@indigoai-us/hq-cli 5.345.104 → 5.345.106

package/CHANGELOG.md

@@ -2,6 +2,102 @@
 
 ## [Unreleased]
 
+## [5.345.106] — 2026-10-12
+- On Windows, `hq core` converts Git Bash paths for commands that declare path operands. Direct calls use the same path rules as the former script wrappers.
+- `hq logout` now ends your session on the HQ server as well as on this
+  device, so a copied token file stops working. If the server can't be
+  reached, logout still signs you out locally and says the old session will
+  expire on its own.
+
+## [5.345.105] — 2026-10-12
+- Session-start health no longer files leftover company journals that this
+  repair pass did not refresh, including older hooks that only read
+  `hq doctor --json`. `hq doctor` still prints those rows; `--fix` still
+  refreshes a stale mhi-media journal when hook files have uncommitted
+  changes.
+- The setup bot can add a pack from the chat. Paste a pack page link
+  (`hqforwork.com/packs/<slug>`), type `hq-pack-<slug>`, a slug, or the pack's
+  name from the catalog, and it confirms in one message, runs
+  `hq packages install <slug> --start-bot`, and hands over to the pack's setup
+  bot by name. An unknown pack gets the three closest catalog names; a failed
+  install gets the reason in plain words and one next step.
+- Bots look for `claude` in the HQ app's toolchain
+  (`~/Library/Application Support/Indigo HQ/toolchain/npm-global/bin/claude`)
+  before `/usr/local/bin`, `/opt/homebrew/bin`, `/usr/bin` and PATH, and log
+  the path they chose when they start.
+- A bot turn that fails because its coding tool is missing posts a plain
+  one-line message with a Retry button on the first attempt instead of
+  retrying silently. Other failures without a repair card get the Retry button
+  too, and the fifth failure in ten minutes posts one summary (reason, next
+  step, pause) rather than a per-turn line plus a pause note.
+- Packs can declare a `setup:` block in `package.yaml`: a setup bot (handle,
+  display name, intro), ordered questions with suggested answers, the script to
+  run with the answers, a done message, and `takeover_setup`. `hq packages
+  install` validates the block and fails naming the field; packs without one
+  install unchanged. `hq packages list` shows `setup: yes|no` for each pack.
+- `hq packages install <source> --start-bot` creates the pack's setup bot as a
+  personal bot running the pack's worker, starts it, and prints `bot: <name>`.
+  An existing bot of that name is restarted and sends its intro again. If the
+  bot cannot start, the install still succeeds, prints the reason, and DMs you
+  one sentence with the next step.
+- A bot running a pack's worker gets the pack's setup instructions in its
+  system prompt: each question with tap-to-answer suggestions (Skip for
+  skippable questions), the exact command to run with the answers, and the done
+  message. When the pack asks to take over and HQ's own setup is unfinished,
+  the bot finishes HQ setup first with the setup worker's skills.
+- Installing a pack from the marketplace (a bare slug or `marketplace:<slug>`)
+  first installs any pack in its `requires.packs` that is not installed yet,
+  from the marketplace. A dependency cycle or a required pack missing from the
+  marketplace fails the install before anything is written.
+- A pack bot sends its first setup question the moment it starts, from code,
+  proposing the company's name when the question asks for it. It no longer
+  spends a model turn checking sign-in and company first.
+- Pack setup bots run lean: their Claude Code turns load none of the computer's
+  plugins, MCP servers or HQ's session, prompt and stop hooks; HQ's tool hooks
+  still run. A plain reply takes about 5 seconds instead of about 20.
+  `hq bot create --lean` does the same for any Claude bot.
+- A `setup:` question can carry a `goal`, and the block a `documents` list. The
+  bot runs the setup as an interview: it follows up once on a vague answer,
+  asks early for the listed documents, and reads them to ask less.
+- Pack setup bots learn from the person's tools before asking. Right after
+  the first answer the bot checks the company's usable HQ integrations, then
