badge should reflect the same underlying unread state, not an independent dismissal ledger.

Inspect configured collaboration connections, scope mapping, authentication and conflict handling before synchronizing content. External project-management records can map to Campaign/Mission work, but the mapping must be explicit.

Recipes are reusable prompt/workflow templates; the Marketplace groups recipes, plugins and skills. Bundled catalogs load locally first; remote refresh has a separate unavailable state.

Plugins extend commands; skills supply scoped agent instructions; MCP servers expose tools. Install through the supported manager and inspect declared capabilities, provenance and update evidence.

Advisories and read-only questions use a governed bus with an audit trail. Announce routes operator messages to active chats; source and destination identity are essential for coordination.

Save the current layout from the palette, inspect Saved Views, and assign an appropriate saved layout as the project's default. A saved view captures pane configuration; it does not copy the conversation transcript.

A Workspace groups projects and filters status/sidebar views. Select it with the workspace switcher; All projects clears that filter. Workspace export/import is distinct from layout saving.

A workbench is an isolated Git working tree used for concurrent implementation. The primary checkout is the integration lane; branches are reviewed, tested, committed and landed back to trunk.

Use the configured command-palette shortcut to discover global actions, Settings destinations and saved layouts. Static destinations are listed below; project-specific or user-defined entries may also appear.

The project tab bar groups Chat, Code, Terminal, Activity, Insights, Actions, Roadmap and Board. Layout preferences can hide/reorder groups, use icons and adjust sizes; Quick Open/shortcuts still reach hidden tabs.

Author .claude/agents instructions in Profiles, separately from Loadout routing and security policy.

Select a project and open Chat to send prompts, choose agent/account/model and inspect streamed replies, tools and attachments. A split layout can hold independently configured panes.

Browse project files and open a file in the configured pane or external application. Hidden and ignored files are controlled by Layout preferences. Resolve an attachment's actual path before handing it to another app.

Search within the selected project's code and files. Use the project selection and query to scope results, then open the matching file and line.

Inspect project log output and recorded activity while diagnosing a run or check. Match timestamps and run identity rather than inferring cause from neighboring rows.

Read the project's timeline of commits, runs and transitions. Use it to locate evidence relevant to a status change and then open the original record.

Browse provider-session history available to the current project and account. Open the original session for inspection or choose a supported resume path.

The internal tab identifier is graph; the visible product name is Matrix. Inspect conversation relationships and candidate context for this project, with source provenance.

Inspect repository changes against the relevant base before review or merge. Confirm the branch, base and dirty worktree as well as the rendered patch.

Open the selected project's terminal surface for commands. Confirm the working directory, account context and command before execution.

Inspect the project manifest and declared commands/conventions. The manifest connects project behavior to CLI checks and orchestration.

Inspect project-level activity and resource summaries as available from their data sources. Use underlying receipts when exact accounting matters.

Find configured project actions and run the intended operation after reviewing its scope. Actions use the project's declared workflow rather than an independent task system.

Browse Campaigns, Missions, executable tasks, dependency state and completion evidence in the project roadmap. Use durable Mission IDs in links or dispatch references.

Read and send project board messages and scoped comments. Board read state feeds the unified notification system; historical messages remain after being marked read.

Open project knowledge and editable notes. Keep reusable context distinct from transient conversation messages, and inspect the configured storage/sync destination.

Keep project resource links discoverable alongside notes and configuration. Open a URL or local resource with its intended handler.

Project health and attention summary; open the source project to investigate a condition.

Review-oriented overview; inspect the referenced work and approvals before accepting a conclusion.

Pending requests for tool authority. Decide on the exact operation, not all future requests.

Active conversation entry points. Opening a chat is different from authorizing a new turn.

Unread board messages. Mark read preserves the underlying board history.

Recorded reusable project knowledge. Treat older lessons as evidence to review, not unconditional authority.

Work progression across tracked projects. Verify underlying run states when exact execution matters.

