@audienti/cli 0.1.144 → 0.2.0-exora.1

package/README.md

@@ -1,917 +1,45 @@
-# Audienti CLI
+# Audienti CLI for Exora
 
-Audienti CLI is the agent-first command-line client for the Audienti production
-API and the local bridge for Audienti's app-hosted MCP endpoint. It lets local
-coding agents and operators inspect accounts,
-create and manage plays, import prospects, build lists, manage task reminders,
-configure list routing rules and account-user automation safety, and work
-supported operator flows.
+Use Node 24. This release connects agents to **https://exo.audienti.com**.
+The `exora` npm tag selects this client; the legacy `latest` tag is unchanged.
 
-The Motion Approach guides post-accept planning and writing. Set it with `audienti motions update motn_123 --approach "Test the premise before asking for a meeting"` or the `approach` field in a create/update payload. A nonblank Approach automatically chooses adaptive planning; blank, nil, or whitespace uses the existing sequence. `--approach ""` explicitly clears it and restores the existing sequence. There is no separate planning-mode selector. Existing sending controls still apply.
-
-`audienti operator queue` and `operator next` show planner and writer questions with candidate options. Answer one with `audienti operator answer <row_id> --choice <id>` or `--answer "Your guidance"`. The command fetches that exact row even when it is not the next move, then submits its current fingerprints. Stale or double answers return 409; an accepted answer returns 202 planning. Answers guide only the current decision. Planning and preview never authorize sending.
-
-## Install
-
-Requires Node.js 20 or newer.
-
-```bash
-curl -fsSL https://cli.audienti.com/install | bash
-```
-
-At a terminal, the installer runs `audienti start` for you. Or install directly
-through npm, then start:
-
-```bash
-npm install --global @audienti/cli
-audienti start
-```
-
-For one-off use, run `npx @audienti/cli --help`.
-
-## First run
-
-```bash
-audienti start
-audienti setup
-```
-
-`audienti start` walks you through the same path as the web:
-
-1. It prints the intro: what Audienti does and how it warms people up.
-2. It opens Audienti in your browser so you can sign in or create an account,
-   then picks your account and account user (automatically when there is only
-   one).
-3. If the account has not paid, it shows the price and credits. Press Enter to
-   pay by card on the pay page (it opens in your browser and the address is
-   printed; the terminal waits until the payment shows), or type a signup code.
-4. It runs setup in the terminal from the step you are on. Step 1, your
-   company: website, who you sell to, what they should say yes to, and your
-   city, state or province, and country when the account needs them. Step 2,
-   verify your targeting: the audience, offer and signals cards. Type `a`, `o`
-   or `s` to change one card, or press Enter to create the experiment.
-5. Step 3, go live: connecting LinkedIn is not in the terminal yet. The command
-   prints the web page where you finish it.
-
-`audienti setup` runs steps 1 and 2 on their own. Nothing is sent without your
-approval.
-
-Without a terminal, `audienti start` never prompts and reports what to do next
-(`payment_required` with the pay page when the account has not paid), and
-`setup` takes flags:
-
-```bash
-audienti setup --url https://acme.com --sell-to "<who you sell to>" --ask "<what they say yes to>" --yes --json
-```
-
-Add `--city <city> --state <code> --country <code>` when setup says it needs
-your place.
-
-Without `--yes`, a non-interactive setup drafts but creates nothing. Running
-`audienti` with no saved login starts `audienti start`; `audienti --help` still
-prints help.
-
-## Authenticate
-
-Authenticate through the browser, then configure the default account context:
-
-```bash
-audienti auth login
-audienti accounts list --json
-audienti accounts create --name Doppel --parent OMALab --json
-audienti accounts select <acct_id>
-audienti users list
-audienti users select <account_user_id|email|name|me>
-```
-
-`audienti auth login` opens Audienti in your browser, creates an API token from
-your signed-in web session, sends it back to a temporary `127.0.0.1` callback,
-and stores it in the local CLI config after validating it.
-
-If you already have a token, you can still configure it directly:
-
-```bash
-audienti auth token <token>
-```
-
-The CLI writes its local configuration to `~/.config/audienti/config.json` with
-owner-only permissions. Do not place a production token in an agent prompt,
-repository file, issue, or CI secret.
-
-The selected account user is used when commands accept `me` or default to the
-current operator, such as `audienti users activity`, `audienti analytics users`,
-and `audienti prospects assign --assigned-user me`.
-
-## Agent Workflows
-
-Before creating, changing or judging an experiment, read the strategy:
-
-```bash
-audienti help methodology
-```
-
-The guide explains premises, signals, informants, validating and gifting,
-cadence, mature-cohort verdicts, and accepted opportunities. Motion mutations
-print a short reminder to stderr; `--json` stdout remains machine-readable.
-`experiments`, `experiment` and `plays` are aliases for `motions`, including help.
-
-For gift research that stays outside the Audienti app:
-
-```bash
-audienti tools gift-research --url https://example.com
-audienti tools gift-research --url https://example.com --json
-```
-
-This prints a research skill and the website for the agent to investigate with
-its own browsing tools. It does not crawl the site, require login, scan through
-Audienti, save gifts or send outreach. The agent returns evidence-linked existing
-assets and clearly marked adaptations that someone would still need to build.
