v2.73.3
Binaries and the merged-pull-request list are on the GitHub releases page.
This release is the fourth milestone (PD4) of the documentation epic that re-frames orun as platform discipline as code. It covers pillar 3, ground agents: coding agents that read the same intent your platform does, so they work inside your standards instead of imitating the nearest example. The binary is unchanged from v2.73.2.
Agents in orun repositories
Agents in orun repositories is now the pillar's anchor. It opens with two tables:
- How an agent learns the standards: the intent and component manifests, the compositions, the catalog, the base literacy, skills, and the task contract, and the command or tool that serves each.
- What keeps it inside them: the golden path's schema and rules
(
orun plan), the agent type's tool policy (the runtime), the task contract'saffects(orun task check), and provenance (orun pr checkand the platform's compliance check).
The agent runtime, task plane, service catalog, and the orun mcp,
orun agent, orun skills, orun task, and orun work pages open with
what each does for grounding.
Corrections
- An agent type's
mayAffectwas described as a blast-radius ceiling. It is recorded and shown in the cockpit, but the runtime does not check it; the task contract'saffects, checked byorun task check, is the enforced limit. orun pr checkwas described as checking a change against its contract. It checks provenance: the branch name, theOrun-Tasktrailer on every commit, and one task per pull request.- The MCP pages no longer hard-code tool counts;
orun mcp toolsprints the roster your binary serves. orun mcp doctoris now documented.
The context pack
context-for-ai/, the copyable pack for agents, now uses the schema's field
names (parameters, not inputs; parameterDefaults, not defaults),
describes policies as recorded rather than enforced, and gains
06-standards-presets-and-agent-guardrails.md: which standards orun
enforces, how presets and Stacks change them, and the guardrails around agent
work.