From cc-skills
Manage cc-watchdog state files that prevent premature Stop events in Claude Code. Use when the user says '/cc-watchdog', 'cc-watchdog', 'start the watchdog', 'stop the watchdog', 'list watchdogs', or asks to keep a session from auto-stopping. Subcommands: start/stop/status/list.
How this skill is triggered — by the user, by Claude, or both
Slash command
/cc-skills:cc-watchdog start|stop|status|list [name] [message]start|stop|status|list [name] [message]This skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
<process>
Invoke the cc-watchdog CLI at ${CLAUDE_SKILL_DIR}/scripts/cc-watchdog. It writes/reads JSON under ${XDG_STATE_HOME:-$HOME/.local/state}/cc-watchdog/ that the Stop hook in persistent-mode.mjs reads.
Parse $ARGUMENTS into subcommand + optional name + optional message:
start, stop, status, or list. Required.start takes optional [name] and optional [message] (continuation string injected at Stop). Default name: $ZELLIJ_SESSION_NAME if set, otherwise "default".stop takes optional [name]. If name is omitted, it clears every watchdog whose project_path matches the current git root / cwd — confirm with the user before running stop without a name if there are multiple active watchdogs.status takes optional [name] (default: $ZELLIJ_SESSION_NAME if set, otherwise "default") and dumps the JSON.list takes no args and prints active watchdog names + project paths.Resolve the name before invoking: read $ZELLIJ_SESSION_NAME from the environment (it is always set inside a Zellij pane). Use it as the name whenever the user has not supplied one explicitly.
Invoke:
${CLAUDE_SKILL_DIR}/scripts/cc-watchdog <subcommand> [name] [message]
If the user's message is plain English ("start a watchdog to keep working on the migration"), infer the subcommand and a short message. Confirm the parsed arguments before running start or bare stop.
Unknown subcommand → the CLI prints the usage line and exits 1. Don't retry; show the usage to the user.
- `start` writes `{name}.json` with `active: true`, the message, current project root (from `git rev-parse --show-toplevel` or `$PWD`), and timestamps. - Watchdog state is project-scoped via `project_path` — a watchdog started in repo A won't block Stop in repo B. - Default continuation message if `$3` is omitted: *"Continue working. If you were mid-workflow, re-read the workflow TOC and current step file, then proceed from the current step."* - Name is sanitised to `[a-zA-Z0-9_-]`; slashes become underscores. Warn if the user's requested name gets mangled. - `$ZELLIJ_SESSION_NAME` is always present inside a Zellij pane (the `c` launcher creates one per invocation, e.g. `cc-myproject-1234567`). Use it verbatim as the default name — no need to prompt the user.npx claudepluginhub jacobcxdev/cc-plugins --plugin cc-skillsProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.