Platform Engineer
Platform Engineer (QA + Infrastructure). Owns the QA test tier (integration + an e2e smoke) and the QA deploy of origin/release, plus Heroku deploys, env vars, CI, observability, and the recovery protocol. Senior reviewer for DevOps/Platform PRs — but never reviews a PR he will then QA (no self-gating).
Copy any phrase to launch it as Steffon.
No skills assigned.
Production smoke seal agent
Dor-check post-deploy gate
Release pipeline hardening
Harden bin/e2e-parallel + introduce two-tier (@smoke) e2e model
Agent CLI trinity: bin/task, bin/preflight, bin/secret
Auto-reclaim merged worktrees on close (scale-down normal flow)
Elastic worktree Redis capacity (steady-state 20, scale in/out by 10)
Final shipped fix supersedes the earlier setup-chrome-pins evidence: install-chromedriver:false + a PATH-eviction step, so Selenium Manager (selenium-webdriver 4.32) resolves the chrome-MATCHED chromedriver from the CfT manifest at runtime — self-healing, nothing to pin. Real proof (CI-config change): green CI run 28825092905 — eviction logged 'chromedriver remaining on PATH: none', then test:system 8/0 (prior chrome149/driver150 SessionNotCreatedError gone), 2994 non-system 0F. Advancing to submitted.
Submitted for pr-review. PR #433 (base release) leads with the task URL. Follow-up to #426: bin/pr-review now announces each spawn with an intent-labeled delegate action (summon primary/light review: <soul>) and its prompts carry the primary-owns-gates/drives-verdict vs light-does-neither split; pr-review.md + pr-review-sop.md specify the two Agent-tool delegate labels for the interactive path; primary/light role SOPs sharpened; per-domain Review Checklist added to Carl/Shannon/Jasper/Steffon/Alex role.md. Cert: full suite 2999/0-fail + system 8/0 + rubocop clean; dor-check met (unit+integration). Untouched: bin/release.rb, result-distillation.md, index.md.
Rework done on feat/test-scope-telemetry-registry (PR #414): (1) gem_release_check now registered in release_scopes + wrapped through run_test_scope in publish_gem, abort-before-publish preserved; (2) qa_smoke completed deferred into the else branch, recorded only after the blocking QA post-deploy hook returns green (failed on boot failure) — no premature green. Regression tests added for both; full suite + rubocop certified green, CI green.
Rework landed: rebased onto origin/release (f660e2bc), SKILL.md resolved as deletion + sentence relocated to agents/avi/sops/production-deploy.md, sync phrase re-applied to new atoms row, Run Deployment paragraph re-landed after anchor, qa-release vocabulary + heartbeats cite nit fixed, tripwire extended to successor SOP. Re-certified 4c09c2b084f6; PR #369 MERGEABLE.
Scout report: merge-ready - Review-lane automation remains review-only; fast mode is capped and DoR confirms backend coverage after CLI repair. Reporter: steffon Outcome: merge-ready Findings: - none Questions: - none Checks: - Regression covers Codex global flag order before exec. - PR #293 is non-draft, clean to release, and remote CI green.
Scout report: merge-ready - QA gate improvement is deterministic and keeps e2e residue out of Rails suite certification. Reporter: steffon Outcome: merge-ready Findings: - none Questions: - none Checks: - Task has full-suite and rubocop evidence after rebase. - PR #285 is non-draft, clean to release, and green on GitHub checks.
DEPLOY FIX (conductor): narrowed post_deploy_cmd from 'bin/rails db:seed' -> 'bin/rails runner load(db/seeds/02_agents.rb)'. The full db:seed (a) failed at db/seeds/54_ai_builder_multiple_candidates.rb:232 (non-idempotent: 'Github login has already been taken') and (b) would seed DEMO News/Content/Tasks into PROD at ship. 02_agents is self-documented idempotent + prod-safe and is the only seed this task needs (retire alex-docs / give alex the docs reviewer seat). Verified the agents change already applied on QA.
SHIPPED to prod (turf-monster-mainnet v108, rel-20260624-a59e5f). Per operator decision, ENABLED the prod geo_settings row post-ship: enabled=true, enforcing=true, banned=[AZ,CO,HI,ID,IA,LA,MO,NE,NV,WA]. Legal blocklist + fail-closed now LIVE on mainnet for money/entry actions.
QA-deployed on rel-20260624-a59e5f @ qa.turfmonster.media (turf v24 @ 8ba6370, /up 200, geo/check resolves CO). Fail-closed governed by GeoSetting.enforcing? (row enabled). PROD-SHIP DECISION: confirm the prod geo_settings row state deliberately — if enabled, undetectable-US visitors now fail CLOSED (intended); if disabled, neither blocklist nor fail-closed enforces. Pairs with the fix-ipinfo prod GeoSetting verification note. Deferred non-blocking nits in review comment.
QA-deployed on rel-20260624-a59e5f @ qa.mcritchie.studio (mcr@15e2d3f, /up 200). Fix verified by suite (273 runs 0F) + dogfooded: merging this + geo onto the assembled RC flipped both cleanly to assembled (no manual fixup needed). Note: the fix only governs the PROD board once mcritchie-studio ships to prod. Awaiting operator prod-ship gate.
Conductor note: task was adopted onto an already-ASSEMBLED RC (rel-20260624-a59e5f). release.add via the reopen! path set release_slug but left stage=reviewed (should be assembled); subsequent adopt! calls no-op'd because membership already existed. Manually moved reviewed->assembled to match reality (genuinely a QA-deployed member, fix verified live). Filed conductor bug as adopt-onto-assembled-stage-stuck.
QA-deployed on release rel-20260624-a59e5f @ qa.mcritchie.studio (Heroku v78, /up 200). Building-collapse class live on /deployments; Building-specific placement locked by flipped integration test. Awaiting operator prod-ship gate.
QA-deployed on release rel-20260624-a59e5f @ qa.mcritchie.studio (Heroku v78, /up 200). Glyph verified by component+unit tests (board slug-row renders 📇 for repositories:['rolio']); no live rolio-repo card on board to render it. Awaiting operator prod-ship gate.
We emailed a one-tap sign-in link to . It expires shortly and can only be used once.
No email? Check spam, or close this and try again.
