hermes-codex-plugin

name	hermes-codex-plugin
description	Use when a task may depend on prior user preferences, project conventions, architectural rules, workflow decisions, previous Codex work, full-text memory search, durable rule saving, or skill drafting/updating through Hermes Codex Plugin, especially to reduce Codex token spend by recalling minimal relevant context.

Hermes Codex Plugin

Use the bundled MCP tools to inspect and manage local memory. Treat memory as a standing source of user preferences, project conventions, architectural rules, workflow decisions, similar solved tasks, and existing reusable skills. Before solving a task, prefer checking whether memory or a skill already contains the relevant rule or workflow, not only when the user explicitly says to search memory. The main benefit is keeping Codex token usage low by retrieving only the relevant memory needed for the current request.

Workflow

1. Search first with hermes_codex_search_chats when prior context from earlier chats may matter.
2. Use hermes_codex_search for narrower local memory search when a project directory filter is useful.
3. Apply durable rules from memory as constraints while editing code, including style and import-placement rules.
4. Save durable facts with hermes_codex_remember.
5. After substantial tasks, write a concise structured task summary and save it with hermes_codex_remember_summary.
6. Use Codex's semantic judgment across languages to decide whether repeated workflow rules should become a skill draft; do not rely only on English keywords.
7. Use hermes_codex_propose_skill to draft a skill from repeated rules.
8. Only write a skill with hermes_codex_write_skill when the user clearly wants it.

Rules

• Keep memories compact and actionable.
• Prefer structured summaries over raw chat transcripts for long-term task memory.
• Do not save secrets, credentials, raw logs, or huge code blocks.
• Treat skill drafts as proposals that need review before use.
• Prefer project-scoped memories for repository conventions and user-scoped memories for stable personal preferences.