-
-Discover the bundled skills and the Audienti marketplace snapshot:
-
-```bash
-audienti skills list
-audienti skills show audienti-gift-research
-audienti skills show signal-prospect-research
-```
-
-Bundled skills print their instructions. Marketplace entries print source links
-and host-specific install commands; the CLI does not install or execute them.
-Upstream repositories own those skills, dependencies and updates. The snapshot's
-source commit is included in `skills list --json`; check the upstream catalog
-for later additions.
-
-The selected marketplace entries are the six social-network finders,
-`signal-prospect-research`, and `sales-sheet-builder`. Before using a social
-finder, agents must check their local browser/search tools, connectors, skills,
-CLIs and network access. `skills show <finder>` includes that environment check
-before the upstream instructions; JSON discovery includes it too. Missing access
-must be reported rather than assumed.
-
-Start with the built-in, production-safe workflow guide:
-
-```bash
-audienti help agent-workflows
-```
-
-Use `--json` whenever another program or agent will consume the result. Inspect
-the target state before mutations, and use the command-specific help before
-creating, changing, or deleting data.
-
-## MCP
-
-Audienti serves MCP from the app at `/mcp`. The package ships a small stdio
-bridge for MCP hosts that need a local command:
-
-```bash
-audienti-mcp
-```
-
-Use it this way:
-
-1. Install the package.
-2. Run `audienti auth login`.
-3. Run `audienti accounts select <acct_id>`.
-4. Configure your MCP host with the packaged `.mcp.json`, or point it at the
-   `audienti-mcp` command.
-5. In the MCP host, list tools and call the account-scoped tool you need.
-
-The bridge reads the same `~/.config/audienti/config.json` as the CLI, forwards
-MCP JSON-RPC messages to the app, and injects the selected account when a tool
-call does not include `account_id`. If you want a tool to run against a
-different account, pass `account_id` in that tool call.
-
-MCP clients that can call HTTP directly may use the hosted endpoint instead of
-the stdio bridge:
-
-```bash
-curl https://app.audienti.com/mcp \
-  -H "Authorization: Bearer $AUDIENTI_API_TOKEN" \
-  -H "Content-Type: application/json" \
-  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
-```
-
-Example tool call:
-
-```json
-{
-  "jsonrpc": "2.0",
-  "id": 2,
-  "method": "tools/call",
-  "params": {
-    "name": "analytics.stages",
-    "arguments": {
-      "account_id": "acct_...",
-      "query": {
-        "interval": "weekly",
-        "cohort_start_date": "2026-07-01",
-        "cohort_end_date": "2026-07-31"
-      }
-    }
-  }
-}
-```
-
-Useful MCP tools to start with:
-
-- `auth.me` confirms the token identity.
-- `accounts.list` shows accessible accounts.
-- `setup.play_preflight` returns LinkedIn connected-account readiness and setup
-  or mapping URLs.
-- `offers.create`, `icps.create`, and `motions.create` set up the offer, ICP,
-  and play.
-- `analytics.stages` returns stage conversion cohorts and stage aging.
-- `analytics.cohort_lists.create` materializes event cohorts as reusable lists.
-
-Common inspection commands:
-
-```bash
-audienti update check
-audienti operator next --plan
-audienti writer test-run <prsp_id>
-audienti analytics motions --json
-audienti analytics icps --json
-audienti motions analytics <motn_id>
-audienti motions run-discovery <motn_id>
-audienti motions runs <motn_id> [--cursor <id>] [--limit <1-10>] [--json]
-audienti motions quick-start --url https://example.com --wait --confirm
-audienti motions abm-companies <motn_id> add --file abm-domains.txt
-audienti motions abm-companies <motn_id> list
-audienti motions update <motn_id> --status closing
-audienti motions update <motn_id> --status paused
-audienti motions update <motn_id> --start-date 2026-09-01 --end-date 2026-09-30 --maximum-company-count 25
-audienti motions update <motn_id> --own-post-engagement true
-audienti motions update <motn_id> --payload motion-principal-list.json
-audienti motions update <motn_id> --payload motion-signals.json
-audienti motions activate <motn_id>
-audienti motions delete <motn_id> --confirm yes
-audienti content programs
-audienti content track <linkedin_post_url>
-audienti content posts
-audienti content engagement <cpwi_id> --json
-audienti content plan <cprg_id>
-audienti content plans --user me
-audienti content plan <rprt_id> --report --user me
-audienti content plan-update <rprt_id> --day 1 --payload edits.json
-audienti content plan-approve <rprt_id> --day 1
-audienti content plan-research <rprt_id> --day 1
-audienti content plan-draft <rprt_id> --day 1 --feedback "Revise the opening"
-audienti content plan-visuals <rprt_id> --day 1 --style <id>
-audienti content post-reply <cpwi_id> --comment <task_id> --body "Approved reply"
-audienti content approve <cpwi_id>
-audienti content comments
-audienti social-cookies sync-messages <scok_id> [--folder <folder>] [--retry] --json
-audienti prospects show <prsp_id> --json
-audienti prospects list --profiles
-audienti prospects list --assigned-user unassigned
-audienti prospects check --all --csv
-audienti prospects assign <prsp_id> --assigned-user me
-audienti prospects set-status <prsp_id> --status not_fit
-audienti prospects lock <prsp_id> --note "Emergency hold"
-audienti tasks list
-audienti tasks add --title "Review Cristina" --due 2026-07-24T11:00 --prospect <prsp_id> --notes "Check the renewal note before replying."
