List view

• Route with Hooks

  Hook-based routing: sentinel-injected instructions via PostToolUse hooks

  No due date

  •1/23 issues closed
  4% complete22 open1 closed

• Observability

  Complete observability coverage across all AutoSkillit systems — context utilization, skill/tool access, sub-agent tracking, guard enforcement. Get all diagnostic data captured in machine-readable formats (JSON, CSV, OTel-compatible) for downstream HTML report consumption.

  No due date

  •6/13 issues closed
  46% complete7 open6 closed

• Coordinator

  EHO Coordinator: AI-judgment-based oversight layer for monitoring and steering parallel agent sessions. See brainstorm gist for full design exploration.

  No due date

  •0/1 issues closed
  0% complete1 open0 closed

• Advanced Planning

  Progressive resolution planner recipe — 3-pass sequential decomposition into GitHub-issue-ready work packages. Adapts helper_agents multi-pass planning strategy to AutoSkillit's recipe model with sub-agent context filtering, sequential elaboration, and feature-gated delivery.

  No due date

  •56/59 issues closed
  94% complete3 open56 closed

• Feature Tagging

  Multi-layer feature gate system for gating experimental features (tools, skills, tests, imports, recipes) behind configurable flags. First test case: franchise. Enables safe promotion to main/stable with in-development features invisible.

  No due date

  •17/17 issues closed
  100% complete0 open17 closed

• Test Path Filtering

  Smart test path filtering system: AST-based import analysis + git diff + non-Python dependency manifest with bucket-based trigger system. Reduces test execution time in headless sessions by 40-76% while maintaining full CI coverage.

  No due date

  •21/53 issues closed
  39% complete32 open21 closed

• franchise mode

  Level-3 (L3) Franchise Orchestrator — interactive Claude session that dispatches headless L2 food trucks via dispatch_food_truck. Canonical plan: .autoskillit/temp/make-plan/franchise_l3_orchestrator_plan_2026-04-11_174213.md. Ticket-ready concretization at Addendum CA-1 through CA-20.

  No due date

  •26/58 issues closed
  44% complete32 open26 closed

• Multi-Provider Support

  Umbrella milestone for decoupling AutoSkillit from Claude Code as the sole backend. Covers two axes: **CLI Backend Abstraction** — support multiple agentic coding CLIs (Claude Code, Codex, Open Code) via a driver-based abstraction over headless sessions and interactive entry (`cook`/`init`), selectable by config. **LLM Provider Abstraction** — support multiple LLM providers (Anthropic, OpenAI, MiniMax, etc.) as the model behind sessions, including per-recipe-step provider selection. ## Tickets ### CLI Backend - #55 — Extract `CodingAgentBackend` protocol from current Claude Code coupling (prerequisite for all below) - #23 — Codex CLI backend support - #820 — Open Code CLI backend support ### LLM Provider - #772 — LLM provider abstraction layer - #773 — Per-recipe-step LLM provider selection - #770 — Evaluate MiniMax M2.7 ## Priority #55 is the top priority — the extraction layer enables everything else.

  No due date

  •44/60 issues closed
  73% complete16 open44 closed

• auto-research

  Research recipe generalization — domain-agnostic in-silico research platform

  No due date

  •82/99 issues closed
  82% complete17 open82 closed