spec-editor Autonomic is in early access. Request access
Paid plugin ยท early access
spec-editor Autonomic is a plugin for the open core. It adds the team that turns a requirement into reviewed, tested, released code, then watches it run and learns from the result. It works on your own subscriptions.
Features
Analysts, an analysis manager, a project manager, a coder, a tester, devops and an on-duty role for incidents.
Manager roles own the queues. A deterministic engine only moves events.
Analysts work on a source independently, and a second pass reconciles their results.
The result is checked against the source: the model does not receive what the source did not say.
A typo, a label or one line takes a short route instead of the full cycle.
One installation runs several projects, each with its own engine, on shared subscriptions with quotas.
Risk classes A to D, managed with spec-editor gate. A human is asked only where evidence is missing.
Every plan is deployed to a blue/green test stand, and its acceptance criteria are run there before promotion.
Screens are accepted by a screenshot taken on the stand, not by the structure of the page.
Rollback exists for a promoted plan.
Builds and stands can run on a remote host. Coders stay on the machine where the subscription is logged in.
Repeating errors in the logs are recognised by their signature and become incidents.
An on-duty role sorts incidents on an inexpensive model and escalates what looks serious.
An incident is bound to the requirement whose code wrote the record, and goes through the same analysis as any other source.
Classifies how hard a task is.
From cheap to strong.
Per source and per role.
Keep spend within a budget you set.
Work fails over to another account when one hits its limit.
Routine operator decisions are taken by policy, with a dry run first.
The KPIs of your model are computed from the events of the pipeline, with their history.
Once a day an observer looks for KPI deviations and proposes hypotheses.
Parameters of the pipeline change through declared experiments with a decision rule. Changes to code go through you.
Roles keep lessons from finished runs in role memory.
The instructions of a role are refined from the outcomes of its runs, and a learned rule is checked by replay before it is kept.
Visual comments right in your running product. A small JavaScript snippet is added to it; a click on an element of the screen creates a source bound to the requirement of that element.
Each role of the delivery team gets the tools and permissions its skills call for, and no more.
The coder works in the shell under a deny list of dangerous commands.
A review stage or a source is approved by a person, and the approval is recorded with who gave it.
Requirements restored from your existing codebase. Started from the interface.
spec-editor coverage shows coverage of requirements by code, both ways.
Works without an external service.
Python, TypeScript, JavaScript, Go, Java, Kotlin and Rust are parsed, so agents see functions and classes, not lines of text.
Assembled from the scenarios and screens of the model.
A chosen scenario is run on a chosen stand and recorded, with chapters and a poster.
From the interface.
In the context of an element.
A protocol for sharing anonymous aggregate statistics exists. It is opt-in.
Use
What happens
Uses: delivery team, check against the source, test stand, review stages
What happens
Uses: review stages, spec-editor gate
What happens
Uses: test stand, acceptance run, rollback
What happens
Uses: breakpoints, delivery team
What happens
Uses: reengineer, coverage, local semantic search
What happens
Uses: cost attribution, classifier, ladder of models, budget guards
What happens
Uses: pool of subscription accounts with failover
What happens
Uses: log signatures, on-duty role, hierarchical logging, delivery team
What happens
Uses: KPI evaluation, observer, experiments, roles that learn
What happens
Uses: user manual, demo video, test stand
What happens
Uses: remote host, test stand
Editions
| Capability | Open core | spec-editor Autonomic |
|---|---|---|
| Model and tools | ||
| Requirements as Markdown files in git, with acceptance criteria and validation | Included | Included |
| Sources from Jira, GitHub Issues, Slack, Discord and Telegram; import of OpenSLO files | Included | Included |
| Export: SRS in the style of IEEE 830, styled HTML, TRLC | Included | Included |
| Web interface, run log, run-time trace map and event graph | Included | Included |
| Editor extensions for VS Code, JetBrains IDEs and Zed | Included | Included |
| Hierarchical logging by the tree of requirements | Included | Included |
| KPIs as elements of the model | Included | Included |
| Model providers: Claude Code CLI, LiteLLM, Ollama | Included | Included |
| Command line | 29 commands | 36 commands |
| MCP server | More than 70 tools | More than 110 tools |
| Delivery | ||
| Analysts, analysis manager, project manager, coder, tester, devops, on-duty role | Not included | Included |
| Independent analysis, a reconciling second pass and a check against the source | Not included | Included |
| Review stages with risk classes A to D | Not included | Included |
| Test stand (blue/green) with acceptance run before promotion; rollback | Not included | Included |
| Visual testing: screens accepted by screenshot | Not included | Included |
| Monitoring: incidents from log signatures | Not included | Included |
| Self-improvement: KPI evaluation, observer, experiments, roles that learn | Not included | Included |
| Remote host for builds and stands | Not included | Included |
| Cost control: classifier, model ladder, attribution, budget guards, account pool with failover, automatic policies | Not included | Included |
| Fast path for small edits | Not included | Included |
| Several projects in one installation, with shared subscriptions | Not included | Included |
| User manual and demo video from scenarios | Not included | Included |
| Breakpoints | Not included | Included |
| Reengineer and spec-editor coverage | Not included | Included |
| Local semantic search over code | Not included | Included |
| Report a bug and chat with an agent from the interface | Not included | Included |
Ownership
Roadmap
The federation connector links the projects of one organisation. It will be sold apart from spec-editor Autonomic.
Early access
Purchase is not open yet. Leave an e-mail address and we will write when it opens. Prices are early and may change. Installation and setup of the plugin are described in its documentation.