-audienti tasks complete <ptsk_id>
-audienti dnc list
-audienti company-rules list
-audienti lists routing-rules <list_id> list
-audienti lists routing-rules <list_id> apply
-audienti users activity --window 7d
-audienti users automation show 136 --platform linkedin --json --account <acct_id>
-audienti users automation update 136 --payload automation-controls.json --json --account <acct_id>
-audienti users automation update 136 --payload automation-controls.json --apply --json --account <acct_id>
-audienti analytics prospects --window 24h
-audienti analytics dashboard --play-tag wine_campaign
-audienti analytics metrics --cohort-preset week-to-date
-audienti analytics cohorts create-list --name "Blank note test" --start 2026-07-20 --end 2026-07-20 --note-mode blank
-audienti analytics users --user me --window 30d
-audienti analytics visibility --window 24h --user me
-audienti analytics content --window week
-audienti tools list
-audienti tools humanize --file draft.txt --tone professional --language English
-audienti tools linkedin-review --url https://www.linkedin.com/in/example --icp <icp_id>
-audienti tools linkedin-review reports
-audienti tools linkedin-review show <rprt_id>
-audienti tools linkedin-review status <rprt_id>
-```
-
-To humanize arbitrary text without exposing the underlying provider key, pass a
-UTF-8 file. Plain output contains only the transformed text, so it can be
-redirected directly to another file:
-
-```bash
-audienti tools humanize --file draft.txt --tone professional > revised.txt
-audienti tools humanize --file draft.txt --language English --json
-```
-
-### Standalone account tools
-
-These run without creating prospects or experiments. Each command returns a run
-id at once and keeps working on the server; `--wait` polls until the run
-finishes and prints the results. If the CLI exits, read the run later with
-`tools runs show|results`. Details expire after 90 days; the summary stays.
-
-```bash
-audienti tools email-find --first-name Ada --last-name Lovelace --company-domain example.com --wait
-audienti tools email-find --file people.csv --request-key batch-2026-10-03   # csv, jsonl or json
-audienti tools linkedin-enrich --url https://www.linkedin.com/in/example --wait --json
-audienti tools signals-find --icp "Mid-size US logistics companies" --question "Who announced warehouse automation?" --count 5
-audienti tools write --purpose "Book a call" --audience "Finance leads" --fact "Approvals take 2 days" --channel email
-audienti tools humanize --file draft.txt --async --wait
-audienti tools runs list --tool email-find
-audienti tools runs results <trun_id> --json
-audienti tools runs export <trun_id> --format csv --output results.csv
-```
-
-`--file` takes JSON (an array or `{"items": [...]}`), JSONL, or CSV with a
-header row. CSV is one row per line; quoted commas are allowed, line breaks
-inside fields are not. With `--wait`, the exit code is 1 on an API error, a
-timeout or an aborted (failed) run; per-item outcomes such as `no_match` do
-not change it.
-
-For active paid Email Find or Signal search work, review a server quote first:
-
-```bash
-audienti tools email-find --file people.csv --quote
-audienti tools signals-find --icp "Logistics companies" --question "Warehouse expansion" --count 5 --quote
-audienti tools email-find --file people.csv --accept-quote <quote_reference>
-audienti tools runs review <run_id>
-audienti tools runs resume <run_id> --item <item_id> --revision <revision> --idempotency-key <unique_key>
-```
-
-Use the same input when accepting the quote. Queued rows retain the accepted price.
-Review reports available credits and the rows that can continue. Select fewer rows
-or top up; a top-up does not restart manually paused work. Use `pause` or `cancel`
-in place of `resume` with the same reviewed selection. `approve-payer` requires the
-new root administrator and does not resume work. `tools runs usage <usage_reference>`
-opens the exact receipt data. Tool history displays credits; supplier costs stay internal.
-Real prices and production charging require separate approval.
-
-Prefer `tools email-find` for an email lookup. `tools get email` still runs the
-prospect import pipeline and creates a prospect.
-
-### Saved network
-
-Your own connected account's saved connections and following, from what
-normal work has already seen. Only the account's owner can read it, and it
-never visits LinkedIn or X. Coverage says partial, not yet observed, or
-capture not supported yet; none of these mean zero.
-
-```bash
-audienti network list --cookie <scok_id> --kind connection --query acme
-audienti network list --cookie <scok_id> --kind follow --direction outgoing --json
-audienti network export --cookie <scok_id> --kind connection --format csv --output connections.csv
-```
-
-To let an agent or operator check whether the local CLI is behind the latest
-published package:
-
-```bash
-audienti update check --json
-```
-
-To work the supported prospect operator queue from the CLI, inspect the next move
-and record the outcome against that same row:
-
-```bash
-audienti operator next --plan
-audienti operator next --done --note "Connection request sent."
