Comet consolidates ideas, state, evidence, and Skill distribution onto a single resumable workbench
## What You Can Do with Comet * Use `/comet` to enter Native or Classic from project configuration and drive a product or code change from idea to archive. * Use `/comet-any` to distill your own workflow into a reusable Skill. * Use `comet eval` to generate pre-publish evidence for a Skill. * Use `comet publish` to review, approve, publish, and distribute `/comet-any` artifacts. * Use `comet status` and `comet doctor` to see the current phase, next step, and recovery recommendations. ```mermaid theme={null} flowchart LR A["Product idea or fix"] --> B["/comet configuration route"] B --> N["Native: strong-model autonomy"] B --> C["Classic: OpenSpec + Superpowers"] N --> E["Verification and archiving"] C --> E F["Reusable workflow idea"] --> G["/comet-any Skill Creator"] G --> H["comet eval evidence"] H --> I["comet publish publishing and distribution"] ``` ## What's Different from Ordinary Prompts Ordinary prompts rely on conversation history. Comet relies on file-based state and inspectable evidence. | Concern | Without Comet | With Comet | | -------------------------- | ------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Interruption recovery | The Agent must recall context, spending tokens before implementation resumes | Recovers from `.comet/config.yaml`, workflow-specific state, and runtime evidence instead of guessing from the original chat | | Phase boundaries | The Agent may skip requirements or verification and modify implementation at the wrong time | Native constrains outcomes and evidence across Shape/Build/Verify/Archive; Classic supplies finer five-phase process constraints | | Auto-advancement | Requires remembering Skill order and taking over between steps | Both workflows resume from disk phase and continue when conditions pass, pausing for user decisions or safety blockers | | Constraint strength | One process rarely fits every model capability tier | Strong models use lighter-method Native; lower capability tiers use Classic, with hotfix/tweak presets for smaller Classic work | | Skill artifact form | Just a single `SKILL.md` file — can't be composed, can't carry dependencies, can't be distributed | `/comet-any` generates a stable, composable **Skill Bundle** — rules, hooks, and Engine metadata can be assembled; it's **reviewable** and **distributable** (landed on 33 AI coding platforms via `comet publish distribute`, just like `comet init`) | | Skill quality verification | Relies on feel — if the Agent runs it once without errors, it passes | `comet eval` produces a **browsable evaluation report**; eval/review/approval/readiness are all **bound to the current hash**, with traceable evidence |
Comet doesn't guess the scene from conversation memory — it recovers work from file-based state and an evidence chain
## Main Entry Points Invoke `/comet` for project changes. It reads project configuration and enters the Native or Classic workflow below. Workflow-specific Skills are permanent internal entries and advanced debugging surfaces, not commands users normally need to remember.