coilysiren/website

Watch

Publish an agent-compose case study for portable agent context #85

New issue

Closed

opened 2026-08-03 00:51:51 +00:00 by coilyco-ops · 1 comment

coilyco-ops commented

2026-08-03 00:51:51 +00:00

Collaborator

Parent

Private Inbox epic

Actor

Design role.

Outcome

Publish an agent-compose case study that shows how Kai turns roles, personalities, skills, and tool inventories into coherent agent runtime context.

Context

agent-compose is a context compiler for AI-agent harnesses. Its value is not a large prompt. It is a controlled build step that selects the relevant operating doctrine and capabilities for a particular seat while keeping ownership, portability, and context budget visible.

This page supplies the composition and developer-experience layer of the larger portfolio story.

What to build

Lead with the context-drift and harness-fragmentation problem.
Show an input-to-output visual: source roles, personality meld, selected skills, tool inventory, then the composed harness bundle.
Explain why composition is a build-time contract rather than runtime prompt accumulation.
Show how role selection, bounded capability discovery, and inspectable output improve developer and operator experience.
Name one hard tradeoff, such as context completeness versus budget or shared doctrine versus role-specific context.
Use one concrete generated artifact or workflow as evidence after verifying it in the canonical repository.
Connect the output to Ward execution and Ward MCP capabilities.

Blocked by

Blocked by the hiring-page refresh so the case study can replace its temporary repository link cleanly.

Shared acceptance boundary

The page uses current canonical repository evidence rather than remembered architecture.
The page names the problem, architecture, hard design judgment, evidence, and current limit.
The page links to the canonical Forgejo source and the hiring page links back to this case study.
Metadata and social preview are intentional, and the page is legible on desktop and mobile.
No private project, opaque identifier, invented metric, or unsupported outcome appears.
Repository validation passes and the change lands through the repository's normal workflow.

Non-goals

Do not duplicate the repository README.
Do not write a changelog or exhaustive feature inventory.
Do not claim that telemetry alone creates trust.
Do not redesign unrelated site surfaces.

## Parent [Private Inbox epic](https://forgejo.coilysiren.me/coilysiren/inbox/issues/319) ## Actor Design role. ## Outcome Publish an agent-compose case study that shows how Kai turns roles, personalities, skills, and tool inventories into coherent agent runtime context. ## Context agent-compose is a context compiler for AI-agent harnesses. Its value is not a large prompt. It is a controlled build step that selects the relevant operating doctrine and capabilities for a particular seat while keeping ownership, portability, and context budget visible. This page supplies the composition and developer-experience layer of the larger portfolio story. ## What to build * Lead with the context-drift and harness-fragmentation problem. * Show an input-to-output visual: source roles, personality meld, selected skills, tool inventory, then the composed harness bundle. * Explain why composition is a build-time contract rather than runtime prompt accumulation. * Show how role selection, bounded capability discovery, and inspectable output improve developer and operator experience. * Name one hard tradeoff, such as context completeness versus budget or shared doctrine versus role-specific context. * Use one concrete generated artifact or workflow as evidence after verifying it in the canonical repository. * Connect the output to Ward execution and Ward MCP capabilities. ## Blocked by Blocked by [the hiring-page refresh](https://forgejo.coilysiren.me/coilysiren/website/issues/83) so the case study can replace its temporary repository link cleanly. ## Shared acceptance boundary * [ ] The page uses current canonical repository evidence rather than remembered architecture. * [ ] The page names the problem, architecture, hard design judgment, evidence, and current limit. * [ ] The page links to the canonical Forgejo source and the hiring page links back to this case study. * [ ] Metadata and social preview are intentional, and the page is legible on desktop and mobile. * [ ] No private project, opaque identifier, invented metric, or unsupported outcome appears. * [ ] Repository validation passes and the change lands through the repository's normal workflow. ## Non-goals * Do not duplicate the repository README. * Do not write a changelog or exhaustive feature inventory. * Do not claim that telemetry alone creates trust. * Do not redesign unrelated site surfaces.

coilyco-ops added the

autonomy

headless

priority

labels

2026-08-03 00:53:15 +00:00

coilyco-ops commented

2026-08-03 02:05:16 +00:00

Author

Collaborator

Landed in ffc01ca433. /work/agent-compose/ now documents the context problem, build-time composition path, context-budget judgment, shipped evidence, current collision limit, and its connection to Ward and Ward MCP. The page has intentional metadata and a fitted social card.

Landed in https://forgejo.coilysiren.me/coilysiren/website/commit/ffc01ca433ca54f4af1f574465a0ccd652b2d0fd. `/work/agent-compose/` now documents the context problem, build-time composition path, context-budget judgment, shipped evidence, current collision limit, and its connection to Ward and Ward MCP. The page has intentional metadata and a fitted social card.

coilyco-ops closed this issue

2026-08-03 02:05:17 +00:00

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

coilysiren/website#85

No description provided.

Rows
Columns