Tracer: one role, YAML to bundle to launched agent, no Go rebuild #332

New issue

Closed

opened 2026-08-25 17:07:56 +00:00 by coilyco-ops · 0 comments

coilyco-ops commented

2026-08-25 17:07:56 +00:00

Member

Parent

#329

What to build

The end-to-end tracer. Author one role as YAML, compose it with a throwaway Python script into the bundle shape #331 proved, mount it with an unmodified acompose, and launch. Then edit the YAML, re-run the script, and watch the change reach a launched agent without rebuilding or re-releasing any Go.

That last sentence is the acceptance test for the entire program in #329, executed once and narrowly.

Pick tpm as the role, because it carries two personalities and four boundaries and therefore exercises meld resolution, boundary deferral, scoped grants, and ownership in one seat.

Scope discipline. The compositor here is throwaway. No packaging, no public API, no repo, no tests beyond what proves the path. Hard-code whatever is faster to hard-code. The deliverable is the proven seam, not code anyone keeps. Slice #333 rewrites it properly.

Acceptance criteria

One role, its two personalities, its four boundaries, and the invariant exist as YAML.
A Python script reads that YAML and emits a bundle that agent-compose verify accepts.
acompose mounts and launches it, and the launched agent carries the correct role skill, meld, and boundary text.
Editing a word in the YAML, re-running the script, and relaunching shows the edit in the agent, with no go build, go install, or release in between. Record the exact command sequence.
The YAML shape is written up as a first-draft schema, including which existing KDL contract each field descends from.
Every place the script cheats is listed, so #333 knows what it owes.

Blocked by

Blocked by #331

## Parent #329 ## What to build The end-to-end tracer. Author one role as YAML, compose it with a throwaway Python script into the bundle shape #331 proved, mount it with an unmodified `acompose`, and launch. Then edit the YAML, re-run the script, and watch the change reach a launched agent without rebuilding or re-releasing any Go. That last sentence is the acceptance test for the entire program in #329, executed once and narrowly. Pick `tpm` as the role, because it carries two personalities and four boundaries and therefore exercises meld resolution, boundary deferral, scoped grants, and ownership in one seat. Scope discipline. The compositor here is throwaway. No packaging, no public API, no repo, no tests beyond what proves the path. Hard-code whatever is faster to hard-code. The deliverable is the proven seam, not code anyone keeps. Slice #333 rewrites it properly. ## Acceptance criteria - [ ] One role, its two personalities, its four boundaries, and the invariant exist as YAML. - [ ] A Python script reads that YAML and emits a bundle that `agent-compose verify` accepts. - [ ] `acompose` mounts and launches it, and the launched agent carries the correct role skill, meld, and boundary text. - [ ] Editing a word in the YAML, re-running the script, and relaunching shows the edit in the agent, with no `go build`, `go install`, or release in between. Record the exact command sequence. - [ ] The YAML shape is written up as a first-draft schema, including which existing KDL contract each field descends from. - [ ] Every place the script cheats is listed, so #333 knows what it owes. ## Blocked by - Blocked by #331

coilyco-ops referenced this issue

2026-08-25 17:08:10 +00:00

Port the semantic layer from Go to the compositor, with the Go tests as the oracle #333

coilyco-ops referenced this issue

2026-08-25 17:08:18 +00:00

acompose refuses a stale bundle and never silently falls back #334

coilyco-ops referenced this issue

2026-08-25 17:08:37 +00:00

Convert all 22 roster entities from KDL to YAML #335

coilyco-ops referenced this issue

2026-08-25 17:10:24 +00:00

Invert the stack: data, compositor, bundle, then Go. Drop KDL, move composition out of the engine, mount the roster #329