Agent branch/review visibility. A branch listing does not prove it is safe or tested.

Provider/account usage evidence. Reported quota, estimates and unknown values are distinct.

Work that needs human review. Dismissal is not a merge or approval.

A proposed next step from current status. A recommendation is not permission to run it.

Potentially stalled work. Inspect heartbeat and logs before restarting owned processes.

Repository context and status. Confirm the actual worktree before applying Git operations.

Time-based prompts for follow-up; not proof an external task ran.

Local resource observations. High utilization needs process evidence before attributing it to a study.

Active monitoring context. A watcher expiring does not necessarily stop the work it observed.

Available cost evidence. Subscription use may have no dollar meter; unknown cost is not free execution.

Global project list, current selection and workspace scope.

File navigation for the selected project; no selection means no project file root.

Search within the selected project; separate from conversation/Matrix retrieval.

Repository changes for the selected project; inspect base and worktree before review.

Global agent/run visibility and entry points.

Command–O: Opens the selected project surface; requires a meaningful selection.

Command–D: Mutes current attention until project state changes; does not fix the condition.

Command–E: Snoozes attention for 24 hours; project history is preserved.

Command–P: Discovers app commands and destinations.

Command–[: Navigates the available project list in the current scope.

Command–]: Navigates the available project list in the current scope.

Control–Tab: Hold to cycle recent projects; this is not the OS Command–Tab switcher.

Command–Shift–B: Opens the unified notification surface, not the OS Notification Center.

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Command palette destination

Provider configuration, named accounts, model/effort defaults, display identity, automatic continuation and exhaustion rules.

Reusable per-role routing, behavior, tool and account references. Loadouts can reference projects and workspace/environment choices without copying credentials.

Tool approval boundaries, local policy, standing directives, protected capabilities and permitted elevated behavior.

Behavior when sending/relaunching, suggestions, smart replies, model reminders, active-turn handling and unattended detach defaults.

Microphone input, local dictation preferences, voice-command confirmation, speech playback voice/rate/volume and auto-read.

Per-type delivery and priority, quiet behavior and notification appearance. Board messages and ordinary feed notifications use their underlying read/dismiss state.

Theme, accent, status colors, density, sender/timestamps, thinking labels, tool folding, fonts, bubbles, backgrounds and rendering preferences.

Sidebar grouping/sort/visibility, tab groups and icons, splits, second-pane behavior, file-link target, hidden/ignored files and infrastructure visibility.

How projects are listed and opened, default interaction behavior and project-oriented preferences. Inspect selected project scope before changing configuration.

Pairing and trusted-device configuration for the supported companion path.

Configurable action bindings and their current overrides. See the shortcut reference for source defaults.

Launch/quit behavior, raw settings access, integrations, recipe registry configuration, dependencies and backup/export tools.

Apply a style in Appearance, then adjust individual controls.

Apply a style in Appearance, then adjust individual controls.

Apply a style in Appearance, then adjust individual controls.

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Button in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in General settings controls

Picker in Chat settings: show Roadmap or Stay on Chat when the last tab moves or closes.

Source initializer: ChatTabFallbackPrefs.defaultMode

Picker in General settings controls

Picker in General settings controls

TextField in General settings controls

TextField in General settings controls

TextField in General settings controls

TextField in General settings controls

TextField in General settings controls

TextField in General settings controls

TextField in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in General settings controls

