Status:
draft - recovered from Donna’s donna-rescue-uncommitted-2026-05-08T215800Z stash for Candi governance review. Treat as docs-mirror placeholder pending typed-artifact promotion under SPEC-078.method.windows-codex-incident-activation
Purpose
Train new Windows/Codex agents to activate Prism governance before local host mechanics consume the whole task. This fragment exists because a new Windows/Codex agent can correctly fix a launcher or app-server issue while still missing the Prism lifecycle obligation: incident-class work is not complete until the incident or RCA is recorded and verified through Prism.Trigger
Use this fragment before and during any Windows/Codex work involving:coder.ps1, PowerShell launchers,.cmdwrappers, or Windows PATH resolution- Codex app-server startup, loopback ports, stale process trees, or app-server injection
prism install,npm,npx, MCP config generation, or editor-host setup- repeated
PeerJoined/PeerLeft, session flap, heartbeat, signal delivery, or MCP bootstrap symptoms - any operator correction that says the issue is broader than a local file or shell problem
Core Rule
If the symptom touches Prism lifecycle, signaling, MCP, install, launcher, or session state, treat it as a Prism incident first and a local Windows bug second. Local debugging may continue, but the incident/RCA artifact must be created as soon as the failure class can be named.Required Sequence
- Name the failure class in one sentence.
- Create a Prism incident/RCA artifact:
- prefer
prism_postmortemfor a detected error or incident - use
prism_noteonly for corrective context attached to an existing incident or whenprism_postmortemis unavailable
- prefer
- Verify exact persistence:
- returned artifact id,
prism_diff, or direct artifact listing
- returned artifact id,
- Verify discoverability:
semantic_recallwith distinctive content terms, not just UUID text
- Continue the local fix.
- Do not call the task complete until the checkpoint, wrap, PR, or signal cites the artifact id.
Windows/Codex Bias Check
Windows/Codex failures often look like local host problems:- PowerShell command precedence
Start-Processbehavior.ps1vs.cmdshim resolutionspawnSync("npm")/spawnSync("npx")behavior- stale Codex app-server process trees
- loopback port allocation
- stale MCP config on a fresh install
Done Gate
A Windows/Codex incident-class task is not done until all five are true:- fix or containment exists
- Prism incident/RCA artifact id exists
- exact persistence was verified
- semantic recall was checked with distinctive terms
- handoff/checkpoint/wrap/signal cites the artifact id
Worked Example
Repeated Amanda/CodexPeerJoined events on mini1 were caused by orphaned old Codex app-server trees running on stale loopback ports while the active session used a newer port. The engineering fix reaped stale same-identity/same-project app-server trees before starting a new Windows Codex app-server.
The method gap was that the engineering fix landed before the incident/RCA artifact. Under SPEC-063 v0.2, the correct behavior is to file the incident as soon as the repeated lifecycle symptom is classified, then update or resolve the artifact as the local fix lands.
Self-Test Prompt
After reading this fragment, the agent must be able to answer:- What symptom tells me this is a Prism incident rather than only a Windows launcher bug?
- Which Prism artifact did I create?
- How did I verify exact persistence?
- Which recall query proves the RCA is discoverable?
- Where did I cite the artifact before declaring the task complete?

