Development

Read in this order:

1. Coding Agents: using the Agently-Skills companion repo with Codex, Claude Code, Cursor, and similar tools.
2. Skills Executor: framework-side runtime skill consumption through Agent APIs, plans, and Actions.
3. Documentation Update Principles: how the main repository docs act as the shared source for repository and website documentation.
4. Agently 4.1.3.6 Release Notes: AgentExecution ownership, Result-first consumption, stream-end hardening, and bounded task-loop slice.
5. Agently 4.1.3.5 Release Notes: settings-owned output defaults, meaningful required values, AgentTurn prompt isolation, and set_turn_prompt(...).
6. Agently 4.1.3.4 Release Notes: structured output parsing hardening, request retry, runtime capability policy, and AgentTaskLoop first public slice.
7. Agently 4.1.3.3 Release Notes: typed settings/options, model profiles, API key pool failover, runtime handler ownership, core package refactors, and image input.
8. Agently 4.1.3.2 Release Notes: bounded AgentExecution task steps, Workspace-backed step context, runtime stall control, and EventCenter RuntimeEvent delivery.
9. Agently 4.1.3.1 Release Notes: Workspace foundation, Recall skeleton, and explicit multi-turn task information management.
10. Agently 4.1.3 Release Notes: final 4.1.3 runtime goals, user-facing code shape, and business value.
11. Release Workflows: current repository automation for docs, installers, and PyPI publishing.

DevTools belongs to Observability because it consumes observation events. Action, MCP, and service APIs live in their own folders.