Toggle in Automatic continuation

Button in Quota exhaustion rules

Picker in Quota exhaustion rules

Picker in Quota exhaustion rules

TextField in Quota exhaustion rules

TextField in Quota exhaustion rules

TextField in Quota exhaustion rules

Toggle in Quota exhaustion rules

Picker in Microphone and voice commands

Toggle in Microphone and voice commands

Button in Speech playback

Button in Speech playback

Picker in Speech playback

TextField in Speech playback

Toggle in Speech playback

Button in Backup and workspace export

Button in Backup and workspace export

Button in Backup and workspace export

Picker in Backup and workspace export

Button in Device pairing

Button in Device pairing

TextField in Device pairing

Button in Notification policy

Button in Provider account preferences

Button in Provider account preferences

Button in Provider account preferences

Button in Provider account preferences

Button in Provider account preferences

Button in Provider account preferences

Picker in Provider account preferences

Picker in Provider account preferences

Picker in Provider account preferences

TextField in Provider account preferences

TextField in Provider account preferences

TextField in Provider account preferences

Toggle in Provider account preferences

Toggle in Provider account preferences

Button in Roles and Loadouts

Picker in Roles and Loadouts

Picker in Roles and Loadouts

Picker in Roles and Loadouts

Picker in Roles and Loadouts

TextField in Roles and Loadouts

Toggle in Roles and Loadouts

SecureField in Protected capabilities

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Button in Policy editor

Picker in Policy editor

TextField in Policy editor

Button in Standing directives

Button in Standing directives

Button in Standing directives

Button in Standing directives

Button in Standing directives

Picker in Standing directives

TextField in Standing directives

Toggle in Standing directives

Choose Always, Only when at bottom (default), or Never. Latest remains available; opening a conversation and explicit navigation are separate actions.

Choose Block, Pills or Document for long user messages. Changes apply to the current conversation view.

Button in Mouse gestures settings

Button in Mouse gestures settings

Button in Mouse gestures settings

Picker in Mouse gestures settings

TextField in Mouse gestures settings

Toggle in Mouse gestures settings

Toggle in Mouse gestures settings

Global defaults only; ask, auto or never. Source declaration only; not runtime certification.

Discovery precedes explicit selection; registration writes local state. Source declaration only; not runtime certification.

Explicit imported content; no native resume or rule activation. Source declaration only; not runtime certification.

Explicit Markdown and inactive rules; no Apply or archive creation. Source declaration only; not runtime certification.

Layout: Follow theme, Glass capsule or Underline (compact). Source declaration only; not runtime certification.

Refresh is unavailable while the physical lookup remains pending. Source declaration only; not runtime certification.

Source initializer: "system"

Source initializer: ""

Source initializer: AgentIdentityPrefs.defaultName

Source initializer: false

Source initializer: false

Source initializer: "system"

Source initializer: "Blue"

Source initializer: 14.0

Source initializer: "Normal"

Source initializer: "standard"

Source initializer: "none"

Source initializer: "System"

Source initializer: "100"

Source initializer: true

Source initializer: true

Source initializer: true

Source initializer: true

Source initializer: true

Source initializer: ControlDensityPrefs.defaultName

Source initializer: "claude"

Source initializer: SecondPanePreference.defaultMode

Source initializer: true

Source initializer: true

Source initializer: "opus"

Source initializer: true

Source initializer: "all"

Source initializer: ""

Source initializer: false

Source initializer: true

Source initializer: ""

Source initializer: "chat-tab"

Source initializer: false

Source initializer: ChatLifecyclePrefs.PauseOnQuit.bounded.rawValue

Source initializer: ""

Source initializer: PrivacyMode.standard.rawValue

Source initializer: false

Source initializer: ""

Source initializer: ""

Source initializer: ChatLifecyclePrefs.ResumeUnsent.autoResend.rawValue

Source initializer: false

Source initializer: "agents"

Source initializer: true

Source initializer: true

Source initializer: false

Source initializer: "sidebar-files"

Source initializer: "status"

Source initializer: false

Source initializer: "open"

Source initializer: false

Source initializer: false

Source initializer: "name"

Source initializer: StatusPalette.style.rawValue

Source initializer: false

Source initializer: false

Source initializer: "medium"

Source initializer: ""

Source initializer: ""

Source initializer: 30

Source initializer: "default"

Source initializer: 10.0

Source initializer: 25

Source initializer: "default"

Source initializer: "status"

Source initializer: "cheap"

Source initializer: true

Source initializer: true

Source initializer: ""

Source initializer: false

Source initializer: SpeechPlaybackPrefs.defaultRate

Source initializer: ""

Source initializer: SpeechPlaybackPrefs.defaultVolume

Source initializer: ""

commandem.chatAutoScroll uses atBottomOnly by default. This global appearance preference does not switch agent accounts.

commandem.longUserBubbleStyle starts as block. Applying a preset may change it; later individual edits are retained.