+  runs a short interview: for each kind of tool the pack lists in
+  `setup.sources` (email, CRM, team chat, docs), which one they use, as
+  buttons. It sends a sign-in link for each tool they picked, one at a time
+  (`hq integrations connect <domain> --no-browser`), reads with read-only
+  tools after a yes, and proposes answers for the person to confirm. It asks
+  only what the data left open, and names tools by kind, not brand.
+- A pack setup ends with a summary: the bot opens its last message with the
+  pack's done message, then says what it created and where, which apps it
+  linked, and what it skipped, with the done suggestion plus up to two next
+  steps as buttons.
+- `hq integrations connect --link-only` prints the sign-in link and exits at
+  once; the sign-in finishes on the HQ console, and `hq integrations list`
+  confirms it. Pack setup bots use it, so a person gets the link right away
+  instead of the bot waiting on a sign-in it has not sent yet.
+- Pack setup asks about all the person's tools in one message with an example
+  answer, and tells the bot that any text it writes is sent, so it does not
+  narrate its steps.
+- A `setup.sources` option can connect through a script the pack ships
+  (`tool`, with `connect`, `status` and `usage`) instead of an HQ integrations
+  domain, or be marked `available: false` with a note. The setup prompt spells
+  out each tool's connect, check and read commands, and the bot never tells the
+  person a listed tool cannot be connected.
+- The tools question is asked by typing, with no buttons and no example
+  answer, and buttons are reserved for real choices.
+- The Pack setup section is a brief, not a script: what the pack needs to
+  learn (topics with goals and common answers), how each of the person's
+  tools connects and is read, how buttons work, the setup command, and what
+  to report when done. The bot decides how to run the conversation and adapts
+  to the person; no fixed order, wording or forced buttons.
+- `hq integrations slack connect [--company <slug>] [--open] [--json]` gets Slack's own sign-in link for your account (the call the console's Add Integration dialog makes) and prints or opens it, so a bot or script can connect Slack without the console dialog. `hq integrations slack accounts` lists the connected accounts.
+- The setup prompt treats a pack's tool list as examples, not a limit. For any other tool the bot works out a connection itself: HQ's catalog and a sign-in link, an access key entered through an HQ secrets link (never the chat), a docs page or MCP server, and then a workaround such as an export dropped into the chat.
+- HQ's setup bot no longer repeats its open question for every setup note the desktop app sends. A note is folded in, and the bot posts only when the note changes what the person needs to know.
+- `hq packs update` installs the newer marketplace version it finds. It used to reinstall the version recorded at install time, so a pack reported "0.4.15 available" and stayed on 0.4.14.
+- Installing a marketplace pack retries a lookup that the server answers with a 5xx or a network error (twice, after 0.5 s and 1.5 s). One brief 502 from the gateway used to end the install.
+- A pack's setup bot is created on Claude Opus 5.5 at medium thinking (`--model claude-opus-5-5 --effort medium`). Left unset it ran on the account's default model, which took up to nine minutes for one setup step. `hq bot set` still changes either.
+- `hq daemon` can supervise isolated Claude Remote Control account pools with per-account inference credentials. Pool desired state now names an existing personal-only `serviceSessionId`; the daemon validates its scope capability and pins `HQ_SESSION_ID` only for Secrets child processes, independent of the mutable current session.
+
 ## [5.345.104] — 2026-10-12
 - Session-start health no longer files a leftover personal-files journal
   warning plus an ordinary HQ Core upgrade as failed doctor repairs.
@@ -10,6 +106,10 @@
 - On Macs, idle `hq lanes` work no longer sits queued on the memory throttle
   just because reclaimable cache made free RAM look too low.
 
