fix(codex): send Pylon's instructions when a thread starts

[rynfar]

What

Codex threads now receive Pylon's runtime instructions as developerInstructions on thread/start and thread/resume. Until now they were sent only inside each turn's collaboration mode, and Codex 0.153.4 does not pass that copy to the model.

Why it matters

On Codex 0.153.4 a Codex thread in Pylon has been running without any of Pylon's guidance: pull request linking, the collaborative browser, devices, delegation, and the pair protocol. Found while testing pair mode, where a Codex lead behaved as if it had never been told about the executor.

Evidence

Same question to a fresh Codex thread (GPT-6-Astra), isolated server, "which application and harness are you running inside?":

• Before: "I'm running in the Codex application within an OpenAI agent harness."
• After: "I'm running inside Pylon through the Codex harness."

codex features list shows collaboration_modes as removed. The provider log confirms Pylon was sending the instructions in turn/start all along.

Asking a model to quote its instructions is not a usable probe: it answered "none" in both cases. The test has to be something only the instructions could tell it.

How

• buildCodexThreadInstructions(toolsAvailable): Pylon's runtime instructions for the tools this session's credential grants, without Codex's own mode template.
• openCodexThread takes developerInstructions and puts it on both the resume and the start request, so the fallback from a failed resume carries it too.
• The per-turn copy is unchanged, for older Codex versions.

Verification

• vp test run on RuntimeInstructions.test.ts, CodexSessionRuntime.test.ts, CodexAdapter.test.ts: 131 passed. The two new tests were written first and seen failing with the implementation reverted.
• Typecheck t3: 0 errors. Format and lint clean on the changed files.
• Live A/B above.

Known limits

• Whether thread/resume applies new instructions to an existing conversation is not verified; a resumed thread was inconclusive.
• An older Codex that honors the per-turn copy now sees Pylon's instructions twice.
• Codex stays refused as a pair lead. Its collaboration.* tools cannot be turned off, so re-allowing it needs a clean paired run first.

Part of #622.

Model: Claude Fable 5.1, in Pylon via the Claude Code harness.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
pylon-marketing	Ready	Preview	Sep 18, 2026 5:46pm UTC

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	14.0 KiB	14.0 KiB	−3 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.2 KiB	−10 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+7 B (+0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	14.0 KiB	14.0 KiB	−1 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.2 KiB	0 B (0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.7 KiB	−1 B (−0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: 6468f56 · PR result: 5f9d104 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 115.6 KiB
• Claude decoded thread snapshot: 116.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.