Context
The Agent Harness session-entry workflow is intentionally short and read-on-demand: .agents/workflows/agent-harness.md sends fresh sessions through ADR 0020, Project board 13, and the target work item. During the 2026-06-15 planning pass, the live issue graph changed under that workflow:
#13377 now exists as the Electron shell umbrella under #13012.
#13033 is now the first build-root/topology-spike leaf under #13377.
.agents/workflows/agent-harness.md Step 3 still says: Electron shell/build-root work routes through #13033.
That was correct before the new umbrella existed. It is now stale enough to misroute future sessions into treating a leaf as the canonical Electron route.
Live latest-open sweep: checked latest 20 open issues immediately before filing at 2026-06-15T18:50Z. Closest matches were #13377, #13033, and #13012; no equivalent workflow-route cleanup leaf found. Exact source/local sweep over .agents, learn, and resources/content found prior #13250 workflow staleness history but no current #13377 route-map ticket. Native A2A in-flight claim sweep could not be run because this Codex surface exposes no list_messages / add_message tools; GitHub state and source are the available authority.
The Problem
The workflow's purpose is to remove cold-entry archaeology. A stale hardcoded Electron pointer defeats that purpose: a future agent can follow the workflow, see #13033 as the Electron route, and miss the accepted umbrella/sub-issue shape where #13377 owns shell packaging/hosting/distribution and #13033 is only the first spike leaf.
This is a route-map bug, not a new Electron implementation scope.
The Architectural Reality
- #13012 remains the Agent Harness parent epic and live decomposition root.
- #13377 is now natively parented under #13012 and owns Electron shell / packaging / main-process Agent OS hosting / distribution, explicitly excluding multi-window choreography.
- #13033 is natively parented under #13377 and remains Ada's first build-root/topology-spike leaf.
.agents/workflows/agent-harness.md is the session-entry surface; it should point to live routing anchors and avoid duplicating full epic bodies.
- ADR 0020 already says #13033 carries the topology spike, so this ticket should only change ADR wording if needed to prevent a conflicting entry contract.
The Fix
One narrow PR:
- Update
.agents/workflows/agent-harness.md Step 3 so Electron shell work routes through #13377, with #13033 named as the first build-root/topology-spike leaf.
- Keep the anti-duplicate instruction: do not file duplicate Electron-support tickets unless live state shows disjoint scope, supersession, or explicit handoff.
- Check
learn/agentos/decisions/0020-agent-harness-concept.md for conflicting session-intake wording. Update only if the ADR currently contradicts the workflow route-map; otherwise leave ADR 0020 stable.
Contract Ledger Matrix
| Target Surface |
Source of Authority |
Proposed Behavior |
Fallback |
Docs |
Evidence |
.agents/workflows/agent-harness.md Step 3 |
#13012 live issue graph + #13377 / #13033 parent-child state |
Route Electron shell work through #13377 and identify #13033 as the first build-root/topology-spike leaf |
If #13377 is closed/superseded later, Step 3 keeps its existing live-state/supersession escape hatch |
Inline workflow text |
Static grep + cold-read check |
| ADR 0020 session-intake / shell wording |
ADR 0020 + this ticket |
Remains non-conflicting with the workflow's route-map; updated only if it still presents #13033 as the full Electron route |
If unchanged, PR body explains why ADR 0020 is concept-stable and workflow owns live route shorthand |
ADR text if changed |
Static grep + diff review |
Decision Record impact
Aligned-with ADR 0020. This does not change the Electron decision, shell boundary, Agent Harness pillars, or guardrails; it keeps the operational entry workflow aligned with the current issue graph.
Acceptance Criteria
Out of Scope
- Implementing the Electron shell.
- Creating new Electron child tickets.
- Rewriting #13012, #13377, or #13033 bodies.
- Changing lead-role rotation, Project board structure, or workflow convention rules.
