Funny how the cycle seems to be coming full circle. We went from explicitly orchestrated agents, to SKILL.md files consumed inside a ReAct loop, and now back to compiling those skills into orchestrated agent workflows.
Ha ha, true. We already had the solution for compliance years ago with LangGraph, byLLM, etc. I think we went with SKILL.md because of the ease of authoring and use. SKILL.md or AGENT.md is genuinely easier to use, with a low barrier to entry for building custom workflows, though not very compliant. SIGIL seeks to address this dilemma by compiling agent skills.
Enforcing steps is fine, but mixing the orchestration layer into the methodology doc is just a mess. Skills should probably stay separate anyway, just keep instructions short
So true. A lot of skills out there already mix methodology and orchestration together. SIGIL’s AG-IR compilation helps extract the methodology and keep that separation clean.
Funny how the cycle seems to be coming full circle. We went from explicitly orchestrated agents, to SKILL.md files consumed inside a ReAct loop, and now back to compiling those skills into orchestrated agent workflows.
Ha ha, true. We already had the solution for compliance years ago with LangGraph, byLLM, etc. I think we went with SKILL.md because of the ease of authoring and use. SKILL.md or AGENT.md is genuinely easier to use, with a low barrier to entry for building custom workflows, though not very compliant. SIGIL seeks to address this dilemma by compiling agent skills.
Enforcing steps is fine, but mixing the orchestration layer into the methodology doc is just a mess. Skills should probably stay separate anyway, just keep instructions short
So true. A lot of skills out there already mix methodology and orchestration together. SIGIL’s AG-IR compilation helps extract the methodology and keep that separation clean.
Does it though? From the writeup it just looks like it pulls out the steps and gates. You're still authoring one SKILL.md either way.
[dead]
[dead]