/hermes3000-writing

name: hermes3000-writing
description: Use Hermes3000 to plan, draft, revise, save, check consistency, and export long-form manuscripts through the Hermes3000 AI writing portal API. Use for novels, fiction series, nonfiction books, whitepapers, long reads, chapter outlines, character/world-building, style guides, consistency memory, and DOCX/PDF/EPUB/HTML exports.
requires:
  bins:
    - node
credentials:
  - id: hermes3000-jwt
    kind: bearer
    required: false
    secret_ref:
      source: store
      id: HERMES3000_JWT
    scope: "hermes3000.ai"
    how_to_obtain: "Set `HERMES3000_EMAIL` and `HERMES3000_PASSWORD` through browser admin at the active `/admin/secrets` route; if browser admin is unavailable, use `/secret set HERMES3000_EMAIL ...` and `/secret set HERMES3000_PASSWORD ...` in browser `/chat` or TUI; local console fallback: `hybridclaw secret set ...`. Then run the bundled auth.login helper so the gateway captures the login JWT into `HERMES3000_JWT` without exposing it to the agent."
metadata:
  hybridclaw:
    category: productivity
    short_description: "Hermes3000 portal workflow for long-form writing."
    tags:
      - hermes3000
      - writing
      - novels
      - whitepapers
      - long-reads

Hermes3000 Writing

Overview

Use Hermes3000 when the user wants a durable long-form writing project managed in the writing portal, not just a one-off draft in chat. Hermes3000 stores book structure, chapter content, consistency memory, timeline data, stats, and exports.

For endpoint details, request shapes, and examples, read `references/api-workflow.md` before making live API calls.

Secret Refs

The bundled helper sends all live Hermes3000 calls through the HybridClaw gateway proxy, matching the Salesforce skill pattern. Login credentials are sent as `<secret:NAME>` placeholders; the gateway resolves them server-side, captures the returned Hermes3000 JWT into `HERMES3000_JWT`, and returns only a capture confirmation. The JWT does not enter the agent context.

Never call Hermes3000 with `curl` or a raw `Authorization: Bearer ...` header. Use `node skills/hermes3000-writing/scripts/hermes3000.cjs ... run ...` for live calls so the gateway injects and captures secrets server-side. Run these helper commands exactly as documented. Do not add Node permission flags such as `--experimental-permission`, and do not rewrite `skills/...` paths to `/workspace/skills/...`; the helper needs runtime-provided gateway access for secret resolution.

Cloud and chat/TUI users may not have a terminal. Only ask the user to set secrets with chat/TUI secret commands; do not ask them to run `node`, `curl`, or `hybridclaw` shell commands. If `HERMES3000_JWT` is missing or blocked, run `auth.login` yourself through the helper. The helper does not prompt for email or password; it resolves `HERMES3000_EMAIL` and `HERMES3000_PASSWORD` from the gateway secret store.

Ask the user to set the login inputs once through browser admin first:

1. Browser admin: open the active HybridClaw admin URL ending in `/admin/secrets` and set `HERMES3000_EMAIL` and `HERMES3000_PASSWORD`. 2. Browser `/chat` or TUI fallback:

/secret set HERMES3000_EMAIL you@example.com
/secret set HERMES3000_PASSWORD <password>

3. Local console fallback:

hybridclaw secret set HERMES3000_EMAIL you@example.com
hybridclaw secret set HERMES3000_PASSWORD '<password>'

Then the agent, not the user, captures the JWT through the gateway:

node skills/hermes3000-writing/scripts/hermes3000.cjs --format json run auth.login

On success, the gateway stores `HERMES3000_JWT`. Subsequent helper operations set `bearerSecretName: "HERMES3000_JWT"` internally so the gateway injects the bearer token server-side.

Operating Rules

1. Preserve author control. Ask for or infer the manuscript premise, audience, language, target length, genre or content type, tone, and export format before creating or replacing durable portal content. 2. Treat credentials and JWTs as secrets. Authenticate through the active runtime's secret or credential mechanism when available. Do not print tokens, paste passwords into chat, or store credentials in project files. 3. Prefer the synchronous generation endpoint for agents: `POST /ai/generate-text`. Use streaming only when the active runtime has a clear SSE reader and the user needs progress output. 4. Save every accepted chapter draft with `POST /books/{id}/content`, then build consistency memory with `POST /consistency/chapter-summary` before writing the next chapter. 5. Use stable chapter UUIDs. When AI-generated chapters do not include `id`, generate UUID v4 values before saving the `chapters` structure. 6. Keep portal state coherent. Update `plot`, `characters`, `style`, and `chapters` before generating chapter prose; update story progress after summaries with `POST /consistency/update-story`. 7. Do not overwrite an existing book, chapter, structure element, or content block unless the user explicitly asks for replacement. Fetch current state first when editing an existing project.

Project Shape

Choose the Hermes3000 `bookType` from the user's goal:

• `prose`: novels, novellas, fiction series, narrative long reads.
• `nonfiction`: nonfiction books, educational material, manuals, essays.
• `whitepaper`: short-form whitepapers, reports, thought-leadership papers.

Represent structure as:

• `plot`: Markdown outline of the argument, storyline, or thesis.
• `characters`: Markdown profiles for fiction characters, or key people,

personas, stakeholders, examples, and cited experts for nonfiction.

• `style`: Markdown writing rules: language, voice, perspective, reading level,

terminology, citation stance, formatting preferences, and banned patterns.

• `places_things`: Markdown world-building, concepts, definitions, datasets,

products, organizations, or repeated motifs.

• `chapters`: Array of `{id, title, summary}` objects