Source initializer: ""

ThemePrefs.toolbarStyleKey → "". Source declaration only; not runtime certification.

Governed agent-to-agent bus (phase 388): stage Tier-1 advisories, ask Tier-2 read-only questions, read the audit ledger.

List .claude/agents/*.md across registered projects; `agents <project> --rename <old> --to <new>` renames a subagent (e.g. the Commander orchestrator); `--create`/`--edit`/`--delete <name>` author one (the Agents-center backend). Agent Builder: `--templates`, `--create --template <slug>`, `--validate`, `--source <name>` (print the exact stored TOML/YAML/JSON, read-only), `--compile --engine <...>`, `--check`, `--import`/`--export --to managed-agents`, `--export --to managed-agents --deploy [--execute]` (originate + publish a local definition to Managed Agents), `--export --to outcome-gate` (stdout-only unreviewed manual draft), `--check-findings -` (stdin-only committed-location check), `--convert --to <toml|yaml|json>`.

List .claude/agents/*.md across registered projects; `agents <project> --rename <old> --to <new>` renames a subagent (e.g. the Commander orchestrator); `--create`/`--edit`/`--delete <name>` author one (the Agents-center backend).

Additionally reviewed CLI source 61d89aeb lists editable .claude/agents/*.md and read-only .github/agents/*.agent.md; --show reads either set, and a same-name editable definition takes listing precedence. Previous baseline 11249a60; the current source review is shown above. The installed-build behavior is not certified.

Analyze an image or a running dev-server URL (screenshotted via headless Chrome) with claude. --against <design> = verify an implementation against its mockup.

List pending tool approvals across all live chats (cross-project inbox).

Allow or deny a pending tool approval (from `commandem approvals`).

Request orchestrator help from inside a project: stages a plan-only help run for owner approval + posts a board note (no run starts; the enqueue safety class). See also: enqueue, handoff, board.

Committee audit: fan one structured audit prompt out to a panel of independent model judges, then synthesize consensus/disagreements/unique findings. Dry-run by default; --execute spends (local/apple judges are free). See also: review, eval-codebase.

Run a diagnostic AI session on attention-bucket projects (dry-run by default).

Run claude toward a goal unattended (auto-approve + auto-continue; not `ask`).

Generate an AI daily briefing from your project state (reads state; not `brief`, which writes CLAUDE.md).

Hard-floor authorization broker (phase 136): mediate a floor command an unattended run needs, authorized by a live human ceremony -- the wall stays whole, this is a door beside it.

Draft a Conventional Commits message from the staged diff (local model first).

Multi-agent conductor: LLM-decompose ONE goal into tailored per-project tasks, dispatch each as an isolated autopilot run, integrate a review (not `orchestrate`'s verbatim broadcast).

Stage an implement-this-design task from a Figma link, a claude.ai artifact link, or a local mockup (design->code intake; plan-only). See also: enqueue, conduct, analyze.

Queue a cross-project agent task (state-only; no run). --tag/--group queue for a whole set.

Send a codebase summary to an LLM for independent quality assessment (dry-run by default).

Export one run's lifecycle, receipt, diffstat, policy decisions, and A2A events as a reviewable JSON bundle.

Inspect the cross-project experiment lifecycle ledger or statistical results without running models or hooks.

Register remote Outpost nodes or Managed Agents surfaces and see their agent runs in `commandem queue` (fleet aggregation; add/ls/rm). For local agent/model/account bindings, see `roles`, `account`, `provider`, or `model` instead.