+## [5.345.103] — 2026-10-12
+- On Macs, idle `hq lanes` work no longer sits queued on the memory throttle
+  just because reclaimable cache made free RAM look too low.
+
 ## [5.345.102] — 2026-10-12
 - Session-start health no longer files a leftover company journal as an
   unrepaired doctor finding, and company manifest rows keep a real check id

package/assets/bot-workers/setup/skills/add-a-pack.md

@@ -0,0 +1,130 @@
+---
+name: add-a-pack
+description: Recognize a content pack the person pastes or names (a hqforwork.com pack link, hq-pack-{slug}, a bare slug, or a display name from the catalog), confirm in one message, install it with --start-bot, and hand over to the pack's bot.
+---
+
+# Adding a pack from the chat
+
+A pack is a bundle of workers, knowledge and a setup bot that HQ installs with
+`hq packages install`. People arrive with one in three ways: they paste a link
+from a pack page, they type the pack's name, or they drop a folder path. You add
+it without them leaving this conversation, and you end by introducing the bot
+the pack brings. This skill only adds that; nothing else about how you run setup
+or answer how-do-I questions changes.
+
+## Spotting a pack in the message
+
+Any of these, anywhere in what they wrote, is a pack reference. Read the slug
+out of it:
+
+- `https://www.hqforwork.com/packs/{slug}`
+- `hqforwork.com/packs/{slug}` (no scheme, with or without `www.`)
+- `hq-pack-{slug}` (the package name)
+- `{slug}` on its own, when it is in the catalog (for example `sample-kit`)
+- a display name from the catalog, in plain words ("add the sample kit",
+  "I want the Sample Kit"). Match case-insensitively; "sample kit" and
+  "sample-kit" are the same name.
+- a local folder path whose last `hq-pack-*` segment names the pack
+  (`./packs/hq-pack-sample-kit`, `~/Downloads/hq-pack-sample-kit`). Keep the
+  path: it is what you install from.
+
+`sample-kit` / "Sample Kit" below is a placeholder: use the real slug and
+display name from the catalog.
+
+Slugs are lowercase letters, digits and hyphens. A link or a package name is
+always a reference, even if the catalog does not list it: the slug is explicit.
+A bare word or a name must match the catalog, so you never guess a pack from
+ordinary conversation ("my co-founder is joining" is not a pack).
+
+### The catalog
+
+The catalog is what `hq packages list` prints under "Available (curated)": each
+line is an install source like `github:indigoai-us/hq-packages#packages/hq-pack-design-styles`
+with a description. The slug is the last segment without `hq-pack-`; the display
+name is the slug in title case (`design-styles` is "Design Styles") unless the
+catalog gives one. Run the list once per conversation when you need to check a
+name; never invent a pack that is not there.
+
+## Confirm first, in one message
+
+Before you install anything, send exactly one message: the pack's display name,
+its one-line description from the catalog (or from the pack page if the person
+pasted a link and the catalog has no description), and what will happen: you
+will add it and its bot will introduce itself. End with the suggestions
+envelope, exactly these two:
+
+```hq-block
+{"v":1,"blocks":[{"kind":"suggestions","items":["Add the {displayName}","Not now"]}]}
+```
+
+Example, for `https://www.hqforwork.com/packs/sample-kit`:
+
+> Sample Kit: {the catalog's one-line description}. It brings its own bot, who
+> will take you through its setup with a few quick questions. Want me to add it?
+
+followed by the envelope with `["Add the Sample Kit","Not now"]`.
+
+"Not now" (or anything that is not a yes): say "Okay, it's here whenever you
+want it" and carry on with whatever you were doing. Do not ask again in that
+conversation unless they bring it up.
+
+## On "Add the ..."
+
+Run, from the HQ root:
+
+    hq packages install {slug-or-path} --start-bot
+
+`{slug-or-path}` is the slug (`sample-kit`), or the local path when the message
+carried one. Quote a path with spaces. Wait for it to finish; it can take a
+minute. Do not narrate the steps while it runs.
+
+Then report in **one** message:
+
+1. What was added, in one line ("The Sample Kit is in.").
+2. Where its pieces live, as Files-tab locations, only if the install output
+   names any (for example Knowledge > {Pack} Playbook). Never a filesystem path.
+3. The hand-over. The install output has a line `bot: {name}` when a bot was
+   started. Name that bot by its display name and say it will take it from here
+   in its own conversation in Messages, for example: "{Bot display name} is
+   starting now and will take it from here; look for its message in Messages."
+   If the output has no `bot:` line, say the pack is installed and name what it
+   added instead.
+
+That ends your part. Do not run the pack's setup yourself, do not re-ask the
+pack's questions, and do not keep the conversation going about the pack: the
+pack's bot owns it from here. Write one line in your progress note
+(`Pack added: {slug}, bot {name}`) so a later turn knows.
+
+## When you do not know the pack
+
+A bare word or name that matches nothing in the catalog: say so in one line and
+offer the three closest names from the catalog, as suggestions, plus "Not now".
+For example:
+
+> I don't know a pack called "growth-kit". The closest I have are Sample Kit,
+> Field Notes and Design Styles.
+
+```hq-block
+{"v":1,"blocks":[{"kind":"suggestions","items":["Add the Sample Kit","Add the Field Notes","Add the Design Styles","Not now"]}]}
+```
+
+If the catalog has fewer than three packs, offer what there is. Never offer a
+name you did not see in the list.
+
+## When the install fails
+
+Say what went wrong in plain words and give one next step, in one message. Take
+the reason from the command's output, not from a guess, and translate it:
+
+- not signed in to HQ Cloud: "I couldn't add the pack because you're not signed
+  in to HQ Cloud. Sign in from the HQ app, then tell me to try again."
+- no network or the registry did not answer: "I couldn't reach the pack
+  registry. Check your connection and tell me to try again."
+- the pack was installed but its bot did not start (the output says so): "The
+  Sample Kit is in, but its bot didn't start: {reason in plain words}. Open
+  Bots in HQ and start {bot display name}, or tell me to try again."
+- anything else: quote the one line of the error that says what happened, in
+  plain words, then: "Tell me to try again, or send me the link once more."
+
+End that message with `["Try again","Not now"]` as the suggestions. Never paste
+a stack trace or a wall of command output into the chat.

