Background skill that scans the last 7 days of memory/sessions/ at session start to detect repeated actions and suggest automations via the proactivity engine. NOT user-invocable. Activates silently at session start before the user's first task. Do NOT activate during /begin onboarding, when fewer than 3 session log files exist, when the user immediately starts a specific task (queue suggestion instead), or when the same pattern was already dismissed twice (permanent stop).
From tandemnpx claudepluginhub binatrixai/tandem-marketplace --plugin tandemThis skill is limited to using the following tools:
evals/evals.jsonreferences/pattern-mapping.mdExecutes pre-written implementation plans: critically reviews, follows bite-sized steps exactly, runs verifications, tracks progress with checkpoints, uses git worktrees, stops on blockers.
Verifies tests pass on completed feature branch, presents options to merge locally, create GitHub PR, keep as-is or discard; executes choice and cleans up worktree.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
Analyze session history to detect repeated actions and suggest automations. This skill runs silently at the start of each new session — never announce that you're scanning unless the user asks.
Activate this analysis:
~/Tandem/memory/sessions/ exists and has at least 3 days of logsDo NOT run:
Read the last 7 days of session files from ~/Tandem/memory/sessions/:
memory/sessions/YYYY-MM-DD.md (today)
memory/sessions/YYYY-MM-DD.md (yesterday)
...up to 7 days back
For each session log, extract the "Skills/commands used" field. Build a frequency table:
Skill/Command | Days Used (out of 7) | Times Total
/email-triage | 5 | 7
/daily-plan | 4 | 4
/monday | 3 | 5
memory-create | 6 | 12
/report | 1 | 1
A pattern is detected when:
Before suggesting, verify the pattern isn't already covered:
~/Tandem/memory/preferences.md for any noted scheduled tasksIf a new automatable pattern is found, queue a suggestion for the proactivity engine to present at the next natural pause point in the conversation.
| Pattern Detected | Suggestion | Scheduled Task |
|---|---|---|
| /email-triage used 3+ days | "I notice you triage email most mornings. Want me to do this automatically every day at [time]?" | Daily email triage at detected time |
| /daily-plan used 3+ days | "You plan your day regularly. Want me to generate a daily plan automatically each morning?" | Daily plan at detected time |
| /monday used 3+ days | "You check Monday.com frequently. Want me to pull your board summary every morning?" | Daily Monday.com sync |
| /weekly-review used on Fridays | "You do weekly reviews on Fridays. Want me to set this up as a recurring Friday task?" | Weekly review on Fridays |
| /research used 2+ times on same topic area | "You've researched [area] multiple times. Want me to set up a recurring competitor/topic watch?" | Weekly research update |
| Same document type created 3+ times | "You create [type] documents regularly. Want me to save a template based on your recent ones?" | Template creation (not scheduled, but memory-based) |
| /email-draft to same recipient 3+ times | "You email [person] frequently. Want me to remember their preferred communication style?" | Memory entry (not scheduled) |
Present via proactivity engine's AskUserQuestion — always in the user's language:
I noticed a pattern in your recent sessions:
[Brief description of what was detected]
AskUserQuestion:
- "Yes, set it up as a daily task at [suggested time]"
- "Yes, but at a different time"
- "Not now — maybe later"
- "No, I prefer doing this manually"
If "Yes, set it up": Create the scheduled task and confirm. Log the automation in memory/preferences.md.
If "Yes, but different time": Ask for preferred time, then create the scheduled task.
If "Not now": Log as "dismissed" in session. Will suggest again after 14 days.
If "No, I prefer manually": Log as "dismissed (permanent)" in session. Will not suggest this pattern again.