Communitygithub.com

forgegod/daidala

A Hermes-native AI workshop for crafted, human-approved work through specialist agents and skills.

¿Qué es daidala?

daidala is a Claude Code agent skill that a Hermes-native AI workshop for crafted, human-approved work through specialist agents and skills.

Compatible con~Claude Code~Codex CLI~Cursor
npx skills add forgegod/daidala

Preguntar en tu IA favorita

Abre un nuevo chat con esta habilidad de agente ya precargada.

Documentación

Daidala Setup

Overview

Guide a user from an installed Daidala plugin to one explicitly confirmed workflow start. Setup prepares and starts; daidala:orchestrate remains the worker contract. This skill works without the web dashboard and uses the same daidala_start request in dashboard and non-dashboard sessions.

When to Use

Load this skill explicitly as daidala:setup when the user wants to install, check, configure, or start Daidala. Do not use it for stage execution or for resuming an already-created workflow.

Procedure

  1. Check the active Hermes version, Daidala plugin state, gateway or dispatcher availability, selected pack, exact required skills, available profiles, and named Kanban boards. Use documented Hermes surfaces and daidala_pack_info; do not inspect private databases or infer names.
  2. Present missing prerequisites with exact commands or tool calls. Installation, board creation, pack setup, and other mutations require a separate preview and explicit confirmation before execution. Never request or display credentials.
  3. Ask the user to select an existing named board. Offer explicit board creation only through the documented Hermes Kanban operation. If creation is declined, keep setup read-only.
  4. Ask for one default existing profile for every executable stage. Show the complete mapping and ask only for explicit overrides of define, plan, implement, verify, review, or deliver.
  5. Ask for the pack, absolute target repository, stable workflow ID, and complete goal. Do not infer a path from prose or generate an unstable identifier.
  6. Ask the user to choose exactly one policy source: no constraints; explicit validated YAML content; or an exact installed policy skill plus its verified complete-directory SHA-256 digest. Never infer a policy source.
  7. Build and display the exact request below. Validate required fields and show the selected board, pack, repository, workflow ID, full stage-profile map, and constraint provenance. Do not call daidala_start yet.
  8. Ask the user to explicitly confirm that exact preview. A general request to configure Daidala is not confirmation. If the preview changes, show it again and obtain fresh confirmation.
  9. After confirmation, call daidala_start exactly once. Report the returned workflow and card identities and point the user to native Kanban. If the dashboard is available, /daidala is an optional observation surface; its absence never changes the request or blocks setup.
  10. Stop after start. Hermes Kanban dispatches workers using daidala:orchestrate; setup does not execute lifecycle stages.

Exact Start Request

Use only the current daidala_start argument names:

  • board_slug: selected existing board;
  • target_repository: absolute clean local Git checkout;
  • goal: complete development goal;
  • stage_profiles: object with exactly define, plan, implement, verify, review, and deliver, each mapped to an existing profile;
  • pack: selected bundled pack;
  • workflow_id: explicit stable identifier;
  • optional constraints_content; or optional constraints_skill together with constraints_skill_digest.

constraints_content and constraints_skill are mutually exclusive. A constraints_skill_digest without constraints_skill is invalid. Do not add aliases, omit executable stages, or translate this request into a second setup schema.

Common Pitfalls

  • Starting before the exact preview receives explicit confirmation.
  • Treating a missing dashboard as a setup failure.
  • Guessing board, profile, repository, workflow ID, or policy source.
  • Using --profile as Daidala's default-profile input; it is a Hermes host option. The CLI uses --default-profile and optional --stage-profile.
  • Installing skills or creating boards without a separate mutation preview.
  • Starting stage work in the setup session instead of leaving it to Kanban.
  • Copying pack lifecycle data instead of calling daidala_pack_info.

Verification Checklist

  • Supported Hermes and enabled Daidala plugin checked.
  • Pack and exact required skills checked.
  • Existing board and profiles selected explicitly.
  • Absolute repository, stable workflow ID, goal, and policy source explicit.
  • Preview contains the complete current daidala_start request.
  • Human explicitly confirmed the unchanged preview.
  • daidala_start called exactly once after confirmation.
  • Dashboard availability did not alter the request.
  • Setup stopped after returning workflow and card identities.

Skills relacionados