Skip to content

weekly sync — 2026-05-04#1

Open
norbert-pap wants to merge 1 commit into
mainfrom
sync/2026-05-04
Open

weekly sync — 2026-05-04#1
norbert-pap wants to merge 1 commit into
mainfrom
sync/2026-05-04

Conversation

@norbert-pap
Copy link
Copy Markdown
Owner

Sources checked (6)

source commits since 2026-04-27 signal
garrytan/gstack 9 ✅ 1 cherry-pick
addyosmani/agent-skills 7 ❌ infra/docs only
anthropics/skills 1 ❌ API doc fix only
VoltAgent/awesome-agent-skills ~8 ❌ community list additions
rohitg00/awesome-claude-code-toolkit 0
Owl-Listener/designpowers 0

Applied changes (1)

eng-review/SKILL.md ← garrytan/gstack PR #1313

v1.26.2.0 fix: plan-eng-review STOP gates always fire + report-at-bottom contract

Hardened all 5 STOP gates (Step 0 complexity check + Architecture + Code Quality + Test Review + Performance) with the anti-rationalization pattern from upstream:

  • AskUserQuestion is a tool_use, not prose — call the tool directly added to each gate header
  • Named blocked next steps: Do NOT proceed to next section or edit the plan file until user responds
  • Anti-rationalization clause: Loading the AskUserQuestion schema via ToolSearch and then writing the recommendation as chat prose is the failure mode this gate exists to prevent
  • Architecture gate only: An issue with an "obvious fix" is still an issue and still needs explicit user approval

Also updated the completion summary: Step 0 label now reads scope accepted as-is / scope reduced per recommendation, and a Lake Score metric was added (X/Y recommendations chose complete option) to track adherence to the Completeness Principle already in the skill.

Porting notes: ExitPlanMode / Decisions to confirm fallback dropped (claudebert has no plan-mode preamble). {{TOOL_RESOLUTION}} template token dropped. Core safety language kept verbatim.


Deferred — noteworthy but not applied (3)

garrytan/gstack PR #1233 — browser-skills runtime v1 (2026-04-29)

53-file PR adding a new domain-skill pattern (browser-skills/hackernews-frontpage/SKILL.md) and expanding the browse binary with CDP bridge, domain-skill commands, skill tokens, and tab isolation. No changes to browse/SKILL.md.tmpl — the skill documentation was not updated to reflect new commands. Deferred: would need a separate pass to determine which new capabilities (if any) should be surfaced in claudebert's browse/SKILL.md; not a clean text cherry-pick.

addyosmani/agent-skills PR #101 — JSON-escape hook content in session-start hook

Fixes hooks/session-start.sh to use jq --arg so SKILL.md with quotes/newlines produces valid JSON. Hook infrastructure change; no skill content. Dropped (claudebert session-start hook is independent).

anthropics/skills PR #1081 — Remove non-existent purpose field from Files API examples

Removes an incorrect field from API documentation examples. No skill content change. Dropped.


Generated by Claude Code

Source: garrytan/gstack PR #1313 → eng-review/SKILL.md

Hardens all 5 STOP gates (Step 0 complexity check + Architecture +
Code Quality + Test Review + Performance) with the office-hours
anti-rationalization pattern:

- "AskUserQuestion is a tool_use, not prose — call the tool directly"
- Named blocked next steps: do NOT proceed to next section or edit
  the plan file until user responds
- Anti-rationalization clause: loading AskUserQuestion via ToolSearch
  then writing prose is the failure mode this gate exists to prevent
- Architecture gate only: "obvious fix still needs explicit approval"

Also updates completion summary: Step 0 label now shows
"scope accepted as-is / scope reduced" and adds Lake Score metric
(tracks completeness-principle recommendations).

ExitPlanMode / Decisions-to-confirm fallback dropped (not in
claudebert). Template tokens dropped.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants