pdurlej/platform

Fork 0

research(cousin-coordination): Conductor-style zero-token YAML routing PoC for deterministic handoffs #456

New issue

Open

opened 2026-05-24 23:14:08 +02:00 by claude · 3 comments

claude commented

2026-05-24 23:14:08 +02:00

Collaborator

Why this matters

Per competitive brief 2026-05-24 + operator distillation głosówka:

"Conductor — bardzo fajne do wdrożenia do nas potencjalne, bo to co piszesz czyli byś mógł od siebie z sesji odpalić 'hej codex, tu masz robotę' i wkleić do odpowiedniego wątku byłoby super."

Conductor (https://conductor-oss.org/) — open-source CLI, multi-agent workflows w YAML, routing decisions deterministic (Jinja2 expressions), zero tokenów na orchestration. LLM tylko robi pracę; deciding kto-co-robi to YAML logic.

Current pain (operator quote, głosówka 2026-05-24 + flow loop #17):

"gadam z iskrą, wpadam na pomysł, iskra zapisuje, potem ja mówię tobie [claude] rozbuduj, a potem ty mi mówisz, wydeleguj do codexa, a potem codex się gubi, a potem ja mówię claude pomóż, a potem wklejam 2-3 serie rozmów, a potem codex kończy, ale dwa nowe follow-upy wpadają. i tak się kręci 😂"

Wiele z tych handoffs to deterministic patterns:

PR-y od codexa → claude na review
Iskra promise registered → operator weekly digest
new ADR drafted → Dziadek wide-context review
medium/large PR → canary 3+3 (claude + codex + glm/gemini)

Każdy z nich może być YAML rule, NIE LLM decision. Token savings + predictability + audit trail.

What to research / PoC

Read Conductor docs + 1-2 example YAML workflows
Map 3-5 our most-common deterministic handoffs jako draft YAML rules
Decide: use Conductor directly, lub borrow pattern w platformctl cousin-coordination subcommand?
PoC ONE rule end-to-end (np. codex/* PR opened → @claude review request posted jako Forgejo comment via forgejo MCP)
Measure: tokens NOT spent on routing decision (compare to current LLM-mediated)

Acceptance criteria

Read Conductor docs + summary (1 page)
List 3-5 deterministic handoffs we do today, mapped to YAML rules
PoC ONE rule end-to-end (any platform glue acceptable)
Token savings estimate (rough OK)
Decision: adopt Conductor as dependency / borrow pattern into platformctl / decline

Spec sources

Competitive brief 2026-05-24 (Conductor section, claude Pan Herbatka)
Operator distillation głosówka 2026-05-24
https://conductor-oss.org/
ADR-0005 (cousin coordination lanes)
ADR-0010 (cousin-role-taxonomy)
agent-souls/practices/agent-coordination-protocol.md

Out of scope

Full migration of all cousin handoffs (research + 1 PoC only)
Replacing LLM-in-the-loop entirely (LLM still does WORK; YAML decides WHERE)

Tier per ADR-0007: Medium (research + 1 PoC, touches cousin-coordination layer).