- Editing
.agents/skills/** or always-loaded AGENTS substrate.
Avoided Traps / Gold Standards Rejected
- Use #13377 as the implementation ticket — rejected because the edit is a tiny workflow-route cleanup and #13377 is an umbrella epic.
- Create duplicate Electron support tickets — rejected; #13033 remains the first implementation leaf under #13377.
- Copy the full Electron plan into the workflow — rejected because workflows are entry sequences, not plan mirrors.
Related
Parent: #13012
Related: #13377, #13033, #13250
Release classification: post-release (Agent Harness planning hygiene; not v13-release-blocking).
Origin Session ID: current @neo-gpt Codex Desktop goal session, 2026-06-15.
Handoff Retrieval Hints: search .agents/workflows/agent-harness.md for Electron shell/build-root; inspect #13377 and #13033 native parent-child state; review #13012 checkpoint 2026-06-15T18:20Z.
Context
The Agent Harness session-entry workflow is intentionally short and read-on-demand:
.agents/workflows/agent-harness.mdsends fresh sessions through ADR 0020, Project board 13, and the target work item. During the 2026-06-15 planning pass, the live issue graph changed under that workflow:#13377now exists as the Electron shell umbrella under #13012.#13033is now the first build-root/topology-spike leaf under#13377..agents/workflows/agent-harness.mdStep 3 still says:Electron shell/build-root work routes through #13033.That was correct before the new umbrella existed. It is now stale enough to misroute future sessions into treating a leaf as the canonical Electron route.
Live latest-open sweep: checked latest 20 open issues immediately before filing at 2026-06-15T18:50Z. Closest matches were #13377, #13033, and #13012; no equivalent workflow-route cleanup leaf found. Exact source/local sweep over
.agents,learn, andresources/contentfound prior #13250 workflow staleness history but no current #13377 route-map ticket. Native A2A in-flight claim sweep could not be run because this Codex surface exposes nolist_messages/add_messagetools; GitHub state and source are the available authority.The Problem
The workflow's purpose is to remove cold-entry archaeology. A stale hardcoded Electron pointer defeats that purpose: a future agent can follow the workflow, see #13033 as the Electron route, and miss the accepted umbrella/sub-issue shape where
#13377owns shell packaging/hosting/distribution and#13033is only the first spike leaf.This is a route-map bug, not a new Electron implementation scope.
The Architectural Reality
.agents/workflows/agent-harness.mdis the session-entry surface; it should point to live routing anchors and avoid duplicating full epic bodies.The Fix
One narrow PR:
.agents/workflows/agent-harness.mdStep 3 so Electron shell work routes through #13377, with #13033 named as the first build-root/topology-spike leaf.learn/agentos/decisions/0020-agent-harness-concept.mdfor conflicting session-intake wording. Update only if the ADR currently contradicts the workflow route-map; otherwise leave ADR 0020 stable.Contract Ledger Matrix
.agents/workflows/agent-harness.mdStep 3Decision Record impact
Aligned-with ADR 0020. This does not change the Electron decision, shell boundary, Agent Harness pillars, or guardrails; it keeps the operational entry workflow aligned with the current issue graph.
Acceptance Criteria
.agents/workflows/agent-harness.mdno longer presents #13033 as the whole Electron shell/build-root route.#13033,#13377, and the stale phraseElectron shell/build-root work routes through #13033.Out of Scope
.agents/skills/**or always-loaded AGENTS substrate.Avoided Traps / Gold Standards Rejected
Related
Parent: #13012 Related: #13377, #13033, #13250
Release classification: post-release (Agent Harness planning hygiene; not v13-release-blocking).
Origin Session ID: current @neo-gpt Codex Desktop goal session, 2026-06-15.
Handoff Retrieval Hints: search
.agents/workflows/agent-harness.mdforElectron shell/build-root; inspect #13377 and #13033 native parent-child state; review #13012 checkpoint2026-06-15T18:20Z.