Context
The current Agent Harness entry workflow exists because closed predecessor #13042 established .agents/workflows/agent-harness.md as a short, read-on-demand mission entry sequence. During this wake cycle, the operator asked to keep coordinating the harness work and specifically called out refining sub-epics/tickets such as Electron support.
Live evidence shows Electron support is not missing: #13033 is the canonical Electron build-root lane under the Agent Harness epic, and it is already assigned. The gap is narrower: the workflow entrypoint does not make that routing obvious, so future agents can re-derive the same conclusion or file duplicate Electron-support tickets.
Release classification: not v13 release-board blocking. Agent Harness board membership: Project 13, because this is a harness entrypoint refinement.
The Problem
The Agent Harness workflow tells sessions to read ADR 0020, inspect board 13, and then open the target work item only. That is directionally right, but the Electron shell decision is now important enough to name in the workflow’s coordination pointers: #13033 is the implementation lane, while .agents/workflows/agent-harness.md is only the entry/routing surface.
Without a compact routing note, future sessions can mistake “refine Electron support” for “file a new Electron ticket” even though the live backlog already has the implementation lane. That is coordination friction, not a product gap.
The Architectural Reality
.agents/workflows/agent-harness.md is a read-on-demand workflow, not always-loaded turn substrate. V-B-A: .codex/hooks/codex-context.mjs loads only .codex/CODEX.md; .claude/CLAUDE.md links to ../AGENTS.md; rg shows the workflow is only referenced on demand and in issue records.
- ADR 0020 accepts Electron as the Harness shell and names the UI/Brain placement split: UI in
apps/, Node-side fleet machinery in ai/, packaging root wrapping the built Body+Brain.
#13033 owns the Electron build root and shell proof. The workflow should route agents to that lane instead of duplicating it.
- The prior workflow ticket
#13042 is closed, so this is a follow-up refinement rather than reopening the convention-establishment leaf.
The Fix
Update .agents/workflows/agent-harness.md with a compact coordination pointer that:
- names
#13033 as the canonical Electron build-root lane;
- says new Electron-support tickets should not be filed unless live inspection shows a disjoint scope or the
#13033 owner/handoff has changed;
- preserves the workflow’s “entry sequence, not skill/atlas” compactness;
- does not add global
AGENTS.md rules or new skill payload.
Contract Ledger Matrix
| Target Surface |
Source of Authority |
Proposed Behavior |
Fallback |
Docs |
Evidence |
.agents/workflows/agent-harness.md routing pointer |
Closed #13042, open #13012, open #13033, ADR 0020 |
Harness entrants see that Electron implementation routes through #13033 before filing or claiming adjacent work |
If #13033 is superseded/reassigned, update or remove the workflow pointer in a follow-up |
Same workflow file only |
rg + live issue checks; diff review; git diff --check |
Decision Record Impact
Aligned with ADR 0020 and ADR 0007. This ticket does not amend either ADR; it keeps the tactical routing note in the on-demand workflow surface rather than global turn-loaded substrate.
Acceptance Criteria
Out of Scope
- Implementing Electron shell/runtime behavior.
- Editing the
#13033 issue body or replacing its owner.
- Adding new global rules, new skills, or Project 13 board policy.
- Closing or re-scoping the parent Agent Harness epic.
Avoided Traps
- Duplicate Electron-support ticket: rejected because #13033 already owns the implementation lane.
- Global substrate rule: rejected because the need is mission-specific and already has an on-demand workflow surface.
- Workflow bloat: rejected; the update should be a small routing pointer, not an embedded implementation plan.
Duplicate Sweep
Semantic/local sweep: ask_knowledge_base(type=ticket) found #13012, #13033, and closed predecessor #13042; no open equivalent workflow-routing refinement found. Exact local sweep found .agents/workflows/agent-harness.md and #13042 history.
Live latest-open sweep: checked latest 20 open issues at 2026-06-15T04:53:46Z; no equivalent found. #13033 is present but is the Electron implementation lane, not this workflow-routing follow-up.
A2A in-flight claim sweep: checked latest 30 Memory Core messages at 2026-06-15T04:53:46Z; recent claims cover #13286, #13289, and #13167 plus review traffic, with no overlapping workflow/Electron-routing claim.