package/assets/bot-workers/setup/worker.yaml

@@ -45,6 +45,9 @@ skills:
   - name: explain-hq
     description: "The fixed explanation of HQ for someone who asks for it at the start of setup. Sent word for word, by you or by the HQ app."
     doc: skills/explain-hq.md
+  - name: add-a-pack
+    description: "Recognize a content pack the person pastes or names (a hqforwork.com pack link, hq-pack-{slug}, a bare slug, or a display name from the catalog), confirm in one message, install it with --start-bot, and hand over to the pack's bot."
+    doc: skills/add-a-pack.md
 
 instructions: |
   # Setup

package/dist/command-catalog.generated.d.ts

@@ -161,6 +161,10 @@ export declare const COMMAND_CATALOG: readonly [{
             readonly flags: "--allow-unsigned";
             readonly description: "Acknowledge an unsigned local directory source; marketplace installs always require a signature";
             readonly hidden: false;
+        }, {
+            readonly flags: "--start-bot";
+            readonly description: "After install, create and start the pack's setup bot (packs with a setup: block); prints `bot: <name>`";
+            readonly hidden: false;
         }, {
             readonly flags: "--branch";
             readonly description: "Follow a ref instead of SHA-pinning (git content-pack flow)";
@@ -551,6 +555,10 @@ export declare const COMMAND_CATALOG: readonly [{
         readonly flags: "--allow-unsigned";
         readonly description: "Acknowledge an unsigned local directory source; marketplace installs always require a signature";
         readonly hidden: false;
+    }, {
+        readonly flags: "--start-bot";
+        readonly description: "After install, create and start the pack's setup bot (packs with a setup: block); prints `bot: <name>`";
+        readonly hidden: false;
     }, {
         readonly flags: "--branch";
         readonly description: "Follow a ref instead of SHA-pinning (git content-pack flow)";
@@ -4793,6 +4801,10 @@ export declare const COMMAND_CATALOG: readonly [{
             readonly flags: "--timeout <seconds>";
             readonly description: "How long to wait for a browser sign-in (default 300)";
             readonly hidden: false;
+        }, {
+            readonly flags: "--link-only";
+            readonly description: "Print the sign-in link and exit without waiting; the sign-in finishes on the HQ console";
+            readonly hidden: false;
         }, {
             readonly flags: "--force";
             readonly description: "Connect again even if this app is already connected";
@@ -5338,6 +5350,58 @@ export declare const COMMAND_CATALOG: readonly [{
             readonly hidden: false;
         }];
         readonly subcommands: readonly [];
+    }, {
+        readonly name: "slack";
+        readonly description: "Connect your own Slack account to HQ (stored in your personal vault)";
+        readonly aliases: readonly [];
+        readonly hidden: false;
+        readonly helpOption: true;
+        readonly helpCommand: true;
+        readonly usage: "[options] [command]";
+        readonly helpText: readonly [];
+        readonly arguments: readonly [];
+        readonly options: readonly [];
+        readonly subcommands: readonly [{
+            readonly name: "connect";
+            readonly description: "Get Slack's sign-in link for your account; the sign-in finishes on HQ";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--company <slug>";
+                readonly description: "Company the Slack workspace belongs to (defaults to your single active company)";
+                readonly hidden: false;
+            }, {
+                readonly flags: "--open";
+                readonly description: "Open the link in the browser instead of only printing it";
+                readonly hidden: false;
+            }, {
+                readonly flags: "--json";
+                readonly description: "Machine-readable output: { url, opened }";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "accounts";
+            readonly description: "List your connected Slack accounts";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--json";
+                readonly description: "Machine-readable output";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }];
     }];
 }, {
     readonly name: "import";
@@ -15520,6 +15584,68 @@ export declare const COMMAND_CATALOG: readonly [{
             readonly hidden: false;
         }];
         readonly subcommands: readonly [];
+    }, {
+        readonly name: "remote-control-pools";
+        readonly description: "Validate, configure, and inspect opt-in Claude Remote Control account pools";
+        readonly aliases: readonly [];
+        readonly hidden: false;
+        readonly helpOption: true;
+        readonly helpCommand: true;
+        readonly usage: "[options] [command]";
+        readonly helpText: readonly [];
+        readonly arguments: readonly [];
+        readonly options: readonly [];
+        readonly subcommands: readonly [{
