feat(cli): npx t3 triage hands a broken install to your own coding agent

[rynfar]

Ports upstream #6563 as a Pylon feature rather than a cherry-pick. t3 triage collects machine facts — version, OS, paths, server liveness — into a scratch directory, then launches claude or codex with a playbook that walks the agent from "what went wrong" through diagnosis to a well-formed issue.

This was CS-9 in the 2026-08-19 upstream review, held back from that batch because it could not be adopted as-is.

Why it needed a port, not a cherry-pick

Upstream's playbook hardcodes pingdotgg/t3code: it searches T3's issues, fetches a newer playbook from T3's raw GitHub URL, and builds a t3code/issues/new link. Shipping that in Pylon would have funnelled Pylon users' bug reports into T3's tracker.

The decision I had to make, and what I assumed

pylon-code/pylon is private, so there is no tracker a user's generated issue can reach. Rather than guess a destination, the repository is configuration:

• PYLON_TRIAGE_REPOSITORY names the repository to clone, search, and file against.
• It is unset by default. With it unset, the context file tells the agent not to post — instead it writes the finished issue to a file next to the context file and hands over the path.

That is the one judgement call in here. If you want triage issues landing somewhere real, set that variable (a public pylon-issues repo would work) and the full flow lights up with no code change.

Upstream's "check for a newer playbook" step is dropped for the same reason — a private repo has no unauthenticated raw URL, so the embedded copy is the only playbook a release can follow. Remaining steps are renumbered, and the cross-reference that pointed at the removed step is rewritten.

Verification

• 6 tests pass, including upstream's byte-identity check between .github/triage/PLAYBOOK.md and the embedded TRIAGE_PLAYBOOK. I regenerated the constant from the file programmatically rather than hand-editing both, so they cannot drift.
• Two new tests cover the configured and unconfigured repository branches.
• Server typecheck clean (verified the t3 package actually ran, not a no-op filter).
• Swept the whole feature for T3 product identity: none remains. Compatibility identifiers (npx t3, T3CODE_HOME, @t3tools/shared) are untouched per AGENTS.md.

Note

.github/ISSUE_TEMPLATE/via-triage.yml ships as-is — it is product-neutral, and while the repo is private only collaborators can use it. It costs nothing and is ready if a public tracker opens.

Model: Claude Opus 5 in Pylon.

---

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

[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	12.6 KiB	12.6 KiB	−25 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.3 KiB	6.3 KiB	−8 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.3 KiB	−17 B (−0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	51.8 KiB	51.8 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	12.6 KiB	12.6 KiB	−4 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.3 KiB	6.3 KiB	−2 B (−0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.3 KiB	6.3 KiB	−2 B (−0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	52.7 KiB	52.7 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: d9a351e · PR result: fedbdc9 · 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: 99.8 KiB
• Claude decoded thread snapshot: 100.6 KiB

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