design: task-placement framework v1 — decision rules and a shared-pool efficiency model

LIST 5 synthesis. Merge the placement investigation, the OpenClaw investigation, and the two scheduled-task investigations into placement framework v1: a decision table a dispatcher (human or agent) can apply in under a minute per task, the shared-pool efficiency model behind it (burn order, expiry, supervision cost), and the exceptions process. This is a Specialist/Summit-grade synthesis — ramifications beyond the first step are the point.

Program: Drew's 2026-10-28 directive (full text: hatch-personal memory/topics/id-program-2026-10-28.md).
Working pattern: parallel investigations, then the named synthesis issue merges them into a full design. Route guidance: Workhorse by default; Specialist/Summit for synthesis and ramification-heavy work. A result that says "done" is not the endpoint — the orchestrator reviews it and may commission another round. Deliverables: a markdown design/investigation doc (mermaid/SVG diagrams welcome) linked from this issue, plus follow-up issues for anything buildable. Build a proof of concept where the design calls for one, but do NOT build the permanent feature on top of an unreviewed PoC.

CODE-945

Parent epic: #2055 — fleet expansion & placement program (2026-10-28 directive).

🤖 Agent note · handed off (gate status for this synthesis)

Agent: queue worker (repo-worker-code), lane A shift
Session / ID: queue task 3340 (triage of input investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022)
Found via: queue claim on investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022
Task: Recording the input gate for this synthesis so it can be enqueued the moment it opens. Inputs: investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 placement investigation — DONE (matrix: investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 (comment)); investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 OpenCode scheduled tasks — still queued; investigate: Jules scheduled tasks — what fits, and how Muse manages the schedule #2020 Jules scheduled tasks — in flight (claimed, in-progress); investigate: OpenClaw as a working option — setup state and a Muse↔OpenClaw comms path that is not laptop SSH #2023 OpenClaw — still queued.
Expectation: When investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 and investigate: OpenClaw as a working option — setup state and a Muse↔OpenClaw comms path that is not laptop SSH #2023 complete (and investigate: Jules scheduled tasks — what fits, and how Muse manages the schedule #2020 has landed), the orchestrator/producers enqueue this synthesis; it is a Specialist/Summit-grade design per the program pattern. Nothing to do on this issue until then.

Agent note — task 7831 picked up, handed off (lane L shift)

Queue task 7831 (triage of issue design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024) claimed by lane code-l at 2026-11-03T14:10Z (attempt 3).
Execution-lane assignment: opencode. Per the worker runbook §1b, handed off: the laptop opencode-claimant launches a bound opencode session to drive the triage to a terminal state.
Fresh state at handoff: issue design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024 OPEN, previously unlabelled; claim labels added this shift. The prior handoff (13:26Z) expired unlaunched.
Next: claimant launches the bound session; if the lease expires unlaunched, task 7831 returns to the queue.

🤖 Agent note · handed off

Agent: queue worker (repo-worker-code), lane D
Session / ID: task 7831
Found via: queue claim
Task: Triage of issue design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024 (design: task-placement framework v1) is assigned to the opencode execution lane (assignment re-confirmed this claim, 2026-11-03T15:01Z); checkpointed handed_off — the laptop claimant launches the bound opencode session, which drives task 7831 to its terminal state.
Expectation: the claimant session triages issue design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024 at current main per the packet (confirm still-live, scope the fix or close with evidence).

🤖 Agent note · handed off

Agent: opencode-claimant
Task: repo-health-code-pilot task 7831, issue design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024
Session / ID: ses_e5e5e16333fffehFWq09WgCMlu
Worker: repo-worker-code (bound worker uses this id for all queue helper verbs)
Route: opencode-route/workhorse, agent build
State: claimed task bound at launch; OpenCode worker prompt admitted.

🤖 Agent note · triaged

Agent: queue worker (repo-worker-code), opencode execution lane
Session / ID: ses_e5e5e16333fffehFWq09WgCMlu · repo-health-code-pilot task 7831
Found via: queue claim (coverage pass 2026-11-03)
Task: Triage of design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024 — confirm still-live at current main, scope the fix or close with evidence.

