Hermes Agent Skills

Two community-vetted skills for Hermes Agent, the self-improving AI agent by Nous Research.

Skill	Description	Codex Score	Status
feishu-document-api	Create, write, and manage Feishu/Lark documents programmatically via the Open API	8/10	✅ Hub-ready
feishu-response-format	Standardized Feishu/Lark response formatting: message types, pipe tables vs interactive cards, Markdown compatibility, MEDIA attachments, Card JSON 2.0 table component	8/10	✅ Hub-ready
hermes-memory-maintenance	Maintain MEMORY.md/USER.md: audit redundancy, GATE rules, quality scoring, drift recovery	9/10	✅ Hub-ready

All skills were audited by Codex (GPT 5.5) source-level review (+ Claude Code review for feishu-response-format) and fixed per findings before submission.

Installation

# Install directly from this repo
hermes skills install https://github.com/duruonanni/hermes-skill-kit/tree/main/feishu-document-api
hermes skills install https://github.com/duruonanni/hermes-skill-kit/tree/main/feishu-response-format
hermes skills install https://github.com/duruonanni/hermes-skill-kit/tree/main/hermes-memory-maintenance

Skills

feishu-document-api

Programmatic Feishu/Lark document creation via the Open API. Covers:

• OAuth 2.0 tenant access token flow
• Block types (text, heading, bullet, code, quote, table)
• Mixed-type batch ordering (consecutive-run pattern)
• Ownership transfer (create → transfer → reveal URL)
• 50-block-per-POST limit handling
• Table creation (block_type 31, max 9 rows)
• In-place document sync (delete-all → rewrite)
• Cron job integration

Scripts: scripts/create_doc.py (simple), scripts/create_structured_doc.py (full workflow) References: 5 docs covering ownership transfer, in-place sync, table discovery, cron patterns, markdown conversion

feishu-response-format

Standardized Feishu/Lark response formatting. Covers:

• Message type selection guide (text / post / interactive / MEDIA:)
• 3 table schemes: pipe tables (🥇 default), interactive cards (Column Set), Card JSON 2.0 Table component (🆕)
• Markdown compatibility matrix (what works and what doesn't on Feishu)
• MEDIA attachment conventions (absolute path rules)
• P0/P1/P2 graded pitfalls ("一表毁所有", Token expiry, security scanner bypass)
• Output checklist for quality assurance
• 53 unit tests for all pure functions

Scripts: templates/send_card.py — interactive card sender with full type hints, HTTP 429 retry, stdin secret pipe Templates: templates/reply_templates.md — multi-step, verification loop, before/after, research summary References: 7 docs covering msg types, API call patterns, Markdown support, pipe table alignment, troubleshooting

hermes-memory-maintenance

Maintain Hermes agent memory files. Covers:

• §-separated entry format (drift detection & recovery)
• GATE-style hallucination prevention rules
• Memory entry quality scoring (0-4)
• Cross-file redundancy detection (MEMORY.md vs USER.md)
• Capacity planning & growth prediction
• Write discipline & aggregation patterns
• Multi-user section-based structure
• Frozen snapshot architecture
• Concurrent write safety (fcntl flock + atomic write)

Scripts: scripts/memory_review.py — automated weekly audit References: consolidation examples, drift recovery guide

Verification

Skills are tested locally and pass:

• skills_list discovery ✅
• skill_view content loading ✅
• Python syntax validation (all scripts) ✅
• Real execution against local memory files ✅

License

MIT — feel free to use, modify, and share.