Active and shipped feature specifications for Honcho. Naming convention: <name>.md is the active spec; <name>.v1.md, <name>.v2.md are prior versions kept for history. Shipped specs move into done/ and are rewritten to reflect the as-shipped state.

Each entry links its Linear issue where one exists (IDs are stable; check Linear for live status).

See also: the Q2-2026 roadmap is archived at archive/roadmap-q2-2026.md (historical); a replacement is being defined in the operating-model reorg.

Active specs

Core API & data model

Agents

Observability

  • telemetry-analytics.md — Instrumentation hardening (Sentry, Prometheus→OTel, accumulate_metric audit). → no live Linear epic (orphan; work scattered in the retired Telemetry Hardening project)

Evaluation & quality

CLI / tooling / SDKs

  • cli.md — CLI & repo restructuring (v2). Supersedes cli.v1.md. → DEV-1534
  • sdk-improvements.md — Typed filter validation, cursor-based pagination, quality enhancements. → pagination shipped (DEV-1186); type-safety DEV-1398 (SDK Quality project)
  • static-schema-hosting.md — Host a JSON Schema for the shared client app config. → DEV-1678
  • mocking-server.md — Honcho mock server for SDK/CI testing (SQLite, no deriver). Harness-plugin testing is covered separately by the committed sandbox; see Harness plugin testing strategy. → DEV-1655 (backlog)

Documentation / GTM

  • readme-seo-changes.md — Honcho README competitive audit. → DEV-1758 (shipped — pending move to done/)

Shipped (done/)

  • [[llm-client-refactor|done/llm-client-refactor.md]] — Decomposition of src/utils/clients.py into src/llm/ package with native-SDK ProviderBackend abstraction. Litellm rejected during implementation. Predecessor: [[llm-client-refactor.v1|done/llm-client-refactor.v1.md]].
  • [[cloudevents-enrichment|done/cloudevents-enrichment.md]] — CloudEvents payload enrichment.
  • [[configurable-embeddings|done/configurable-embeddings.md]] — Configurable embedding provider/model.
  • [[deriver-polling-resilience|done/deriver-polling-resilience.md]] — Deriver queue polling resilience.

Integration specs

Integration specs (e.g. for partner SDKs) live one level up at projects/honcho/integrations/:

  • [[ai-sdk-spec|integrations/ai-sdk-spec.md]] — Vercel AI SDK integration
  • [[ai-sdk-spec-revisions|integrations/ai-sdk-spec-revisions.md]] — Revisions to the AI SDK spec

Workflow

  • New specs: write under specs/, status Draft, owner field set
  • Major revisions: rotate the existing file to <name>.v1.md (or next number) and write a new <name>.md
  • Shipping: move to done/ and rewrite to describe the as-shipped state with a “history” pointer to the prior .vN.md

See Wiki Conventions for the broader spec → Ultraplan → phased PRs flow.

22 items under this folder.