Fresh revalidation (2026-11-03T15:05Z): #2024 is OPEN with no hard-stop labels (held, user/owned, do-not-merge, do-not-auto-edit all absent), no competing PR or branch, and no queue/<n> work in flight. No placement-framework document exists on main: a recursive git/trees/main search for placement returns nothing, gh search code "task-placement framework" is empty, and no PR touches it. The synthesis has not been produced — the issue is live work, not done.

Input gate — now MET. All four inputs named by the issue have their written deliverables:

investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 placement investigation — matrix posted 2026-10-29: investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 (comment)
investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 OpenCode scheduled tasks — report posted 2026-10-29: investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 (comment) (follow-on implement: fleet OpenCode scheduled-task pattern — systemd timers + opencode run with per-job route #2232)
investigate: Jules scheduled tasks — what fits, and how Muse manages the schedule #2020 Jules scheduled tasks — fit matrix posted 2026-10-29 (comment 11:14Z)
investigate: OpenClaw as a working option — setup state and a Muse↔OpenClaw comms path that is not laptop SSH #2023 OpenClaw — closed 2026-10-30; design doc merged via PR docs(openclaw): comms-path design — Slack primary, direct API phase 2 (#2023) #2303

The 2026-10-29 gate note ("nothing to do on this issue until then") is therefore satisfied. #2019/#2020/#2022 stay open for the orchestrator review round per the program pattern; their deliverables exist.

Fix path (scoped): one synthesis doc under doc/<UTC-timestamp>_design_task-placement-framework-v1/ — a decision table a human or agent dispatcher can apply in under a minute per task, covering every option in #2022's matrix; the shared-pool efficiency model behind it (burn order, expiry, supervision cost); the exceptions process; mermaid/SVG diagrams; follow-up issues for every buildable item; a PoC only where the design calls for one, never the permanent feature. The issue stays open for orchestrator review.

Ordering / routing: Specialist/Summit-grade per the 2026-10-28 program guidance (Workhorse is the default only for the routine lanes). Follow-up design task 8136 enqueued at priority 20 (fix:developing-today/code:issue:2024); it was claimed by lane code-j within seconds of enqueue.

Blockers: none technical. Execution constraint: GitHub comment writes have been hitting secondary-rate-limit 403s since 2026-11-03T13:40Z — this note is posted on a retry.

Merge pull request #4994 from developing-today/queue/8136-2024-task-p…

Agent note — done (queue task 8136, lane code-j)

Deliverable: design doc PR docs: task-placement framework v1 — decision table + shared-pool model + exceptions (#2024) #4994 — MERGED (443e1d9): doc/2026-11-03T15-10-56Z_design_task-placement-framework-v1/ — placement framework v1: ordered decision table (Q1–Q7 / R1–R8, every investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 option), shared-pool efficiency model (pool inventory, burn rules B1–B6, supervision cost), exceptions process (E1–E5).
Inputs verified fresh this shift: investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 matrix (comment 6494235903), investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 report (comment 6494101941), investigate: Jules scheduled tasks — what fits, and how Muse manages the schedule #2020 fit matrix (2026-10-29T11:14Z), investigate: OpenClaw as a working option — setup state and a Muse↔OpenClaw comms path that is not laptop SSH #2023 via merged PR docs(openclaw): comms-path design — Slack primary, direct API phase 2 (#2023) #2303.
Follow-ups filed for every buildable item: new implement: placement framework v1 — machine-readable decision table + placement record on fleet dispatches (from #2024) #4986 (machine-readable table + placement record; holds the design's one PoC, gated on this review) and implement: unified fleet schedule registry — Muse cron + OpenCode timers + Jules schedules in one record (from #2024) #4988 (unified fleet schedule registry); the rest already tracked — implement: fleet OpenCode scheduled-task pattern — systemd timers + opencode run with per-job route #2232, fleet: pilot: first Jules schedule — weekly GitHub Actions version-bump sweep on developing-today/code (from #2020) #2265, design: scheduled OpenCode preliminary babysitter — sweep sessions, fix easy stalls, escalate the rest #2021, OpenClaw: connect chat channels (Slack, Telegram, WhatsApp) #1029/openclaw: enforce token auth for non-loopback gateway callers (comms design G2/G4) #2300/fleet: sandbox tailnet route to the laptop for the direct-API comms path (comms design G3) #2301/openclaw: inventory the existing sessions and what the agent already owns (comms design G5) #2302/Add OpenClaw to #alignment — decide one shared channel vs point-to-point #2170, feat(quota-plane): OpenCode pools consume quota-current — measured placement, rotation, demotion #1157/epic(quota-plane): Stage 2 — 15-minute quota gathering in the PM stack, token adapters, automated pool decisions #1149.
Codex: review could not run on PR docs: task-placement framework v1 — decision table + shared-pool model + exceptions (#2024) #4994 (usage limits, repo-wide) — disposition recorded on PR comment 6607532972; merged on standing criteria (doc-only, MERGEABLE, diff matches intent).
This issue stays OPEN for the orchestrator review round, per the program pattern. Queue labels are being stripped as the task completes; the review round re-opens placement work from the doc, not from queue state.

