weekly sync — 2026-06-15

[norbert-pap]

Applied changes

gstack#1916 — Mandatory unresolved-decisions sentinel

Source: garrytan/gstack#1916
Target skills: eng-review/SKILL.md, prod-review/SKILL.md, code-review/SKILL.md

Every review report must now always end with exactly one of:

• NO UNRESOLVED DECISIONS — when all ASK items were settled before moving on
• A **UNRESOLVED DECISIONS:** bullet block — one bullet per open item, stating what the decision was and what breaks if left unresolved before implementation starts

Previously eng-review and prod-review only noted this if the user interrupted (conditional); code-review had no such section at all. The gstack change makes the sentinel mandatory in all six of its review consumers after a token-reduction pass had quietly dropped it. Same principle applied here: a clean review that omits the sentinel is now indistinguishable from one that silently defaulted.

---

Deferred / noteworthy (considered, not applied)

gstack#1966 — Cross-model adversarial review default-on

Source: garrytan/gstack#1966
Makes OpenAI Codex "outside voice" default-on in plan reviews with a codex_reviews master switch. Falls back to a Claude subagent when Codex is unavailable. Not portable: claudebert has no Codex auth integration. The concept (cross-model review as default, not opt-in) is worth noting if claudebert ever adds a second-model review path.

gstack#1929 — browse js/eval --out render-to-file

Source: garrytan/gstack#1929
Adds js "<expr>" --out <file> and eval script.js --out <file> to write evaluation results to disk rather than returning them over the command channel. Includes offline render path docs ("don't bundle your own Chromium"). Deferred because claudebert's browse binary is a separate implementation — documenting flags the binary doesn't support would be misleading. Revisit when the browse binary is updated.

gstack#1912 — Rename "Boil the Lake" → "Boil the Ocean"

Source: garrytan/gstack commit 41c6d3eb
Refactors the Completeness Principle heading and body to use "Boil the Ocean" framing: "boil the ocean one lake at a time" — the complete thing is the goal, broken into achievable lakes. Gstack's version is more concise but drops the "100% test coverage is the vibe code unlock" language that's useful in claudebert. Deferred pending a judgment call on whether to adopt the shorter framing or merge both.

anthropics/skills#1293 — frontend-design skill improvements

Source: anthropics/skills#1293
Improvements to Anthropic's reference frontend-design skill (generates UI). Claudebert's design-review reviews existing UI — different purpose and workflow. No direct port applicable.

VoltAgent/awesome-agent-skills — Aegis (523★)

Source: https://github.com/GanyuanRan/Aegis (523 stars, active)
"Make AI coding agents architecture-aware: baseline-first, evidence-verified, drift-checked, and safe across long tasks." High-signal methodology skill. Would require creating a new claudebert skill file, not cherry-picking to an existing one. Worth evaluating as a future addition to the pipeline.

---

Sources checked (6)

Source	Commits since 2026-06-08	Outcome
garrytan/gstack	12	1 applied, 3 deferred
addyosmani/agent-skills	28	dropped (infra/docs/new agents outside pipeline mapping)
anthropics/skills	3	1 deferred
VoltAgent/awesome-agent-skills	3	1 deferred (Aegis 523★), 2 dropped (below 50★)
rohitg00/awesome-claude-code-toolkit	0	nothing
Owl-Listener/designpowers	2	dropped (design-library index update)

---

Generated by Claude Code