Hand a task from this project (cwd, or --from) to another: stages an unattended run on the target with the origin's context attached AND posts a board handoff note. Pull-based (nothing runs until the runner). See also: enqueue, tell, board.

Install, inspect, or remove Commandem token tiering for raw Claude sessions.

Enqueue a task from a GitHub issue (or failing CI run). Fetches title+body via gh, queues an AgentRun.

Hard kill the agent runtime + drop a cross-process stop sentinel.

Measure whether injected context (directives, recipes, memory seeds, or this project's full brief) changes agent behavior: with/without A/B, deterministic checkers.

Show or switch the LOCAL model (auto-discovers ollama / LM Studio / vLLM; persists for the CLI + app).

Auto-continue a STALLED session commandem OWNS (bounded + killable). Refuses an external session (adopt via `commandem chat`). See `commandem stalled`.

Propose, authorize, dispatch, and track a quota-sized, dependency-ordered OPERATION of roadmap work (Mission 850).

Bounded parallel fan-out: run independent sub-tasks concurrently, each an isolated autopilot run (deliberate, not a swarm). Or broadcast one --goal ACROSS projects with --all/--group/--tag.

Fleet policy: editable never/ask-first/allowed command rules across every repo + agent. All tiers editable (incl. floor lifts via unsafe-allow); lifts are disclosed, never silent. `grants` lists the human-approved exact-operation grants the approval cards save (shared across compatible agents).

Capture a lesson from the latest git commit (runs as a .git/hooks/post-commit script).

Act on a GitHub PR (diff/view/approve/request-changes/comment/merge) via gh.

List queued/in-flight/finished runner runs.

Search memory across every project (the cross-project moat).

Record a cross-project lesson from a project (decision/mistake/pattern/fix).

Stage a new queued run from one failed or killed local AgentRun; preserves its dispatch and verify contract.

AI code review of uncommitted changes, a PR (--pr N), or a range (dry-run by default).

Show or set role bindings and task preferences (manifest > global > default).

Pick the best available model for a task (the inspectable side of --model auto).

Opt-in PLAN-ONLY unattended agent loop draining the queue (not `run`).

Quota-maximizing worker: propose (and --stage plan-only) roadmap-derived agent work sized to your UNUSED weekly Claude allowance. Dry-run by default; nothing runs until `commandem runner`. See also: enqueue, runner, usage.

List .claude/skills across projects + user level; `skills new <name> [<pattern>]` scaffolds a SKILL.md + eval.toml; `skills install <name|dir|proj:name>` installs a builtin/published/fleet skill; `skills search`/`skills publish <dir>` use the marketplace.

Suggest your next chat message from a project's own tracked conversations (retrieval; --generate for a local model; --reply for smart-replies to the agent's last message).

Inspect and control a bounded swarm of worker runs.

Post a note to another project on the message board (pull-based; never triggers a run).

Resolve a provider-neutral chat transition, review or seal a Matrix bridge at a request boundary, offer a reviewed quota handback, or record delivered-bridge outcome and relationship feedback.

Manage native login and isolated account profiles for supported agents.

Export all commandem state to a portable tar.gz archive (or restore from one).

Check the claude CLI version vs latest; --install an hourly notifier.

Daily health checkup of the CLIs commandem drives (claude/codex/gemini/gh): version drift + auth state.

Foreground check-runner loop (launchd-managed in production).

Manage per-directory trust for a cwd .env (direnv-style): a ./.env loads only after you --allow this dir.

Start a focus session on a project (default 25m Pomodoro).

Generate a groff man page (commandem.1) for the command surface. --json emits the verb tree instead.

Get or set global commandem preferences (e.g. default_mode).

List snoozes/focus; --include-owned lists agent reminders; Project --add creates one; --disable cancels.

Read-only agent-profile inventory, or opt-in infra status with --only gh|wrangler|notion|linear. Other kinds and interactive setup are not implemented. See account, doctor --deps, clis and provider.