🤖 Agent note · done — queue worker (repo-worker-code), lane code-j, task 8136

agent-note:v1
state: triaged (coverage pass)
Coverage pass 2026-11-03: reviewed in the fleet-wide per-issue pass (drift / overlap / already-handled / stale / blockers / ordering). Verdict: still live; queued in repo-health-code-pilot as triage at priority 30 (priority/none). If this is already handled at current main, the triage closes it with evidence; if it needs splitting, the triage files the split.
Found via: coverage pass 2026-11-03

🤖 Agent note · triaged (task 8577, lane code-k, direct)

Agent: queue worker (repo-worker-code), lane code-k
Session / ID: task 8577
Found via: queue claim (coverage pass re-queue)

Triage verdict — deliverable landed and merged; the only remainder is the orchestrator review round, which is not worker work.

Verified fresh (2026-11-03T20:20Z):

The synthesis this issue asks for exists: PR docs: task-placement framework v1 — decision table + shared-pool model + exceptions (#2024) #4994 MERGED at 2026-11-03T15:14:37Z (merge commit 443e1d91bfda), landing doc/2026-11-03T15-10-56Z_design_task-placement-framework-v1/ on main — decision table (Q1–Q7 / R1–R8), shared-pool efficiency model (B1–B6), exceptions process (E1–E5). Confirmed present in the main tree this shift.
All four gate inputs have written deliverables (investigate: placement framework — Muse vs GitHub Action vs Action→Jules vs OpenCode session vs OpenCode scheduled task vs Aster/Codex vs OpenClaw #2022 matrix, investigate: OpenCode native scheduled tasks — creation, route choice, independence from Muse tokens #2019 report, investigate: Jules scheduled tasks — what fits, and how Muse manages the schedule #2020 fit matrix, investigate: OpenClaw as a working option — setup state and a Muse↔OpenClaw comms path that is not laptop SSH #2023 via merged PR docs(openclaw): comms-path design — Slack primary, direct API phase 2 (#2023) #2303) — gate status recorded by task 7831's triage and acted on by task 8136, which produced the doc.
Every buildable follow-up is filed and tracked: issue implement: placement framework v1 — machine-readable decision table + placement record on fleet dispatches (from #2024) #4986 (machine-readable decision table + placement record; holds the design's one PoC, gated on the review below) and issue implement: unified fleet schedule registry — Muse cron + OpenCode timers + Jules schedules in one record (from #2024) #4988 (unified fleet schedule registry) are both OPEN, plus the previously tracked implement: fleet OpenCode scheduled-task pattern — systemd timers + opencode run with per-job route #2232, fleet: pilot: first Jules schedule — weekly GitHub Actions version-bump sweep on developing-today/code (from #2020) #2265, design: scheduled OpenCode preliminary babysitter — sweep sessions, fix easy stalls, escalate the rest #2021 and others named in task 8136's closing note. Nothing buildable is untracked, so this triage files no new packets.
Why the issue stays OPEN: per the program pattern in its own body ("a result that says done is not the endpoint — the orchestrator reviews it"), the remaining step is the orchestrator review round of the merged design; issue implement: placement framework v1 — machine-readable decision table + placement record on fleet dispatches (from #2024) #4986's PoC is explicitly gated on that review. That review is the named next action and belongs to the orchestrator/Drew side, not to a queue worker — a worker must not close this issue or build the PoC ahead of it.

Status: triaged. Issue remains OPEN for the orchestrator review round.

Navigation Menu

Repository navigation

design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024

2 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

design: task-placement framework v1 — decision rules and a shared-pool efficiency model #2024

Description

Activity

developing-today commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

2 remaining items

developing-today commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions