cli.md

commands

The bercail CLI, the shell helpers, and the skills your agents call by name. bercail help prints a short version in your terminal, keys included.

install / CLI

Command	What
bercail install	Full install from the release mise or packslip unpacked
./install.sh	Full install
./install.sh --yes (-y)	Non-interactive
./install.sh --no-bstack	New install without bstack (desk default stays cursor-agent)
./install.sh --bstack	New install with bstack (the default)
./install.sh --help	Installer help
bercail help	This reference
bercail doctor	Deps, Herdr layout, mod and keys, skills, webhook sinks and follower
bercail harness	JSON: which of claude, codex, cursor-agent, opencode are on PATH (reports, does not choose)
bercail bstack setup / status / on / off	bstack: opencode on your own plans. No subcommand = status
bercail desk restore / --dry-run	Bring back the desk agents a reboot or a core restart took, each on its session in its own space. See desks after a reboot. --dry-run lists each desk and what would happen to it
bercail update	Re-sync configs, Herdr layout and mods, helper, skills, and bstack; restart the notify follower. --force (-f) also replaces a changed layout and the worktrunk config, and writes the shell block despite conflicts
bercail review-pr <PR url-or-branch> / run / card / ui / state / select / checkout / rereview / open / dir	Review a whole PR stack by intent and decisions: one visible review desk session, cards and a stack summary beside its worktree. See review
bercail reconfigure	Re-read config.toml, refresh skills
bercail dry-run	Show what install would do, without doing it
bercail uninstall	Remove managed files and the shell marker
bercail notify status / add NAME / repo add NAME [REPO] / repo remove NAME REPO / check [PATH] / test NAME / openclaw [NAME] / run	Agent status webhooks. No subcommand = status; openclaw points a sink at a local OpenClaw Gateway; run is the follower the desk mod keeps alive
bercail shep status / setup / use pi / use openclaw [--agent ID | --create [ID]] [--session KEY] / handoff --to pi|openclaw / run	Shep: which runtime Shep runs on, and the handoff between them. No subcommand = status; run is what the Shep view starts

shell

worktrunk creates the tree; a Herdr space follows.

Command	What
dev (or d)	Open the Herdr app if needed, show the space for $PWD (made if missing), and start its desk agent if it has none
t	Open the Herdr app
wtc [branch]	Create (or switch to) the worktree for branch + its space. New branches start from HEAD. With no branch, the current one moves to a new worktree and this checkout goes back to the default branch
wts [branch] (or wt.)	Switch worktree (fzf if omitted)
wtd [branch]	Force-remove the worktree (default: the current branch) + close its space

skills

Call by name. Skills install to ~/.agents/skills; bercail adds a link where the configured agent looks elsewhere (~/.claude/skills, ~/.codex/skills, …).

Skill	What
handoff	Run a beads issue on a stage, headless or as a --branch desk (--base stacks it on a parent branch). See handoff
review	Wait for human tuicr notes; publish to GitHub only if asked. See review
bstack	Playbooks for rigorous work, routed by role. See bstack
council	Adversarial review on request; only findings a rerun confirms are reported
how / why	How code works now; why it is that way, from git, PRs, and beads. Read-only
unslop	Clean machine-writing tells out of prose and a diff
autopilot	Drive an open PR to green: CI, review threads, conflicts. Pushes; never merges
desk-restore	Put a desk with no record back on its session by hand, matched by folder and content

autopilot's PR snapshot, pr-state.sh, is in Ship with autopilot. The review helpers:

~/.agents/skills/review/scripts/wait-comments.sh --repo . [--timeout N]
~/.agents/skills/review/scripts/publish-github.sh --repo . \
  [--event comment|approve|request-changes] [--body TEXT] [--dry-run]

Manual skill install: npx skills add simoncrypta/bercail --skill handoff -g