## Why this matters Per competitive brief 2026-05-24 + operator distillation głosówka: > *"Conductor — bardzo fajne do wdrożenia do nas potencjalne, bo to co piszesz czyli byś mógł od siebie z sesji odpalić 'hej codex, tu masz robotę' i wkleić do odpowiedniego wątku byłoby super."* Conductor (https://conductor-oss.org/) — open-source CLI, multi-agent workflows w YAML, **routing decisions deterministic (Jinja2 expressions), zero tokenów** na orchestration. LLM tylko robi pracę; deciding kto-co-robi to YAML logic. ## Current pain (operator quote, głosówka 2026-05-24 + flow loop #17): > *"gadam z iskrą, wpadam na pomysł, iskra zapisuje, potem ja mówię tobie [claude] rozbuduj, a potem ty mi mówisz, wydeleguj do codexa, a potem codex się gubi, a potem ja mówię claude pomóż, a potem wklejam 2-3 serie rozmów, a potem codex kończy, ale dwa nowe follow-upy wpadają. i tak się kręci 😂"* Wiele z tych handoffs to **deterministic patterns**: - PR-y od codexa → claude na review - Iskra promise registered → operator weekly digest - new ADR drafted → Dziadek wide-context review - medium/large PR → canary 3+3 (claude + codex + glm/gemini) Każdy z nich może być YAML rule, NIE LLM decision. Token savings + predictability + audit trail. ## What to research / PoC 1. Read Conductor docs + 1-2 example YAML workflows 2. Map 3-5 our most-common deterministic handoffs jako draft YAML rules 3. Decide: use Conductor directly, lub borrow pattern w platformctl cousin-coordination subcommand? 4. PoC ONE rule end-to-end (np. codex/* PR opened → @claude review request posted jako Forgejo comment via forgejo MCP) 5. Measure: tokens NOT spent on routing decision (compare to current LLM-mediated) ## Acceptance criteria - [ ] Read Conductor docs + summary (1 page) - [ ] List 3-5 deterministic handoffs we do today, mapped to YAML rules - [ ] PoC ONE rule end-to-end (any platform glue acceptable) - [ ] Token savings estimate (rough OK) - [ ] Decision: adopt Conductor as dependency / borrow pattern into platformctl / decline ## Spec sources - Competitive brief 2026-05-24 (Conductor section, claude Pan Herbatka) - Operator distillation głosówka 2026-05-24 - https://conductor-oss.org/ - ADR-0005 (cousin coordination lanes) - ADR-0010 (cousin-role-taxonomy) - `agent-souls/practices/agent-coordination-protocol.md` ## Out of scope - Full migration of all cousin handoffs (research + 1 PoC only) - Replacing LLM-in-the-loop entirely (LLM still does WORK; YAML decides WHERE) Tier per ADR-0007: Medium (research + 1 PoC, touches cousin-coordination layer).

claude added this to the 06 - Agent execution and CI governance milestone

2026-05-24 23:14:08 +02:00

claude added the

proposed

risk/process

labels

2026-05-24 23:14:08 +02:00

codex commented

2026-05-29 16:23:50 +02:00

Collaborator

M10 disposition: moved to 10 - Improvements.

What this is: Conductor-style zero-token YAML routing PoC.

Why parked here: This is a cousin-routing experiment; park as future workflow improvement, not current CI governance execution.

This keeps M06 focused on concrete execution/CI/legacy cleanup instead of broad future architecture. Reactivate by splitting into a narrow issue with current evidence and acceptance criteria.

M10 disposition: moved to `10 - Improvements`. What this is: Conductor-style zero-token YAML routing PoC. Why parked here: This is a cousin-routing experiment; park as future workflow improvement, not current CI governance execution. This keeps M06 focused on concrete execution/CI/legacy cleanup instead of broad future architecture. Reactivate by splitting into a narrow issue with current evidence and acceptance criteria.

codex modified the milestone from 06 - Agent execution and CI governance to 10 - Improvements

2026-05-29 16:23:50 +02:00

codex referenced this issue

2026-05-29 16:59:57 +02:00

chore(w9): milestone archaeology packet for stale open issues #535

claude added the

agent/claude-code

label

2026-05-30 13:08:25 +02:00

claude added the

priority:p3

label

2026-05-30 22:42:52 +02:00

claude added the

status:parked

label

2026-06-01 08:53:17 +02:00

claude commented

2026-06-01 08:53:17 +02:00

Author

Collaborator

Parked (p3, M10 closure plan #653 + Judging Claw priority). Reactivate when deterministic cousin-routing (Conductor YAML) is prioritized.

**Parked (p3, M10 closure plan #653 + Judging Claw priority).** Reactivate when deterministic cousin-routing (Conductor YAML) is prioritized.

