@@ -25,7 +25,7 @@ Today, writing agent instructions is fragmented on two axes:
2525 unstructured rationale all drop silently at 7B. You don't find out
2626 until the agent misbehaves in production.
2727
28- Seven packages solve that in one pipeline with a feedback loop:
28+ Eight packages solve that in one pipeline with a control layer and feedback loop:
2929
3030- ** Four build-time tools** turn your authored source into every harness's file layout:
3131 [ ` @razroo/agentmd ` ] ( ./packages/agentmd ) validates * structure* ,
@@ -34,6 +34,9 @@ Seven packages solve that in one pipeline with a feedback loop:
3434 [ ` @razroo/iso-route ` ] ( ./packages/iso-route ) compiles * one model policy* into each harness's config.
3535- ** One wrapper** runs the whole build chain:
3636 [ ` @razroo/iso ` ] ( ./packages/iso ) chains the above into a single ` iso build ` .
37+ - ** One orchestration library** handles durable runtime control:
38+ [ ` @razroo/iso-orchestrator ` ] ( ./packages/iso-orchestrator ) provides resumable
39+ steps, keyed mutexes, and bounded fan-out for side-effectful agent workflows.
3740- ** Two feedback tools** close the loop after deploy:
3841 [ ` @razroo/iso-eval ` ] ( ./packages/iso-eval ) scores * did the agent complete the task?* and
3942 [ ` @razroo/iso-trace ` ] ( ./packages/iso-trace ) parses production transcripts to show * what the agent actually did* .
@@ -155,6 +158,14 @@ the repo now supports a tighter loop:
155158 transcripts do not yet expose stable model metadata. Zero upload —
156159 everything is local reads and user-controlled output.
157160
161+ - ** [ ` packages/iso-orchestrator ` ] ( ./packages/iso-orchestrator ) ** — ` @razroo/iso-orchestrator `
162+ Durable orchestration primitives for the runtime layer above a single
163+ agent session. Persists workflow state to local disk, memoizes
164+ load-bearing ` step() ` results, provides keyed mutexes for "same
165+ entity" exclusion, and offers bounded ` forEach() ` fan-out so domain
166+ packages can move invariants out of prompt prose and shell scripts.
167+ Library-first today: no CLI, no harness-specific task-dispatch adapter.
168+
158169Each package is independently published on npm and works on its own.
159170They're in one repo because they're designed to compose.
160171
262273 ├── iso-harness/ # one source, every harness
263274 ├── iso/ # one command for the whole pipeline
264275 ├── iso-route/ # one model policy → per-harness config
276+ ├── iso-orchestrator/ # durable runtime control above one agent session
265277 ├── iso-eval/ # behavioral eval on the produced harness
266278 └── iso-trace/ # parse + query real agent transcripts (observability)
267279```
@@ -313,7 +325,7 @@ build, and `npm publish --provenance`.
313325## End-to-end example
314326
315327[ ` examples/pipeline/ ` ] ( ./examples/pipeline ) is an executable demonstration
316- that exercises ** all seven packages end-to-end** in one `npm run
328+ that exercises ** seven of the eight packages end-to-end** in one `npm run
317329test: pipeline ` invocation: ` agentmd lint` + ` render` → ` isolint lint` →
318330` iso-route build ` (from a bundled ` models.yaml ` that extends the
319331` standard ` preset) → ` iso-harness build ` (which consumes iso-route's
0 commit comments