Claude Code statusLine provider: caches REAL subscription usage (rate_limits) + prints a status line. Set in ~/.claude/settings.json.

Explicit self-update via go install @latest (never automatic).

Print the build-stamped version; --check notifies if newer.

Mute a project (or one of its failing checks) from the attention bucket until its state changes.

Stage an operator note for asynchronous delivery into LIVE chat sessions; inspect the durable outcome with --history.

Execute the plan a prior `ask` stored, runner-awaiting runs, or a Managed Agents tool confirmation.

Run a headless claude agent in --mode plan (dry-run) by default.

Live bidirectional agent proxy as NDJSON (claude default; --agent codex|gemini|antigravity|goose for a resume-per-turn session).

Run a project's check_command on demand (or --tag/--group to run a whole set).

Run the pre-deploy readiness gate (git clean, tests, health), then the project's deploy_command.

Managed Deployments (hosted cron) for a linked Managed Agent: ls|show|create|pause|resume|rm. D5 guardrails: both a per-run AND a daily budget cap are required to create one; create/pause/resume/rm are refused inside an unattended run or a plugin; create is dry-run until --execute; rm ARCHIVES (there is no delete).

Run a shell command across projects in parallel and aggregate the output (the cross-project broadcast console). Or `exec <pattern> "<cmd>"`.

Dequeue + start the next queue_file entry for a project.

Open a project in a coding agent (claude default), VS Code, Terminal, or Pathem. From inside tmux, opens a new tmux window instead of Terminal.app.

Run git pull across all (or one) registered project(s).

Branch a chat session rewound to before message N (edit/rewind engine).

Start a project's run_command (detached, in its own process group; not `runner`, the agent loop).

Defer a project from the attention bucket for a duration.

Message a project's RUNNING autopilot: the note rides its next turn, marked [operator].

Stop a project's running command (by process_pattern).

Cross-project activity feed: commits, agent runs, check transitions, deploys (newest first). Optional <pattern> scopes to one project; --news-json requires an exact registered name.

Read the cross-project message board (all, or one project's thread + what it's working on).

Full-text search across every project's Claude conversation transcripts (vs `search`'s code grep).

Scan projects' source CONTENT for dangerous patterns (eval/exec injection, unsafe deserialization, shell injection, XSS sinks, disabled TLS verification, weak crypto, hardcoded credentials). Critical/high findings raise attention.

Comment on a durable local target thread, configure its destination default, migrate legacy Notes/comments, or comment on a legacy activity-feed event.

Score a task's difficulty (free, deterministic) + suggest a lighter claude model/effort for simple work.

List or show a project's claude config files.

Semantic retrieval over ONE project's code/docs (relevance-ranked; vs `search`'s exact grep).

Inspect the bounded, redacted context manifest captured at one model-request boundary.

Provider/Profile reported run cost plus separately labelled Claude transcript list-price estimates.

Show which registered projects depend on each other.

Show uncommitted changes across all (or one) registered project(s).

Weekly work summary across registered projects (workbenches hidden by default).

Catalog, draft, inspect, validate, resolve, or preflight an evidence-backed Gate contract without running an agent.

Watch registered repos' git metadata (config + indexes); log each change with a key diff + process snapshot.

Tail polaris's check transition log (history.jsonl).

Unified review inbox: ONE urgency-ranked list of everything fleet-wide that needs you (tool approvals, agent plans, killed/failed/worktree runs, PRs awaiting review).

Show exhaustion, near-limit preparation, and handback rules with each field's source; simulate a quota window or set scoped JSON on stdin.

Run each project's links_check_command + report broken links (--install schedules it).

Tail the most recent log file for a project.

Report local, content-free lifecycle coverage, export a redacted diagnostic bundle, or derive read-only Plan/Week outcomes.

HTTP health checks against projects with health_url configured.

