berd-help
GitHub协助用户操作、排查 Berd 桌面应用问题,涵盖设置、代理、自动化及 berctl 命令使用。强调基于证据和源码验证,不猜测界面路径或过时文档内容。
Trigger Scenarios
Install
npx skills add block/berd --skill berd-help -g -y
SKILL.md
Frontmatter
{
"name": "berd-help",
"metadata": {
"berdBundled": true
},
"description": "Help with Berd the desktop app: how-to, troubleshooting, settings, agents, skills, automations, projects, sessions, providers, connections, feedback, or berdctl. Use for app questions, not harness-specific behavior."
}
Berd help
Use this skill to help users operate, troubleshoot, and script Berd — the
desktop app — from a normal chat. Topics below point to references/ files;
read the one relevant to the user's question rather than guessing, and read
more than one if the question spans topics.
First Principles
- Before giving exact product instructions, try to verify the behavior from
available evidence. If source code is available, inspect it first. If
source code is not available, use visible app state, screenshots, error
messages, Doctor output, or
berdctl --helpoutput. - Do not invent UI paths, menu names, frontmatter fields, feature existence,
fallback workflows, or
berdctlcommands/flags. If you cannot verify the exact path, say that and ask for a screenshot, selected project/source access, or the specific screen the user is on. Forberdctl, verify against--helprather than recalling flags from memory. - Treat the current app behavior as the source of truth. Prefer evidence from
visible UI, runtime state, error messages, Doctor output, local files,
berdctl --help, and available source over memory or stale documentation. - Give practical steps the user can try now. Keep the answer short unless the problem needs a deeper diagnosis.
- Do not claim access to private GitHub source unless it is actually available in the current environment.
- Berd changes fast. Prefer naming where to look (a file, a command's
--help, a visible settings section) over naming specific section names, labels, flags, or IDs in this skill's own prose — those are exactly the details most likely to move. When a reference file below does name something specific, treat it as a hint of where to look, not a guarantee it still matches current behavior; verify before repeating it to the user.
Scope
This skill is about Berd the app — not about any one harness running inside
a session (goose, claude-acp, codex-acp, copilot-acp, amp-acp,
cursor-agent). A question about the app around the harness (where a
setting lives, how a session started, how to export a chat, how to file a
Berd bug) is in scope. A question about a specific harness's own behavior,
output, or errors is not — treat that as harness-specific and say so rather
than guessing at harness internals.
What To Check
For how-to questions:
- Identify the feature area from the Topics list below and read that reference file.
- If source is available, search for the feature's UI, command, hook, tests, or i18n strings before answering.
- Use the app's current state when available: selected agent, project, model, provider, active session, attached files, visible errors, or screenshots.
- Explain the shortest path to the action, then mention important
constraints or gotchas. If a
berdctlcommand does the same thing faster or more reliably than a click path, offer it — seereferences/berdctl.md.
For troubleshooting:
- Ask for the exact error, screenshot, provider/model, project path, or Doctor result when the symptoms are ambiguous.
- Suggest low-risk checks first: rerun, reconnect, verify settings, check selected model/provider, restart the app, or run Doctor when relevant.
- Separate likely causes from confirmed causes.
- If the problem looks like a real app bug rather than user error, see
references/reporting-problems.mdinstead of guessing at a fix.
Topics
Read the matching reference file for a specific question. Read more than one when a question spans topics (for example, an automation failure that might be a provider problem needs both):
references/berdctl.md— the berdctl CLI, when to prefer it over UI stepsreferences/agents.md— agents/personasreferences/skills.md— skills, source kinds, precedencereferences/automations.md— automationsreferences/projects.md— projects and workspacesreferences/sessions.md— sessions ("chat"), the right rail, files, worktrees, terminalreferences/ai-providers.md— provider connections and error causesreferences/connections-and-extensions.md— connections vs. extensionsreferences/settings.md— settings structure and gotchasreferences/reporting-problems.md— filing feedback or a bug
Working With Bundled Builder Skills
- If the user wants to create, edit, or inspect an agent/persona, use or
recommend the
agent-builderskill. - If the user wants to create, edit, or inspect a reusable skill, use or
recommend the
skill-builderskill. - Do not duplicate those workflows inside this skill or its references. This skill helps users understand, troubleshoot, and drive the app around them.
Source Of Truth
When source code is available, use it to verify behavior — the reference
files above may lag actual behavior since Berd changes fast. Good places to
inspect include UI feature folders, Tauri commands, distro bundled skills,
settings views, berdctl command modules
(src/features/berdctl/commands/impl/), and tests. If source code is not
available, rely on current app context and --help output, and be
transparent about uncertainty.
Version History
- 977d35f Current 2026-08-19 11:30


