name: handydandy repo: https://github.com/percyhendrux123-prog/hermes-handydandy sha: 7cc035707f18590db3957fbcc73e3e4064bc8f38 version: "0.3.0" description: "Session-size notices and structured handoffs. After each turn it checks the estimated input tokens and message count Hermes reports; when a limit is reached it either adds one notice to the reply (advise, the default) or asks the agent, on your next message, to finish your request and write a HANDOFF-*.md file that you open yourself in a new session. No extra model calls, and it never starts a session on its own. Disclosure — writes state.json, state.lock, events.log and short-lived atomic-save temp files under /plugin-data/handydandy/ (profile and session ids, paths, thresholds and error details; no conversation text); asks the agent to write handoff files to handoff_dir (default /handoffs/), which may hold work details from the conversation; optionally appends one row per completed handoff to a session_log_csv the owner sets (off by default). Reads its own state, the first byte and modification time of recorded handoff files, plugin settings via ctx.get_config, and HANDYDANDY_ and HERMES_HOME environment variables. No network calls, no shell commands, no background processes, no stored credentials, no telemetry; it does not auto-approve tools or change safety settings." maintainer: percyhendrux123-prog tier: community category: general requires_hermes: ">=0.22.5" docs_url: https://github.com/percyhendrux123-prog/hermes-handydandy/blob/7cc035707f18590db3957fbcc73e3e4064bc8f38/README.md readme: true platforms: [macos, linux] known_issues: - "Handoff mode is an instruction to the agent, not a guarantee; the instruction repeats each turn until a readable, non-empty handoff file newer than the notice is observed." - "The percentage-of-context check is off unless the owner sets context_limit to the model's context window; the plugin does not look up model metadata." - "The state file keeps the 200 most recent session records, so a much older session can be warned again." capabilities: provides_tools: [] provides_hooks: - pre_api_request - transform_llm_output - pre_llm_call provides_middleware: [] requires_env: []