skills(sage-saint): correct gotchas after repair pass - #134

PR #134 had already merged (10:12 UTC, squash bc97b7f) before this review ran, so it became a post-merge direct review (one encrypted docs file, no code — no agent fan-out). I decrypted both revisions of modules/agents/skills/sage-saint/SKILL.md with the fleet key and diffed the plaintext: the change retracts a false CLI gotcha, records now-fixed release-pipeline failures, adds a colima /tmp trap, and adds ~64 lines of new runbook sections. Verified: sops envelope integrity (recipient set unchanged and matching .sops.yaml, MAC/lastmodified updated, committed blob is ciphertext); the 'now-fixed' pipeline claim cross-checked against the release repo's workflow history (fix commit 0c2123e9, 2026-08-24, matches). The corrections are accurate and well-provenanced. Two findings, both fixed: P3 comment rot — an untouched upgrade-runbook step still called a verify-attestations failure 'cosmetic', advice that post-fix would mask real regressions; P4 — a stack-map row named only one of the two sandbox-container naming vintages the PR itself documents. Inline comments were impossible/inappropriate (the GitHub diff is ciphertext, and plaintext must stay out of public comments), so findings went in the summary comment, generically worded. Fixes landed on a fresh branch off main as commit 5cdb2d7 in follow-up PR #135 (#135); full pre-commit suite passed on that commit (gitleaks, trufflehog, glassine check, check-crypt-patterns, and all local behaviour suites). Summary comment with 'Verdict: approve' posted: #134 (comment)