Concurrent operational snapshot for the Monitor workspace: runs, approvals, fleet, capacity, health, notifications, and receipts.

Cross-project planning snapshot: roadmaps, Week commitments, needs-you work, and provider capacity.

Machine compute pressure: load, memory, heaviest processes (attributed to projects). The "why is it laggy?" answer.

Inspect project privacy policy, run a bounded local sensitivity scan, or preview a Profile-aware data-use decision without launching.

List all projects in polaris's registry.

Reconstruct one run's bounded causal timeline from durable runs, transcripts, policy, gates, limits, and receipts.

Cross-project git-state rollup: which repos are dirty, ahead/behind origin, or carry extra branches. Parallel; read-only.

Show GitHub PRs awaiting your review across all (or one) registered project(s).

List recent screenshots, or --watch for new ones to analyze.

git grep across every registered project, in parallel. With --in and --replace, preview a literal cross-file replacement; --apply requires the preview hash.

List sessions for a project; --agent selects Claude, Codex, Gemini, or Local; --view tails one.

List stuck sessions across providers and Profiles; owned ones can `commandem nudge`, external ones adopt through the reported provider identity.

Show git stashes across all (or one) registered project(s).

Per-project stats: messages, tokens, git commits/files.

Cross-project status (sources: registry + git + gh + claude). Optional pattern scopes the result; --since shows what changed.

Long-lived NDJSON stream of status snapshots + attention deltas.

Estimated time per project from git commit clustering.

Provider/Profile reported run tokens plus Claude archive session/model analysis with estimated list-price cost.

Trace one provider-owned session with explicit message, tool, and result fidelity.

Interactive terminal UI (SSH-friendly); live status + dispatch.

Aggregate Claude usage across projects. Set ANTHROPIC_ADMIN_KEY to also report ground-truth API tokens + billed cost from a Console org.

Scan projects' declared dependencies for known vulnerabilities via OSV-Scanner (route, never rebuild). Critical/high findings raise attention (--install schedules it).

Flag Claude sessions with model-tier overspend (model OUTRANKS what the router would pick) or a low cache-hit rate (below what the session's turn count should plausibly achieve). Observation-only; high-confidence by default, --all adds medium. See `tokens`/`route`.

One concurrent weekly snapshot: needs-you inbox, roadmap candidates, verified completions, reminders, usage, and digest.

Manage declared custom ACP (Agent Client Protocol) CLI profiles: ls|add|rm. A saved profile routes as `--agent acp:<name>` for `chat`, `autopilot`, and `open` (acp_agents.toml; see docs/acp-agents.md).

Add a project to polaris's central registry from its commandem.toml.

Adopt an existing repo (e.g. one teleported from a claude.ai/code web session) as a tracked project, seeding a roadmap from its conversation.

Archive a project (hide from status); --restore to un-archive; --list to show archived.

Automated git bisect run for a registered project.

Sync polaris-awareness + standing directives into every project's CLAUDE.md (writes files; not `brief-me`).

Audit registry state; --deps preflights tools and --missions migrates legacy Mission references.

Manage first-class project groups (named collections of projects).

Audit engineering hygiene across registered projects.

Set a project's avatar emoji (or --clear for the monogram).

Bootstrap commandem from your claude history: register every project you've used claude in.

Scaffold a commandem.toml manifest in the current dir.

Show or edit a project's commandem.toml convention fields (comment-preserving).

Import beacon's project registry into polaris's.

Create a documents, simple, blank, or coding workspace, or use a GitHub template repo.

Per-project scratchpad: append, show, or clear timestamped notes.

Configure the standard directive modules (privacy/secrets/finish/commit) that `brief` propagates.

Security personas: bundle directives + policy floor + autonomy into one posture.

Manage named LLM API profiles: ls|add|rm|discover|test|export|import. Bundles contain credential references, never secret values; import diagnoses missing local credentials without erasing existing references.

Reusable prompt library: render a curated/your-own template with project context; search/install/publish to a registry.