+            readonly name: "validate";
+            readonly description: "Validate a host-local desired-state JSON file without saving it";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options] <file>";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [{
+                readonly name: "file";
+                readonly required: true;
+                readonly variadic: false;
+                readonly description: "";
+            }];
+            readonly options: readonly [];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "configure";
+            readonly description: "Save a validated host-local desired-state JSON file";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options] <file>";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [{
+                readonly name: "file";
+                readonly required: true;
+                readonly variadic: false;
+                readonly description: "";
+            }];
+            readonly options: readonly [];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "show";
+            readonly description: "Show configured pool metadata (never secret values)";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--json";
+                readonly description: "Print JSON";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }];
     }, {
         readonly name: "service";
         readonly description: "Entry points the daemon runs for services that have no command of their own";
@@ -15567,6 +15693,82 @@ export declare const COMMAND_CATALOG: readonly [{
                 readonly hidden: false;
             }];
             readonly subcommands: readonly [];
+        }, {
+            readonly name: "remote-control-pool";
+            readonly description: "Keep one configured Claude Remote Control account pool running in its private mount namespace";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--account-id <id>";
+                readonly description: "Claude account ID from hq lanes accounts";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "remote-control-pool-host";
+            readonly description: "Run one pool's forwarder and Claude host inside its private namespace";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--account-id <id>";
+                readonly description: "Claude account ID from hq lanes accounts";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "remote-control-pool-forwarder-host";
+            readonly description: "Supervise one pool's personal HQ Secret injected inference forwarder";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--account-id <id>";
+                readonly description: "Claude account ID from hq lanes accounts";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
+        }, {
+            readonly name: "remote-control-pool-forwarder";
+            readonly description: "Run the loopback-only /v1/messages inference forwarder";
+            readonly aliases: readonly [];
+            readonly hidden: false;
+            readonly helpOption: true;
+            readonly helpCommand: false;
+            readonly usage: "[options]";
+            readonly helpText: readonly [];
+            readonly arguments: readonly [];
+            readonly options: readonly [{
+                readonly flags: "--account-id <id>";
+                readonly description: "Claude account ID from hq lanes accounts";
+                readonly hidden: false;
+            }, {
+                readonly flags: "--port <port>";
+                readonly description: "loopback listener port";
+                readonly hidden: false;
+            }, {
+                readonly flags: "--secret-env <name>";
+                readonly description: "HQ Secrets injected environment variable name";
+                readonly hidden: false;
+            }, {
+                readonly flags: "--ready-file <path>";
+                readonly description: "private daemon state file for listener ownership";
+                readonly hidden: false;
+            }];
+            readonly subcommands: readonly [];
         }, {
             readonly name: "codex-remote-control";
             readonly description: "Keep the box's Codex remote-control runner going in a tmux session, restarting it after a Codex update";
@@ -15785,6 +15987,10 @@ export declare const COMMAND_CATALOG: readonly [{
             readonly description: "Where the bot keeps its memory: synced (personal/workers/<name>/memory in HQ) or local (~/.hq/bots/<name>/memory, this computer only)";
             readonly hidden: false;
             readonly defaultValue: "synced";
+        }, {
+            readonly flags: "--lean";
+            readonly description: "Claude bots: run each turn without this computer's plugins, MCP servers and HQ's session hooks (HQ's tool hooks still run), so replies take seconds";
+            readonly hidden: false;
         }, {
             readonly flags: "--no-daemon";
             readonly description: "Do not install the launchd agent (macOS)";