Added

Team onboarding pack: an owner or admin picks shared skills and agents and MCP servers on the new Team → Onboarding tab (/v1/org/onboarding), and every member's hnl installs them into the member's enabled harnesses on the next hnl cloud sync, when the background service starts and hourly after that, so updates reach everyone. Skills and agents go where hnl add puts them (a skill or server the member made with the same name is never overwritten); MCP servers go into omp, Cursor, Codex and Claude Code. A pack never holds a secret: env and header values must be 1Password references (op://vault/item/field) or ${VAR} names (anything else answers 400 secret_value), and a server that needs a 1Password secret starts through op run. hnl team pack lists what the pack installed, hnl team pack sync applies it now and hnl team pack remove <name> removes an item from this machine; the removal is reported to the organisation and left alone, unless the item is marked required by policy (then removal is refused and a deleted copy comes back). Owners and admins see how many machines installed, removed or failed each item; who did what only with the owner's Agent inventory switch on. Free and Pro see the upgrade prompt (403 feature_gated, feature onboarding_pack).
Team budgets: owners and admins set a monthly USD budget for the whole organisation or a team on Team → Spend (/v1/org/team-budgets); its spend is the sum of the members' gateway spend this month, shown only while the owner's Spend switch is on (403 visibility_off otherwise).
Prompt library: star any prompt in History (its record view, or the star beside each prompt in the session panel) and it lands on the new Library page (nav and ⌘K "Library"), where you give it a title and tags, search and filter by tag, copy it, or unstar it. Use the same library from a terminal with hnl prompts (list, --tag, --query, --json), hnl prompts show <id> and hnl prompts copy <id>, and from any harness through the HarnessLink MCP tools prompt_library_list and prompt_library_get — ask for "my starred review prompt" and the model fetches it. The prompt text is kept when starred, so it stays even after the History prompt is archived or locked. Free keeps 5 starred prompts (starring a 6th shows the upgrade prompt); Pro, Team and Enterprise keep any number, and nobody sees another member's library. GET/POST /v1/prompts/starred, GET/PATCH/DELETE /v1/prompts/starred/:id; Download my data includes prompts/starred.jsonl.
Saved queries with alerts (Pro): saved queries on Query, History, Runs and Memory are now kept with your account instead of in one browser (views a browser saved before are uploaded the first time you open the page there). On Query, the bell beside a saved query — or "Alert me on new matches" when saving — sends you an email and a Telegram message when the query finds new matches: how many, the newest three and a link back. At most one message per query per hour and 24 a day; a batch that could not be delivered is retried, and a delivered one is never sent again. Settings → Saved queries lists them all with an alert switch each. GET/POST /v1/saved-queries, PATCH/DELETE /v1/saved-queries/:id; turning an alert on below Pro answers 403 feature_gated.
Recall that comes to you (Pro): when a session starts, hnl context-hook hands the harness up to three of your past memory entries that match it — "You fixed this in <workspace> on <date>: <title>" with the entry attached — matched on the newest prompts of the workspace, and again on the session's first prompt, matched on that prompt (GET /v1/recall; matching only, nothing generated; your own entries plus, on Team, those in projects your owner shared). The network gets 250 ms; past that the workspace's last answer is used. Turn it off on one machine with memory.recall: false, or everywhere in Settings → Profile → Suggested past fixes (Free sees the upgrade prompt there). Each suggestion counts toward "recalled N memories".
Weekly retrospective (Pro): the weekly digest email is now "Your week in HarnessLink" — prompts, sessions and active hours, where your time went per workspace (share of active hours), prompts per harness, spend by model and what the gateway saved, memories banked, words that keep recurring in your own prompts, and the next step of each work-state checkpoint still open. Every figure comes from your own synced rows and a section with nothing in it is left out; the optional 2–3 sentence summary is labelled AI-written. Overview shows the same as a "This week" card (GET /v1/digests/retro, 403 feature_gated below Pro). Members of paid Pro, Team and Enterprise organisations are subscribed by default (Mondays 08:00 UTC); turn it off in Settings → Profile → Email digests and it stays off. Free keeps the plain stats digest.
Machine restore in one command: hnl snapshot uploads this machine's harness setup — Claude Code, Codex, omp, pi and HarnessLink config files, MCP server configs, skills, agents, hooks, commands, rules and prompts — and the background service takes one snapshot a day. hnl restore --apply on a new machine writes them back exactly (--from <snapshot id|device|hostname> picks one, --force overwrites files that differ, --install-harnesses installs harnesses that are missing). Secret values never leave the machine: credential files are never read, every value in an env/headers table or under a secret-named key is replaced by [REDACTED:secret] unless it is a reference (op://…, ${VAR}), and restore lists the files where a secret needs filling. hnl snapshot --dry-run shows what would travel; hnl snapshot list lists your snapshots. Devices shows a Snapshots list with the restore command. Free keeps the latest snapshot; Pro keeps every one.

Added

Flag notifications