Related
Parent: #13012
Electron implementation lane: #13033
Closed predecessor: #13042
Origin Session ID: unavailable in the current Codex Desktop surface after compaction; use the retrieval hints below.
Handoff Retrieval Hints: "Agent Harness Electron lane routing #13033 workflow", ".agents/workflows/agent-harness.md #13042 #13033".
Context
The current Agent Harness entry workflow exists because closed predecessor
#13042established.agents/workflows/agent-harness.mdas a short, read-on-demand mission entry sequence. During this wake cycle, the operator asked to keep coordinating the harness work and specifically called out refining sub-epics/tickets such as Electron support.Live evidence shows Electron support is not missing:
#13033is the canonical Electron build-root lane under the Agent Harness epic, and it is already assigned. The gap is narrower: the workflow entrypoint does not make that routing obvious, so future agents can re-derive the same conclusion or file duplicate Electron-support tickets.Release classification: not v13 release-board blocking. Agent Harness board membership: Project 13, because this is a harness entrypoint refinement.
The Problem
The Agent Harness workflow tells sessions to read ADR 0020, inspect board 13, and then open the target work item only. That is directionally right, but the Electron shell decision is now important enough to name in the workflow’s coordination pointers:
#13033is the implementation lane, while.agents/workflows/agent-harness.mdis only the entry/routing surface.Without a compact routing note, future sessions can mistake “refine Electron support” for “file a new Electron ticket” even though the live backlog already has the implementation lane. That is coordination friction, not a product gap.
The Architectural Reality
.agents/workflows/agent-harness.mdis a read-on-demand workflow, not always-loaded turn substrate. V-B-A:.codex/hooks/codex-context.mjsloads only.codex/CODEX.md;.claude/CLAUDE.mdlinks to../AGENTS.md;rgshows the workflow is only referenced on demand and in issue records.apps/, Node-side fleet machinery inai/, packaging root wrapping the built Body+Brain.#13033owns the Electron build root and shell proof. The workflow should route agents to that lane instead of duplicating it.#13042is closed, so this is a follow-up refinement rather than reopening the convention-establishment leaf.The Fix
Update
.agents/workflows/agent-harness.mdwith a compact coordination pointer that:#13033as the canonical Electron build-root lane;#13033owner/handoff has changed;AGENTS.mdrules or new skill payload.Contract Ledger Matrix
.agents/workflows/agent-harness.mdrouting pointer#13042, open #13012, open #13033, ADR 0020rg+ live issue checks; diff review;git diff --checkDecision Record Impact
Aligned with ADR 0020 and ADR 0007. This ticket does not amend either ADR; it keeps the tactical routing note in the on-demand workflow surface rather than global turn-loaded substrate.
Acceptance Criteria
.agents/workflows/agent-harness.mdnames #13033 as the canonical Electron build-root lane for the Agent Harness.#13042.AGENTS.md,.agents/skills/**, or ADR mutation is introduced for this narrow routing note.git diff --checkand a targeted read of the rendered workflow text.Out of Scope
#13033issue body or replacing its owner.Avoided Traps
Duplicate Sweep
Semantic/local sweep:
ask_knowledge_base(type=ticket)found #13012, #13033, and closed predecessor#13042; no open equivalent workflow-routing refinement found. Exact local sweep found.agents/workflows/agent-harness.mdand#13042history.Live latest-open sweep: checked latest 20 open issues at 2026-06-15T04:53:46Z; no equivalent found.
#13033is present but is the Electron implementation lane, not this workflow-routing follow-up.A2A in-flight claim sweep: checked latest 30 Memory Core messages at 2026-06-15T04:53:46Z; recent claims cover #13286, #13289, and #13167 plus review traffic, with no overlapping workflow/Electron-routing claim.
Related
Parent: #13012 Electron implementation lane: #13033 Closed predecessor: #13042
Origin Session ID: unavailable in the current Codex Desktop surface after compaction; use the retrieval hints below.
Handoff Retrieval Hints:
"Agent Harness Electron lane routing #13033 workflow",".agents/workflows/agent-harness.md #13042 #13033".