-```
-
-`operator queue`, `operator failed-drafts`, `network-ops queue`, and
-`inbox-ops queue` print a
-continuation command when more rows remain. Read-only `operator next` (including
-`--plan`) also prints one when its page has no focal move but more rows remain;
-when a move is present, inspect or act on it and rerun `operator next`.
-Follow the continuation command even when the current page has no ready
-rows. It preserves the account and filters and uses the API's exact `--page` and
-`--offset` (which can be zero), or its opaque `--cursor`. Do not combine a cursor
-with an offset. If the scan limit is reached, narrow the filters; an incomplete
-scan does not mean the queue is empty. JSON output preserves the API response.
-Pagination flags are not accepted when recording outcomes or requeueing drafts.
-
-Network Ops has its own owner-scoped CLI surface for pending inbound LinkedIn
-connection requests. It shows the already-synced LinkedIn profile headline when
-one is present, keeps the exact invitation-bound message when one was captured,
-and queues the same canonical provider actions as the product UI:
-
-```bash
-audienti network-ops queue [--page <n>] [--offset <n>|--cursor <token>]
-audienti network-ops accept <row_id>
-audienti network-ops decline <row_id>
-```
-
-`reject` is accepted as an alias for `decline`. An accepted API response means
-the provider action was queued; it is not confirmation that LinkedIn completed
-the action. The same queue can be read with
-`audienti operator queue --opportunity-kind network`.
-
-Inbox Ops has its own owner-scoped CLI surface built for clearing a backlog in
-bulk. `inbox-ops queue` follows every page, numbers each row, and saves the
-numbered list locally for the selected account. The bulk verbs select rows by
-number, range, row id, `--domain`, or `--sender`, print a manifest, and apply
-only with `--yes` (or an interactive confirmation). `--dry-run` validates every
-row on the server and reports planned, skipped, and rejected rows without
-changing anything:
-
-```bash
-audienti inbox-ops queue
-audienti inbox-ops queue --group-by domain
-audienti inbox-ops ignore 1-25 --dry-run
-audienti inbox-ops ignore 1-25 --yes
-audienti inbox-ops filter-domain 3,7 --yes
-audienti inbox-ops filter-domain --domain alerts.example.com --yes
-audienti inbox-ops filter-sender 12 --yes
-audienti inbox-ops allow-sender 4 --yes
-audienti inbox-ops allow-domain --domain customer.example --yes
-audienti inbox-ops filters
-audienti inbox-ops rule <row_id> --scope sender --disposition filter
-audienti inbox-ops rule set --scope sender --key news@example.com --disposition filter
-audienti inbox-ops rule remove --scope domain --key example.com
-```
-
-Numbers stay valid until `inbox-ops queue` runs again. `ignore` hides one
-thread until a newer reply arrives; `filter-*` and `allow-*` write durable rules
-that apply to every current and future matching message across the owner's
-accounts. Rule verbs given only `--domain` or `--sender` write the keyed rule
-directly, even when the matching rows are no longer listed. Every row is
-authorized individually on the server: rows another owner controls, stale rows,
-and malformed ids are rejected one by one while the rest still apply. Pass
-`--page`, `--offset`, or `--cursor` to read exactly one page instead.
-
-All Inbox Ops commands update only the authenticated owner's personal/global
-Inbox Ops preferences or private ignore receipts. They do not archive mail, call
-the provider, or add a DNC entry.
-
-To manage simple reminders for yourself:
-
-```bash
-audienti tasks list
-audienti tasks list --status completed
-audienti tasks add --title "Review target account" --due 2026-07-24T11:00 --list <list_id> --notes "Check the latest notes."
-audienti tasks complete <ptsk_id>
-```
-
-To manage the same ordered rules shown in a list's Routing Rules UI, use a JSON
-payload for create and update operations:
-
-```bash
-audienti lists routing-rules <list_id> list
-audienti lists routing-rules <list_id> create --payload routing-rule.json
-audienti lists routing-rules <list_id> update <rule_id> --payload routing-rule.json
-audienti lists routing-rules <list_id> move <rule_id> up
-audienti lists routing-rules <list_id> remove <rule_id>
-audienti lists routing-rules <list_id> apply
-```
-
-```json
-{
-  "name": "VP prospects to Alice",
-  "enabled": true,
-  "action_kind": "assign_user",
-  "target_account_user_id": "alice@example.com",
-  "conditions": {
-    "seniorities": [
-      {"id": "5", "name": "Vice President", "negative": false}
-    ],
-    "seniority_match_mode": "at_least"
-  }
-}
-```
-
-Use `action_kind: "route_to_list"` with `target_list_id` to route to another
-working list. `apply` queues the existing list-routing background job, just like
-the UI; it reports that work was enqueued, not that every prospect has finished
-routing. Location and company_location labels must resolve to a known geography
-at save time; unknown labels return 422. Run `audienti lists routing-rules help`
-for the full condition contract.
-
-To inspect activity for prospects that entered the account during a specific
-cohort while keeping a separate activity window:
-
-```bash
-audienti analytics prospects --cohort-start 2026-07-01 --cohort-end 2026-07-07 --window 7d
-```
-
-To compare recent weekly prospect cohorts by their current pipeline stages:
-
-```bash
-audienti analytics prospects cohort-analysis --weeks 4 --motion <motn_id>
+```sh
+npm install -g @audienti/cli@exora
+export AUDIENTI_API_URL=https://exo.audienti.com/api/v1
+audienti agent connect --name "My agent" --profile "$HOME/.audienti/exo-agent.json"
 ```
 
-To inspect the current outcomes of outbound operations created in a date range,
-use the account-scoped metrics report:
+Open the returned approval URL yourself, sign in, check the code, and approve.
+Give the agent the profile path, not your password or a human API token.
 
-```bash
-audienti analytics metrics --cohort-start 2026-08-24 --cohort-end 2026-08-30 --interval daily
-audienti analytics metrics --cohort-start 2026-08-24 --cohort-end 2026-08-30 --interval weekly
-audienti analytics metrics --cohort-preset week-to-date --user me --motion <motn_id> --play-tag <tag> --list <list_id> --offer <offr_id> --icp <icp_id> --social-cookie <scok_id> --platform linkedin --action messaging.message_sent --outcome succeeded_after_retry
-audienti analytics metrics --cohort-preset week-to-date --json
+```sh
+audienti agent poll --profile "$HOME/.audienti/exo-agent.json"
+audienti workspace list --profile "$HOME/.audienti/exo-agent.json"
+audienti workspace use WORKSPACE_UUID --profile "$HOME/.audienti/exo-agent.json"
+audienti tools balance --profile "$HOME/.audienti/exo-agent.json"
+audienti profiles list --profile "$HOME/.audienti/exo-agent.json"
+audienti skills show audienti
 ```
 
-This is an `events.created_at` root-operation cohort, and each operation is
-classified by its current outcome across all of its attempts. It is distinct
-from the `AccountProspect.created_at` entry cohorts used by dashboard, stages,
-and prospect cohort analysis. The command
-makes one authenticated request to the account Metrics API. Plain text formats
-only the returned cohort ranges, Operations, First-try success, Recovered, Final
-failures, In progress, Attempts, selected daily or weekly grid rows, action rows,
-and account-wide Social Cookie rows. `--json` prints the API response deeply
-unchanged; the CLI does not recalculate cohorts, outcomes, counts, or rates.
-
-The six leaf outcomes are `first_attempt_success`, `succeeded_after_retry`,
-`failed_without_retry`, `failed_after_retry`, `in_progress`, and `unresolved`.
-The `success` and `failure` values are umbrella filters over their corresponding
-two terminal leaf outcomes.
-
-For example, the human output uses the server-provided values directly:
-
-```text
-Outbound metrics
-Cohort: 2026-08-24 to 2026-08-30 (events.created_at)
-Current: 2026-08-24T00:00:00-05:00 to 2026-08-31T00:00:00-05:00
-Previous: 2026-08-17T00:00:00-05:00 to 2026-08-24T00:00:00-05:00
-Operations: 1347
-First-try success: 1275 (94.7%)
-Recovered: 40 (3%)
-Final failures: 16 (1.2%)
-In progress: 16 (1.2%)
-Attempts: 1400
-```
-
-To materialize a connection-request activity cohort as a reusable list selector,
-then reuse it in analytics:
-
-```bash
-audienti analytics cohorts create-list --name "Connection requests 2026-07-01 to 2026-07-07" --event connection_request_sent --start 2026-07-01 --end 2026-07-07
-audienti analytics dashboard --list <list_id>
-audienti analytics prospects --list <list_id> --window 30d
-audienti analytics users --user me --list <list_id> --start 2026-07-01 --end 2026-07-31
-```
-
-For prospect, dashboard, and stages analytics, use `--cohort-start` and
-`--cohort-end` when the cohort is based on when people entered Audienti, a
-motion, or a receiving segment. Use `analytics cohorts
-create-list` followed by `--list` when the cohort is based on an event that
-happened later, such as connection requests sent in a date range. Rebuild the
-list when the event definition or date window changes so the analytics question
-stays auditable.
-
-`audienti setup play preflight` includes the selected sender's server-derived
-working schedule under `social_cookie.automation`, including `active_days`,
-normalized `working_hours`, effective `time_zone`, and current
-`in_working_hours` state. The human-readable output reports the same timezone,
-today's window, and whether provider execution is currently inside the window.
-It also reports the configured user fallback location, the connected account's
-configured and last-verified effective proxy geography, and the proxy source
-without exposing proxy URLs, credentials, egress IPs, or browser session data.
-For authorized LinkedIn account owners and admins, `social_cookie.capabilities.linkedin`
-reports Audienti's stored `premium` and `sales_navigator` booleans plus the
-`checked_at` timestamp. Text output shows yes, no, or unknown and when detection
-last ran. Null or missing values mean unknown, including older server responses
-and unavailable access. These are stored observations that may be stale;
-preflight does not perform a fresh provider check. `--json` preserves the server
-payload unchanged.
