Progress
This page separates what foxctl currently documents as usable behavior from work that is still evolving through plans, evals, and benchmark validation.
Current
Section titled “Current”| Area | Current state | Docs |
|---|---|---|
| Production docs | Starlight site with Cloudflare Pages deploy path, docs build checks, link checks, and production verification | Verification |
| Repo retrieval | Semantic search, repoindex, DAG grep, codemaps, and evidence packaging | Repoindex and DAG grep |
| Skills runtime | Job-tracked, ephemeral, and direct execution paths with stable envelopes | Runtime and install |
| Agents and rooms | Agent lifecycle, orchestration, room timelines, mailbox asks, and collaboration flows | Agent orchestration |
| Storage | CAS, persistence, vectors, Turso, and Postgres behavior | CAS and persistence |
| Integrations | LLM providers, MCP, OpenAPI, plugins, hooks, chat adapters, and Obsidian bridge | Integration status |
| Benchmarks | Curated Go benchmark runner for hot packages and repeatable local captures | Benchmarks |
In progress
Section titled “In progress”| Area | Why it matters | Plan-backed status |
|---|---|---|
| Durable execution recovery | Keeps agent and runtime work recoverable after crashes | Active plan-backed runtime work |
| Runtime side-effect safety | Makes retries and event append behavior idempotent | Active plan-backed safety work |
| Refactor intelligence | Finds hotspots, repeated change patterns, and deterministic refactor targets | Active implementation and validation work |
| Slop function detection | Flags AI-generated sprawl and low-confidence code shape | Active detection backlog |
| RLM helper runtime | Supports recursive helper pipelines, LongCoT evals, and smolvm experiments | Experimental |
| ACA memory evolution | Improves self-corrective retrieval and memory derivation | Active research and plan-backed work |
| Room workpacks and milestones | Makes multi-agent work auditable through evidence lanes and exit policies | Active room workflow backlog |
| OpenSandbox integration | Adds sandboxed workspace execution as an adapter surface | Planned integration work |
Promotion rule
Section titled “Promotion rule”Do not describe in-progress work as current operator behavior until it has:
- A stable command or API contract
- Deterministic tests or evals for the expected behavior
- Documentation in the matching current docs area
- A benchmark or operational check when the feature affects hot paths
Plan documents remain useful for contributors, but current docs must stay tied to behavior that exists in the repository.