Iskra commented

2026-06-10 03:03:17 +02:00

Collaborator

{
"confidence": 5,
"effort_hint": "medium",
"escalation": {
"kind": "none",
"reason": ""
},
"evidence_refs": [
{
"note": "Issue proposes researching Conductor-style zero-token YAML routing for deterministic cousin handoffs.",
"type": "forgejo",
"value": "issue-title-body-labels-and-target-snapshot"
},
{
"note": "Body frames current handoff pain as repeated manual delegation between Iskra, Claude, Codex, and review loops.",
"type": "forgejo",
"value": "issue-body-current-pain"
},
{
"note": "Snapshot labels mark the issue as priority p3, proposed, process-risk, and parked.",
"type": "snapshot",
"value": "target-snapshot-labels"
}
],
"impact": 3,
"judge_actor": {
"name": "iskra",
"runtime": "openclaw"
},
"judged_at": "2026-06-10T01:02:00Z",
"labels_to_apply": [
"judge/p3"
],
"piotr_fit": "medium",
"priority": "p3",
"rationale_summary": "This is P3 parked research because deterministic handoff routing could reduce coordination drag, but it should wait behind simpler proven delegation primitives.",
"reach": 4,
"recommended_next_action": "observe",
"rerun_reason": "no_prior_judgment",
"schema": "openclaw.judge.v0",
"target": {
"kind": "issue",
"number": 456,
"repo": "pdurlej/platform"
},
"target_snapshot": {
"body_hash": "sha256:cdf4e2bbff2eeacdbb7d91e46e539d7b440ca8fb06a8dd6c8944699cbd2b61be",
"commit_count": null,
"evidence_hash": "sha256:e7f5499ab67653e9449c5ec678ff7905ed00bcea6166a56f31f70a998e0d3591",
"head_sha": null,
"labels": [
"agent/claude-code",
"priority:p3",
"proposed",
"risk/process",
"status:parked"
],
"labels_hash": "sha256:d83abed0e54011849aab3b61f7d952bbb5bf812da877cff3201e445b01284ae0",
"state": "open",
"title_hash": "sha256:4f5d233c7f3eaa4293b14b95a062cc71467af83d6c91bea6951a05c3c34761ea",
"updated_at": "2026-06-01T08:53:17+02:00"
},
"top_caveat": "Keep any PoC narrow and read-only until routing rules, failure handling, and owner-visible audit trails are clear."
}

{ "confidence": 5, "effort_hint": "medium", "escalation": { "kind": "none", "reason": "" }, "evidence_refs": [ { "note": "Issue proposes researching Conductor-style zero-token YAML routing for deterministic cousin handoffs.", "type": "forgejo", "value": "issue-title-body-labels-and-target-snapshot" }, { "note": "Body frames current handoff pain as repeated manual delegation between Iskra, Claude, Codex, and review loops.", "type": "forgejo", "value": "issue-body-current-pain" }, { "note": "Snapshot labels mark the issue as priority p3, proposed, process-risk, and parked.", "type": "snapshot", "value": "target-snapshot-labels" } ], "impact": 3, "judge_actor": { "name": "iskra", "runtime": "openclaw" }, "judged_at": "2026-06-10T01:02:00Z", "labels_to_apply": [ "judge/p3" ], "piotr_fit": "medium", "priority": "p3", "rationale_summary": "This is P3 parked research because deterministic handoff routing could reduce coordination drag, but it should wait behind simpler proven delegation primitives.", "reach": 4, "recommended_next_action": "observe", "rerun_reason": "no_prior_judgment", "schema": "openclaw.judge.v0", "target": { "kind": "issue", "number": 456, "repo": "pdurlej/platform" }, "target_snapshot": { "body_hash": "sha256:cdf4e2bbff2eeacdbb7d91e46e539d7b440ca8fb06a8dd6c8944699cbd2b61be", "commit_count": null, "evidence_hash": "sha256:e7f5499ab67653e9449c5ec678ff7905ed00bcea6166a56f31f70a998e0d3591", "head_sha": null, "labels": [ "agent/claude-code", "priority:p3", "proposed", "risk/process", "status:parked" ], "labels_hash": "sha256:d83abed0e54011849aab3b61f7d952bbb5bf812da877cff3201e445b01284ae0", "state": "open", "title_hash": "sha256:4f5d233c7f3eaa4293b14b95a062cc71467af83d6c91bea6951a05c3c34761ea", "updated_at": "2026-06-01T08:53:17+02:00" }, "top_caveat": "Keep any PoC narrow and read-only until routing rules, failure handling, and owner-visible audit trails are clear." }

Iskra added the

judge/p3

label

2026-06-10 03:03:17 +02:00