CSPR Personal Recommender
CSPR is a cross-source personal recommender skill bundle.
It contains four skills:
interest-profiler: infer and refresh local preference memory from browsing evidence, notes, and feedback.source-scout: collect fresh cross-source recommendation candidates using profile memory, feeds, search, Agent-Reach, and optional subagents.news-editor: rank candidates and write a concise personalized Markdown newspaper.memory-curator: update preference memory from accumulated feedback and newspaper outcomes.
Version 1.1.1 also includes:
scripts/cspr-refresh-history.sh: incrementally scan local browser history and refresh preference memory.scripts/cspr-daily-newspaper.sh: run the daily cross-source discovery/editor pipeline and send the generated newspaper through OpenClaw's configured Telegram channel.src/cspr/andpyproject.toml: the local Python CLI source needed by the scripts.
Runtime state, browsing history, generated newspapers, tokens, local configuration, virtual environments, caches, and credentials are not included.
Reproduce locally
Prerequisites:
uvopenclaw- Python 3.11+
- Browser history available on the local machine
- Agent-Reach tools installed if you want the full cross-source route coverage
- Telegram delivery configured in OpenClaw, or
CSPR_TELEGRAM_TARGETset
From the plugin root:
uv sync
scripts/cspr-refresh-history.sh
scripts/cspr-daily-newspaper.sh
Useful overrides:
CSPR_HOME_DIR="$HOME/.cspr" scripts/cspr-refresh-history.sh
CSPR_OPENCLAW_MODEL="your-model" CSPR_OPENCLAW_THINKING="high" scripts/cspr-daily-newspaper.sh