-For LinkedIn, `social_cookie.automation.pacing` includes effective weekly
-quotas, daily targets, motion active days, the outstanding-invitation cap, ramp
-configuration, current outstanding inventory, any inventory blocker, and current
-invitation capacity calculated by the server. A `null` weekly quota means
-unlimited.
-
-For an email social cookie, `social-cookies sync-messages` requests the shared
-account-scoped cookie/folder sync slot and prints the server's durable status.
-Pass `--folder` to select a provider folder and `--retry` only when an explicit
-retry is intended. The command does not call a provider directly; `--json`
-returns the accepted request and `sync_states` payload unchanged.
+Approval initially grants read access. For CRM writes or paid tools, grant the agent
+workspace access on your Agents page. Workspace policy and current permissions still
+apply. Only a verified human Admin can purchase or redeem credits through Billing.
+Paid calls spend shared workspace credits; development credits are local-only.
 
-Use `users automation` to inspect or change one account user's LinkedIn safety
-policy without changing the saved account selection:
-
-```bash
-audienti users automation show 136 --platform linkedin --json --account acct_example
-audienti users automation update 136 --payload automation-controls.json --json --account acct_example
-audienti users automation update 136 --payload automation-controls.json --apply --json --account acct_example
-```
-
-`update` is preview-only unless `--apply` is present. The CLI always sends
-`platform: "linkedin"` and an explicit `apply` boolean after the file fields,
-so a payload file cannot silently opt into applying a change. The server scopes
-the connected account through the selected account plus account user, preserves
-omitted controls, returns the before and proposed/actual after states, and
-records an audit event only for an applied update. `--json` prints the server
-payload unchanged.
-
-Follow (including unfollow) and invitation withdrawal are enabled by default. Set
-`follow_autopilot_enabled` or `withdraw_connection_autopilot_enabled` to `false`
-to disable that action. These preferences still respect automatic sending and
-other execution limits; follows also require visibility automation.
-
-An `automation-controls.json` warm-up payload can use every automation gate,
-independent action limits, an aggregate visibility limit, and a weekly ramp:
-
-```json
-{
-  "automation_controls": {
-    "automatic_sending_enabled": false,
-    "visibility_operations_autopilot_enabled": true,
-    "follow_autopilot_enabled": true,
-    "withdraw_connection_autopilot_enabled": true,
-    "post_comment_autopilot_enabled": false,
-    "connection_request_autopilot_enabled": false,
-    "inmail_autopilot_enabled": false,
-    "direct_message_autopilot_enabled": false,
-    "email_sending_autopilot_enabled": false,
-    "manual_action_handoff_enabled": false,
-    "risk_cooldown_enabled": true
-  },
-  "action_limits": {
-    "profile_view": { "hourly": 2, "daily": 4, "weekly": 20 },
-    "follow": { "hourly": 2, "daily": 3, "weekly": 15 },
-    "like": { "hourly": 1, "daily": 2, "weekly": 10 },
-    "invite": { "hourly": 1, "daily": 2, "weekly": 10 },
-    "message": { "hourly": 1, "daily": 2, "weekly": 10 },
-    "comment": { "hourly": 1, "daily": 1, "weekly": 5 },
-    "visibility": { "hourly": 3, "daily": 8, "weekly": 40 }
-  },
-  "visibility_ramp": {
-    "enabled": true,
-    "starting_daily_limit": 8,
-    "weekly_increment": 2
-  }
-}
-```
-
-`follow` covers follows and unfollows; `like` covers likes and unlikes;
-`message` covers messages and InMail. `visibility` is the aggregate cap across
-profile views, follows/unfollows, and likes/unlikes. The show/preview readback
-includes configured, default, effective, used, and remaining hourly, daily,
-and weekly values plus the currently binding limits.
-
-Motion update payload mode can replace the selected principal and backing list:
+To connect an MCP client, use this command and environment after selecting a workspace:
 
 ```json
 {
-  "principal_account_user_id": 136,
-  "list_id": "list_abc123"
+  "command": "audienti-mcp",
+  "env": { "AUDIENTI_PROFILE": "/absolute/path/to/.audienti/exo-agent.json" }
 }
 ```
 
-The principal and list are resolved inside the selected account. Use
-`"list_id": null` to clear the backing list.
-
-To inspect the account's current prospect mix by motion type and compare it
-with the rolling seven-day recorded-source contribution:
-
-```bash
-audienti analytics motions
-audienti analytics motions --json
-```
-
-Current Unattributed counts mean no current motion association. Rolling
-seven-day Unattributed counts mean no attributable recorded source motion; the
-server falls back to the current motion only when recorded source data is
-blank.
-
-To inspect the account's current prospect mix by source ICP and compare it
-with the rolling seven-day source contribution:
-
-```bash
-audienti analytics icps
-audienti analytics icps --json
-```
-
-Unattributed means no attributable account source ICP. The server falls back
-to the current motion ICP only when recorded ICP source data is blank.
