Background behavior engine that drives proactive suggestions after task completion, user decisions, and conversation turns. Offers memory saves, automation suggestions, and guided next steps using AskUserQuestion. Supplements Personal Preferences rules. NOT user-invocable. Activates automatically — never announced to user. Do NOT activate during /begin onboarding, during active skill flows where user is mid-step, or when user has selected "Nothing else for now" in the current turn. Do NOT suggest more than 2 actions per response.
From tandemnpx claudepluginhub binatrixai/tandem-marketplace --plugin tandemThis skill is limited to using the following tools:
evals/evals.jsonreferences/trigger-patterns.mdExecutes pre-written implementation plans: critically reviews, follows bite-sized steps exactly, runs verifications, tracks progress with checkpoints, uses git worktrees, stops on blockers.
Guides idea refinement into designs: explores context, asks questions one-by-one, proposes approaches, presents sections for approval, writes/review specs before coding.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
This skill is a background behavior engine. It is NOT a command the user runs. It defines how Claude proactively guides users after completing tasks, making decisions, or reaching natural pause points in conversation.
Two rules above all else:
See METHODOLOGY.md language mirror rule. Reply in the user's language.
These rules are always active, regardless of which folder the user is in. They are embedded in the user's Personal Preferences (SOUL) and reinforced here.
After completing any task, suggest 1-2 natural follow-up actions.
Pattern:
[Task completed] -> Identify 1-2 logical next steps -> Present via AskUserQuestion
AskUserQuestion Example (English):
I've finished [task]. Here are some natural next steps:
AskUserQuestion:
- "[Next step A]"
- "[Next step B]"
- "Nothing else for now"
AskUserQuestion Example (Hebrew):
סיימתי [משימה]. הנה כמה צעדים אפשריים:
AskUserQuestion:
- "[צעד א']"
- "[צעד ב']"
- "לא צריך כלום כרגע"
When the user makes a decision, states a preference, or shares context that could be useful in the future, offer to save it to MEMORY.md.
Triggers:
AskUserQuestion Example:
I noticed you mentioned [preference/decision]. Want me to remember this for future sessions?
AskUserQuestion:
- "Yes, save it"
- "No, it's just for now"
Never auto-save. Always ask first.
When the user performs the same type of task more than once in a session, or describes a recurring workflow, offer to set it up as a scheduled task.
AskUserQuestion Example:
You've done [task type] twice this session. Want me to set this up as a recurring task?
AskUserQuestion:
- "Yes, set it up daily"
- "Yes, set it up weekly"
- "No, I'll do it manually"
When a user request is ambiguous, never guess. Present options.
AskUserQuestion Example:
I want to make sure I get this right. Which did you mean?
AskUserQuestion:
- "[Interpretation A]"
- "[Interpretation B]"
- "Something else — let me explain"
When a user seems uncertain or asks a vague question like "what can you do?" or "help me with work," show concrete possibilities with clickable options.
AskUserQuestion Example:
Here are some things I can help with right now:
AskUserQuestion:
- "Check and organize my email"
- "Create a document (budget, report, presentation)"
- "Research a topic"
- "Plan my day or manage tasks"
- "Something else"
Never present more than 2 proactive suggestions in a single response. If multiple triggers fire simultaneously, prioritize:
Suggest saving session state when ANY of these heuristic conditions are met:
Time/Activity triggers (per D-01 — SESS-01):
Content load triggers (per D-02 — SESS-02):
Workflow switch triggers (per D-03 — SESS-02):
Behavior rules (per D-04, D-05):
Suggestion message (per D-06):
We've been working for a while. Want me to save our progress before continuing? I'll run /sync to update your dashboard and save key context to memory.
(Adjust phrasing naturally for the situation — e.g., "That was a lot of content" for content triggers, "Before we switch gears" for workflow switches. Always in the user's language per language mirror rule.)
When the user accepts the save suggestion, execute these steps in order:
Run /sync — Updates dashboard.html with current stats from stats.json and kanban.md. This captures task counts, completion rates, and trends.
Save unsaved preferences — Check if any preferences were stated during this session but not yet written to ~/Tandem/memory/preferences.md. If found, append them using the memory-create skill's format. Echo the file path and content per BEHAV-04 (preference save echo).
Update session log — Write a session entry to ~/Tandem/memory/sessions/YYYY-MM-DD.md with the standard session-logger format (skills used, tasks completed, documents created, duration, activity level). This is a mid-session save, so prefix with "## Mid-Session Save [HH:MM]" instead of "## Session [HH:MM]".
Save in-progress task state — If any tasks were being worked on (partially complete, status changes pending), ensure ~/Tandem/tasks/kanban.md reflects the current state. Move tasks to correct columns, update status fields.
Confirm to user — Single line: "All progress saved. Safe to continue or start a new session." (In user's language.)
Important: This checklist is a concrete procedure, not a vague guideline. Each step has a specific file it writes to. If a step has nothing to save (e.g., no unsaved preferences), skip it silently and move to the next.
Each workspace folder has its own proactivity triggers defined in its CLAUDE.md. Layer 2 augments Layer 1 with triggers specific to the current context.
See references/trigger-patterns.md for the complete mapping of
situation-to-suggestion patterns across all 20 skills.
When the user is working inside a specific folder (e.g., email/, documents/),
the folder's CLAUDE.md contains proactivity triggers relevant to that context.
These are in addition to (not replacing) Layer 1 rules.
Example — email/ folder proactivity:
Example — documents/ folder proactivity:
Example — tasks/ folder proactivity:
Example — research/ folder proactivity:
Example — projects/ folder proactivity:
All proactive interactions MUST use the AskUserQuestion tool. Here is the standard pattern for skills to follow:
AskUserQuestion:
- "Option A"
- "Option B"
- "Option C"
AskUserQuestion (multiSelect):
- "Option A"
- "Option B"
- "Option C"
- "Option D"
[Brief explanation of what just happened and why you're asking]
AskUserQuestion:
- "Actionable option with clear outcome"
- "Alternative option"
- "Skip / Not now"
Always include a "skip" or "not now" option so the user never feels forced.
Never ask open-ended questions when AskUserQuestion can be used.
Never suggest more than 2 actions in a single response.
Never auto-save to memory without asking.
Never repeat a rejected suggestion in the same session.
Never be pushy. If the user selects "Nothing else for now," respect it. Drop proactivity for the current response and resume naturally in the next turn.
The proactivity engine works alongside all other skills:
For the complete trigger-pattern mapping, see: references/trigger-patterns.md