ACA context architecture
Status: Current architecture guide.
ACA is foxctl’s context architecture for combining repo evidence, memory, sessions, and knowledge artifacts without turning every note into an instruction.
The short version:
ACA control plane -> top-of-mind, handoffs, observations, tensions, proposals
Knowledge plane -> Obsidian notes, vault index, bridge metadata, reviewed promotions
Retrieval plane -> repoindex hints, semantic note search, contextengine evidence packsWhat ACA coordinates
Section titled “What ACA coordinates”| Layer | Purpose |
|---|---|
| Repo evidence | Code, docs, symbols, graph edges, and snippets |
| Memory | Reusable facts, decisions, and task continuity summaries |
| Obsidian bridge | External knowledge graph and vault reconciliation |
| Retrieval policy | Evidence ranking, provenance, and safety boundaries |
| Agent surface | Context bundles consumed by skills and agents |
Current command surface
Section titled “Current command surface”| Need | Command family |
|---|---|
| Current workspace orientation | foxctl orient, foxctl context show, foxctl context report |
| Capture continuity | foxctl capture, foxctl context task-history-summary |
| Retrieve blended context | foxctl context retrieve |
| Inspect retrieval misses | foxctl context retrieve-inspect, foxctl context retrieve-inspect-suite |
| Review durable proposals | foxctl context proposals, foxctl context proposal merge |
| Refresh vault knowledge | foxctl obsidian index build, foxctl obsidian graph build |
| Reconcile docs and vault notes | foxctl obsidian bridge reconcile, bridge apply, bridge tidy |
Typical refresh flow
Section titled “Typical refresh flow”After repo docs, repo graph, or bridge metadata changes:
foxctl obsidian graph build --workspace . --vault-path "/path/to/vault"foxctl obsidian graph promote --workspace . --vault-path "/path/to/vault"foxctl obsidian bridge reconcile --workspace . --vault-path "/path/to/vault"foxctl obsidian index build --vault-path "/path/to/vault"Relationship to contextengine
Section titled “Relationship to contextengine”Context engine is the typed evidence substrate. ACA decides which workspace state and knowledge should be retrieved; contextengine can record evidence packs, retrieval episodes, feedback, impact edges, and staleness markers.
Use ACA when the question is “what does this workspace know right now?” Use contextengine when the feature needs typed evidence, lane fusion, or retrieval feedback records.
Production boundaries
Section titled “Production boundaries”- Evidence is not instruction unless a policy explicitly promotes it.
- Generated summaries need provenance.
- Retrieval should prefer typed signals over ad hoc keyword routing.
- Planned memory-promotion behavior must stay labeled planned until promoted to current docs.
- Vault bridge operations draft and patch metadata deliberately; they do not silently rewrite canonical prose.