From verified-development
Continue a paused feature: read handoff.json + continue-here.md, brief, route next action.
How this skill is triggered — by the user, by Claude, or both
Slash command
/verified-development:continue [--force][--force]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You MUST use this when the user types `/continue`, asks to "resume", "continue where we left off", "pick up where we stopped", or when SessionStart detects a fresh handoff.json. Read state, brief the user in ONE message, then route to the next action.
You MUST use this when the user types /continue, asks to "resume", "continue where we left off", "pick up where we stopped", or when SessionStart detects a fresh handoff.json. Read state, brief the user in ONE message, then route to the next action.
Read .verified/state.md for feature and phase. If no feature is set, tell the user there is nothing to resume — they should run /progress for current status or /specify to start something new — and stop.
node ${CLAUDE_PLUGIN_ROOT}/hooks/lib/handoff.js read .verified/features/<feature>
Parse it. Validate via the helper (validate command on stdin) — if validation fails the file is corrupt; offer to delete it and run /progress to rebuild.
This is one of three cases:
/continue --force (proceed without handoff context) or future /forensics. Stop unless --force was passed.next_action is set (/plan, /implement, etc.) and no in-flight artifacts exist. This is a clean handoff between phases — just report the next action and prompt the user to invoke it. Do not invoke it yourself.feature: none. Same as no-feature case above.Run git rev-parse --short HEAD. Compare with git_head in handoff.
If they differ, the working tree has moved since the pause. Warn the user, list what changed (git log --oneline <handoff-head>..HEAD), and offer: "(a) continue anyway, (b) rewind to handoff sha, (c) /continue --force". Do not pick for them. Stop unless they answer.
If blockers contains any with severity: blocking, list them. Do NOT proceed past the brief. The resuming agent must show it understood by either:
--force does NOT override blocking blockers — they are intentional safety gates.
Read .verified/features/<feature>/continue-here.md. Treat it as the narrative companion to the JSON. If they disagree, trust the JSON.
Format:
Continuing `<feature>` (`<phase>`).
Completed N/M tasks: <list of completed task IDs>.
Next: <single concrete next action — task title or slash command>.
Blockers: <list, or "none">.
Decisions made mid-phase:
- <decision 1>
- <decision 2>
(or "none")
<one-line summary of "what I just did" from continue-here.md>
Keep it tight — 6–10 lines. The user wants situational awareness in one glance.
Tell the user the next concrete command to run (/implement to continue, or the specific task if mid-/implement). Do NOT auto-invoke it. The user should approve before continuing work — they may have stopped for a reason that's still relevant.
Set active_phase to the current phase (we're now in flight again). Update last_activity to today. Leave next_action as-is until the user actually proceeds.
--force skips orphan-state detection and git_head mismatch warning, but NEVER skips severity: blocking blockers.active_phase + last_activity update at the end. Do NOT clear the handoff — that happens when the phase completes.npx claudepluginhub mikemyl/verified-development --plugin verified-developmentProvides 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.