-
-To see whether one motion is producing prospects by day, and where each
-produced-day cohort currently sits in the funnel:
-
-```bash
-audienti motions analytics <motn_id>
-```
-
-To count the people and distinct companies currently targeted by one campaign
-tag without waiting on the web dashboard filter:
-
-```bash
-audienti analytics dashboard --play-tag wine_campaign --cohort-start 2026-07-01 --cohort-end 2026-07-07
-```
-
-To queue an immediate discovery run for a motion through the same launch gate
-used by the operator surface:
-
-```bash
-audienti motions run-discovery <motn_id>
-```
-
-Outbound, inbound, and LOPA motions are accepted only when their current
-configuration and discovery producer are executable. An accepted response
-includes a durable `run` receipt that starts in `pending` before the discovery
-job is queued. Rejected launches exit nonzero and report the authoritative
-`reason`, optional `next_eligible_at`, and `suggested_action`; `--json` preserves
-that structured response on stdout for automation. Completed receipts keep
-candidate `submitted_count` separate from the current persisted
-`promoted_count`, and expose completed, failed, and blocked scope counts.
-
-To check whether a motion has executable configuration before preparing,
-activating, or running discovery:
-
-```bash
-audienti motions status <motn_id>
-audienti motions status <motn_id> --json
-```
-
-The JSON response includes `executable_configuration.valid` plus a reason when
-the config is incomplete, such as missing outbound signals or missing LOPA
-tracked profile URLs. Its latest discovery receipt uses the same submitted,
-promoted, and scope outcome fields as `run-discovery`.
-
-To create and launch a quick-start motion from a company URL:
-
-```bash
-audienti motions quick-start --url https://example.com --wait --confirm
-```
-
-To create or update an ICP from a rich JSON payload:
-
-```bash
-audienti icps create --payload icp.json
-audienti icps update <icp_id> --payload icp-patch.json
-audienti icps list --status all
-audienti icps archive <icp_id>
-audienti icps restore <icp_id>
-```
-
-`icps update --payload` accepts the same human-readable facet keys as create.
-Supplied facet collections, such as `company_sizes_attributes`, replace that
-collection in place; omitted fields and collections remain unchanged. Invalid
-lookup values fail server-side without partially applying the patch.
-Archive is reversible and is the normal cleanup action. It removes the ICP from
-new selection, closes active primary motions so admitted work can drain,
-archives inactive primary motions, and preserves secondary motion links.
-Restoring the ICP does not reactivate motions.
-
-To audit one account user's outbound actions, optionally narrowed to one motion
-and one AccountProspect.created_at cohort:
-
-```bash
-audienti analytics users --user me --start 2026-07-01 --end 2026-07-07 --cohort-start 2026-06-01 --cohort-end 2026-06-30 --motion <motn_id>
-```
-
-To start a report-backed writer session for one prospect, including the no-reply
-path, planned actions, channel changes, and planned message rows:
-
-```bash
-audienti writer test-run <prsp_id>
-```
-
-The output includes a `Report: rprt_...` id. Treat that report id as the writing
-session handle while the generated report is retained.
-
-To draft every message into that report-backed session, opt into full report
-mode:
-
-```bash
-audienti writer test-run <prsp_id> --mode report --report <rprt_id>
-```
-
-For writer debugging, draft only one selected timeline row on one branch into
-the same report:
-
-```bash
-audienti writer test-run <prsp_id> --mode step --branch no-accept --step 3 --report <rprt_id>
-```
-
-The CLI queues the report job on the API and polls until the report finishes, so
-slower writer calls do not depend on a single long HTTP request. Use
-`--timeout-seconds <n>` to adjust the CLI wait budget.
-
-To queue work into the report and come back later:
-
-```bash
-audienti writer test-run <prsp_id> --mode step --branch no-accept --step 3 --report <rprt_id> --no-wait
-```
-
-If the CLI stops waiting before the server job finishes, fetch the completed
-report later with:
-
-```bash
-audienti writer test-run show <prsp_id> <rprt_id>
-```
-
-To update a prospect's attached profile channels through the same paths used by
-the prospect show page:
-
-```bash
-audienti prospects add-profile <prsp_id> --url prospect@example.com
-audienti prospects add-profile <prsp_id> --url +12025550123
-audienti prospects add-profile <prsp_id> --url https://www.linkedin.com/in/example
-audienti prospects report-bad-profile <prsp_id> <prof_id>
-```
-
-To queue a personal LinkedIn profile authority review and ICP-fit positioning
-blueprint, then check whether it is waiting on enrichment, running, completed,
-or failed:
-
-```bash
-audienti tools list
-audienti tools linkedin-review --url https://www.linkedin.com/in/example --icp <icp_id>
-audienti tools linkedin-review reports
-audienti tools linkedin-review show <rprt_id>
-audienti tools linkedin-review status <rprt_id>
-```
-
-To reassign or clear ownership for existing prospects:
-
-```bash
-audienti prospects assign <prsp_id> --assigned-user <account_user_id|me>
-audienti prospects assign <prsp_id> --assigned-user unassign
-```
-
-To preview moving a prospect and its account-scoped history into another
-account where you are also an administrator:
-
-```bash
-audienti prospects move-account <prsp_id> --account <source_acct_id> --target-account <target_acct_id>
-audienti prospects move-account <prsp_id> --account <source_acct_id> --target-account <target_acct_id> --assigned-user me --target-motion <motn_id> --target-list <list_id> --apply
-```
-
-The command is preview-only unless `--apply` is present. Apply always requests
-a fresh preview first and submits the returned `manifest_digest`, so a move is
-rejected when the underlying state changes. `--json` prints the preview or final
-apply response unchanged.
