Personal Secretary

Problem It Solves

Long-term agent collaboration often lacks durable context. After a task is finished, user preferences, project conventions, and completed work may remain buried in chat history. Weekly or monthly reviews then become manual reconstruction, while scheduled reports can miss runs caused by shutdown, sleep, or untriggered automations.

How It Works

The skill builds a local memory and reporting loop. User-level long-term memory is stored in ~/.workbuddy/MEMORY.md for preferences and collaboration habits. Project long-term memory lives in /.workbuddy/memory/MEMORY.md for workspace conventions. Daily work logs are appended to /.workbuddy/memory/YYYY-MM-DD.md under an append-only rule. Initialization creates directories such as inbox, processing, and outputs/汇报, plus state files like reports-state.md. After each substantive task, the agent appends a log entry. If the user explicitly creates daily, weekly, or monthly report automations, the system aggregates the logs and writes report files. When scheduled runs are missed, the session-level compensation rule reads reports-state.md, backfills missing reports at session start, and deduplicates entries.

Boundaries And Notes

The default mode is intended for a single workspace, with reports stored under that workspace's outputs/汇报. Multi-workspace aggregation requires creating and registering a separate reporting hub, so it is an optional advanced setup. The design avoids credentials, network access, and unauthorized automations; daily logs are append-only, and memory files are modified conservatively.