hermes-memory-plugin-integration
Connect Hermes plugins to memory safely.
- License 许可
- MIT
- 所属仓库
- zapabob/Skills
- 最近更新
- 2026-10-26
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- hermes-memory-plugin-integration
- author
- Hermes Agent
- license
- MIT
- version
- 1.0.0
- metadata
- { "hermes": { "tags": [ "hermes", "plugins", "memory", "ebbinghaus", "testing" ], "related_skills": [ "ebbinghaus-memory", "hermes-agent-skill-authoring" ] } }
- platforms
- [ "linux", "macos", "windows" ]
- description
- Connect Hermes plugins to memory safely.
03 SKILL.md 正文
Hermes Memory Plugin Integration Skill
Use this skill when a Hermes plugin should read from or write to an agent memory system without leaking private context. It is for plugin integration design, implementation review, and tests around memory-aware generated output.
This skill does not make memory mandatory for plugin operation. A plugin should keep its normal behavior when memory is disabled or unavailable unless the user explicitly requests a hard dependency.
When to Use
- Use when modifying a Hermes plugin that should use Ebbinghaus recall, sleep, consolidation, or writeback.
- Use when social posts, briefings, cron outputs, or personal-agent embodiments need durable continuity.
- Use when reviewing whether plugin prompts clearly separate trusted memory from untrusted public content.
- Do not use for ordinary session memory questions that do not involve plugin code.
Prerequisites
- The target plugin has a documented settings surface for optional features.
- The memory provider exposes a stable tool or local store for recall and writeback.
- Tests can create a temporary memory fixture instead of touching a real user database.
terminal,read_file,search_files, andpatchare available for code inspection and edits.
How to Run
Begin with the memory substrate:
- Prefer an official provider API or Hermes tool when one is available.
- If only a local store exists, bridge through it defensively and keep the path configurable.
- Keep the plugin usable when memory is missing, disabled, or malformed.
For Ebbinghaus-backed prototypes, use HERMES_HOME or get_hermes_home() to resolve state. Do not hardcode the default profile path.
Quick Reference
| Concern | Rule |
|---|---|
| Recall | Keep snippets bounded, ranked, and truncated. |
| Prompting | Label memory as trusted context and public input as untrusted. |
| Writeback | Store concise artifacts with source and tags. |
| Secrets | Never store credentials, cookies, tokens, or raw auth state. |
| Tests | Use temp SQLite or provider mocks, never real memory stores. |
Procedure
- Find the plugin's generation or delivery path and identify where memory can be optional context.
- Add configuration for enablement, provider or DB path, and recall limit.
- Implement bounded recall with graceful failure when memory is unavailable.
- Add prompt boundaries so private memory cannot be mistaken for user-approved public text.
- Write generated artifacts back only after a draft or live action is created, with clear provenance tags.
- Add targeted tests for enabled recall, disabled recall, missing-store behavior, and writeback.
Pitfalls
- Do not make publishing, cron delivery, or UX-critical paths fail because optional memory is offline.
- Do not add heavy dependencies just for simple ranking if token overlap is enough.
- Do not let memory alignment bypass dry-run, approval, or live-action gates.
- Do not write tests against the user's real
HERMES_HOME. - Do not store private session dumps as public artifact memory.
Verification
- Memory disabled: the plugin behaves exactly as before.
- Memory missing or malformed: the plugin continues without context.
- Memory enabled: relevant snippets enter the prompt with clear boundaries.
- Writeback stores concise content, source, tags, and timestamps.
- Tests run with temporary state and do not touch real user memory.
References
references/lm-twitterer-ebbinghaus-bridge.md
04 同仓库更多 Skills
xurl
X/Twitter via xurl CLI: post, search, DM, media, v2 API.
writing-plans
Write implementation plans: bite-sized tasks, paths, code.
whisper
OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual ASR.
weights-and-biases
W&B: log ML experiments, sweeps, model registry, dashboards.
watchers
Poll RSS, JSON APIs, and GitHub with watermark dedup.
serving-llms-vllm
vLLM: high-throughput LLM serving, OpenAI API, quantization.