-
-To make emergency prospect state changes without going through a motion:
-
-```bash
-audienti prospects set-status <prsp_id> --status nurture
-audienti prospects set-status <prsp_id> --status not_fit
-audienti prospects lock <prsp_id> --note "Emergency hold"
-audienti prospects unlock <prsp_id>
-```
-
-To manage account DNC and company disposition rules from the CLI:
-
-```bash
-audienti dnc add prospect@example.com
-audienti dnc import --file dnc.txt
-audienti company-rules create --linkedin-url https://www.linkedin.com/company/example --disposition monitor
-audienti company-rules create --domain example.com --disposition not_fit --user me
-audienti company-rules apply --all
-```
-
-The direct intake action supports the same destination names and optional experiment:
-
-```bash
-audienti prospects intake --url https://www.linkedin.com/in/example --new-list "Target list" --new-transition "Fit transition" --json
-```
-
-Use `--motion <motn_id>` for an existing experiment. Explicit new names, including blank names, are sent to the API for validation.
-
-To import multiple LinkedIn people through the same per-prospect import path:
-
-```bash
-audienti prospects import-batch --file prospects.csv --new-list "Target list" --new-transition "Fit transition" --assigned-user me
-```
-
-CSV files should include a `linkedin_url` or `url` header. Optional row columns
-`list_id`, `motion_id`, `new_list_name`, `new_transition_name`, and `assigned_user_id` override command defaults. The command parses the file and sends one batch request; the API owns validation, destination creation, row outcomes, and queueing.
-
-## Compatibility
-
-The CLI talks to the versioned Audienti `/api/v1` contract at
-`https://app.audienti.com` by default. A release occurs only after the matching
-server deploy succeeds.
-
-The canonical source lives with the Audienti application under
-`packages/audienti-cli`. This public repository is a CI-managed mirror; direct
-changes are unsupported and cause the next source release to fail safely.
-
-## Plugins
-
-This repository includes Codex and Claude Code plugin manifests plus an
-`audienti` skill. The plugin provides workflow instructions; it does not grant
-credentials or silently authenticate an agent.
-
-## License
-
-Copyright (c) 2026 OMALab, Inc. All rights reserved. See [LICENSE](LICENSE).
-
-Batch import accepts up to 50 rows per API request. Split larger files before importing. Row destination IDs or new names override batch defaults; failed rows stay visible in the response.
-Bulk destinations: `audienti prospects set-destination <prsp_id> --type list --destination <list_id>` adds by default. Use `--name "New list"` to create a list, `--type experiment --name "Existing relationships"` to create a transition without setup, or `--mode move` for explicit reassignment. Add preserves existing memberships; experiment Add skips people assigned elsewhere.
-
-Create a child workspace with `audienti accounts create --name <name> --parent <name-or-id> [--json]`. The parent can be a visible name, unique name fragment, numeric ID, or prefixed ID. You must administer a root team account. All parent admins receive access to the child; ordinary members do not. Ambiguous names show matching IDs and stop. A duplicate child name under the same parent fails clearly. The response includes the new account ID. Creation keeps your selected account unchanged: use `audienti accounts select <new_acct_id>` or `--account <new_acct_id>` before creating a motion there.
-
-### Credit history and root approvals
-
-`audienti credits history` lists credit charges. `credits show <usage_id>` opens an
-exact receipt. A root billing role does not grant access to private saved results.
-`credits approvals` shows action prices and changed-payer requests. After the
-root billing admin explicitly agrees, use `credits accept-price <price_id>` or
-`credits approve-payer <operation_id> --revision <revision> --request-key <unique_key>`.
-Approval never resumes waiting work. Increased prices also require the notice period.
+Restart the MCP process after changing workspace or credentials. Tools may appear
+in the catalog before the agent has permission to invoke them. The server remains
+authoritative. Agent-executable skills are in `skills/audienti`; helpers do not call
+providers or execute hidden workflows.
 
-Omalab staff can use `credits refund-review <usage_id>` to inspect the exact charge,
-result and previous reviews. After confirming a defect, `credits refund <usage_id>
---cumulative-principal <exact_credits> --evidence-token <uuid> --reason <reason>
---evidence <exact_result_evidence>` records the cumulative credit return. Customer
-billing admins cannot use these staff actions. This never refunds cash.
+Start with saved-record reads and inspect the credit balance. Explicit enrichment
+uses the configured hosted provider; registration of an external agent-managed
+connection does not configure a hosted provider or imply successful synchronization.