Mask likely secrets (API keys, tokens, private keys) in stdin -> stdout.

Run the self-hostable artifact registry (serve) or moderate submissions (queue/approve/reject/rekey).

Un-track a project from polaris's registry (in-repo commandem.toml kept).

Rename a registered project (roadmap + comments preserved; groups + ack/snooze state migrated).

Show or edit a project's roadmap: its Campaigns and the Missions that support them.

Scan a directory tree for git repos and register them in bulk.

Filter-defined project groups: named expressions over snapshot fields (dirty, attention, tag:client, ahead>2) every shell renders as a sidebar section.

Edit a project's tags in the registry index.

List all tags in use across registered projects with counts.

Full-palette theme engine: import a VS Code theme .json; the app applies it.

Scaffold .vscode/tasks.json with polaris/claude tasks.

Durable per-agent git worktrees: concurrent agents WITHOUT long-lived branches (`land` returns each branch to trunk; `prune` removes idle ones).

Purpose-scoped workspaces: while one is ACTIVE, every surface (status/sidebar/stream/menubar) shows ONLY its projects -- a scoped instance without a separate registry. `--all-workspaces` (any verb) bypasses. See also: group, smartgroup.

Versioned triggers, schedules, effects, quota posture, and exact-revision history.

HTTP control surface (loopback-only, bearer-token, read-only verbs).

Attach ANY MCP server to a project (`mcp add|ls|rm <project>`), search the official MCP registry (`mcp search <query>`), or run commandem's own MCP server over stdio (bare; --install-desktop wires Claude Desktop).

Run the Outpost: this binary headless on your own always-on box (VPS/home server) -- hosts web + runner, clones repos, BYO subscription. Works with your Mac closed.

Watch the attention bucket and send macOS notifications.

Pair a phone with the web surface: print a QR + one-time code that mints a per-device revocable token.

List installed plugins, or browse/install/publish community plugins via a registry.

Post attention-bucket changes to Slack via Incoming Webhook.

Sync status to notion/linear, or inspect provider-neutral planning and collaboration sync.

Sync commandem state across machines via iCloud Drive.

Declarative reactions to attention events (notify/slack/enqueue; dry-run unless --watch).

Serve a mobile-responsive web dashboard for project status and dispatch.

Receive GitHub webhook POSTs and invalidate project cache on events.

A registered working directory plus its manifest, roadmap and local state. Registration is not repository publication.

A named project grouping/scope filter. It does not save a split-pane layout or copy a chat.

An agent-isolated Git working tree; reviewed changes land into the primary integration checkout.

A stored layout/pane configuration, optionally selected as a project default. Not an archive of conversation messages.

Reusable provider/account-related configuration. A native account owns its authentication/session root; an endpoint profile is a different configuration.

Reusable routing, role, account-reference and behavior choices. Credentials stay in provider-owned or protected local storage.

A grouping of related outcomes and Missions. It is not a provider-specific chat session.

A durable identified unit of roadmap work, with executable tasks, dependency state and completion evidence.

A named check whose observed evidence constrains an action. A passing check proves only its tested contract.

A durable record of a run's identity, result and available verification evidence; not a guarantee of successful implementation.

Reviewed context delivery to a destination conversation. Cross-provider handoff is different from native resume.

Provenance/accounting for included, excluded and truncated sources; not a model attention trace.

The conversation relationship/context surface. Deterministic retrieval is not a trained graph neural network.

A learned graph model requiring qualified data, topology and held-out evaluation. A training pipeline alone does not qualify a deployed model.

A local retention/session posture with provider-specific limits, separate from provider training policy.

Evidence or a suggestion from another agent, not an owner-authorized instruction or tool approval.

Provider-reported quota, observed exhaustion and local cost/token estimates are separate evidence types. Unknown is not zero.

A transport, dependency, host or harness failure. It is not an adjudicated model result and must be reported separately.