Subagent that scans CopilotKit + AG-UI releases shipped in (and just after) the report window, diffs the git tags to extract what each release actually fixed, and returns a fix-map the weekly-report writer cross-checks against open issues. Catches issues that are quietly already-fixed in a shipped release, and is the authoritative source for the "current release vX.Y.Z" version claims. Invoked by weekly-report (spawned in parallel with the Discord/GitHub pulls); runs as a general-purpose subagent so the orchestrator's context stays small.
中文简介正在准备中。当前显示原文说明。
查看安装方法安装前可以查看交给 AI 代理的指令内容。
Pull every CopilotKit + AG-UI release in the cycle, work out what each one actually fixed, and hand back a compact fix-map plus the authoritative latest version per repo. The weekly-report writer uses this two ways: (1) cite versions that are 100% correct, and (2) catch issues we'd otherwise mark OPEN that have quietly already shipped a fix.
Runs as a subagent spawned by weekly-report in parallel with the Discord + GitHub pulls (Subagents A/B). No other subagent depends on it; the orchestrator only needs the fix-map by the clustering/resolution step.
A wrong "current release vX.Y.Z" on the lead Top-issue card is a credibility hit, and an issue we headline as "open, no fix" that actually shipped a fix two days ago makes the whole report look stale. Both are avoidable by reading the releases.
Release notes are frequently empty stubs. Real example (week of 2026-06-19→26): v1.61.0 had full notes, but v1.61.1 and v1.61.2 both had bodies that read only Release v1.61.x. A notes-only scan would have reported "nothing fixed" for two of three releases.
So the source of truth is the git tag diff + commit/PR parse, not gh release view bodies. Read notes when they exist (they're nicer prose), but always back them with the diff.
| Repo | Release tag shape | Authoritative version source |
|---|---|---|
CopilotKit/CopilotKit | semver vMAJOR.MINOR.PATCH (e.g. v1.61.2) | npm view @copilotkit/react-core version + gh release list |
ag-ui-protocol/ag-ui | date tags release/YYYY-MM-DD (e.g. release/2026-06-24) | npm view @ag-ui/core version, npm view @ag-ui/langgraph version + gh release list |
Note AG-UI ships several sub-packages on independent versions (@ag-ui/core, @ag-ui/client, @ag-ui/langgraph, @ag-ui/a2ui-middleware…). A langgraph-only bump (e.g. @ag-ui/langgraph 0.0.41→0.0.42) is exactly the kind of thing that fixes a reported langgraph issue — report per-package, not one blob.
List releases + find boundaries.
gh release list --repo CopilotKit/CopilotKit --limit 12
gh release list --repo ag-ui-protocol/ag-ui --limit 12
Identify <prev> (last release before window start) and <latest> (newest now) per repo.
Confirm the live latest version (this is the value the writer cites):
npm view @copilotkit/react-core version
npm view @ag-ui/core version ; npm view @ag-ui/langgraph version
Use the highest released semver. Never a remembered one.
Diff the tags and extract fixed issues — the core step:
gh api repos/CopilotKit/CopilotKit/compare/<prev>...<latest> --jq '.commits[].commit.message'
gh api repos/ag-ui-protocol/ag-ui/compare/<prev>...<latest> --jq '.commits[].commit.message'
From the commit messages, capture every Fixes #N / Closes #N / Resolves #N (case-insensitive) and every (#PR) reference. For squash-merged PRs the message holds both the PR number and the Fixes #N. When a commit cites a PR but no issue, fetch the PR to recover the closed issue:
gh pr view <pr> --repo <repo> --json title,closingIssuesReferences,mergedAt,url
Attribute each fix to the release whose tag contains its merge (use the per-release compare <v_{n-1}>...<v_n> when you need to know which version shipped it).
Map notes when present (CopilotKit point releases often have none — that's expected):
gh release view <tag> --repo <repo> --json publishedAt,body
<tag> · <date> · 1-line headline (or "stub notes — diffed").#NNNN · fixed in <version> · via PR #MMMM · <repo> · 1-line what.
Group CopilotKit vs AG-UI.#NNNN we'd mark OPEN appears fixed in <version> — verify.weekly-report)NOTE: appears fixed in vX.Y.Z (PR #MMMM) — verify rather than silently moving it. The human verifies before it's reclassified (a Fixes #N in a commit isn't always a complete fix). Don't auto-move to Resolved.shipped in vX.Y.Z to each ✅ Resolved this week row, sourced here.weekly-report — orchestrator; spawns this in parallel with the Discord/GitHub pulls and carries the cross-check + version-citation rules.deep-read-issue — per-issue fix-PR detection; release-scan is the complementary shipped-release view (did the merged PR actually make it into a tagged release?).还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
After a weekly report is built, diff it against the PRIOR report to find items that appear in BOTH weeks, then print the Owner (and prospect "Passed to") value recorded last week so it can be carried forward. Print-only — never edits the owner fields; Nathan verifies and re-tags in Notion. Invoked by weekly-report near the end, and standalone on "carry forward owners", "who owned this last week", "owner continuity".
中文简介正在准备中。当前显示原文说明。
Subagent flow for deep-reading GitHub issues + their linked/closing PRs to the bottom. Captures CURRENT state, maintainer status updates, reporter repro, blocking reviewer concerns, hidden second bugs, cross-repo duplicates. Invoked by weekly-report orchestrator; runs as a general-purpose subagent so the orchestrator's context stays small.
中文简介正在准备中。当前显示原文说明。
Deep enrichment for community-sourced enterprise PROSPECTS (the 🎯 Prospective enterprise customers subsection). For each prospect, resolve their LinkedIn from their GitHub self-linked accounts FIRST (search only if none is self-linked), verify the employer matches, capture their full name, then list the company website + company size (ARR / latest funding round / employee count). Runs as a general-purpose subagent (many web searches) so the orchestrator's context stays small. Invoked by weekly-report (enterprise section) and the enterprise skill.
中文简介正在准备中。当前显示原文说明。
Subagent for GitHub author enterprise enrichment. Runs gh api users/<login> for each unique handle and classifies enterprise vs indie. Invoked by weekly-report or enterprise skills; runs as Explore subagent so the orchestrator's context stays small.
中文简介正在准备中。当前显示原文说明。
Enterprise-specific view across CopilotKit + AG-UI. "Enterprise" here = CopilotKit's COMMERCIAL surfaces (Premium / CopilotKit Enterprise / Intelligence Platform / paid tiers), NOT CopilotKit running inside a big company. Tracks the commercial product surfaces (from product-surface-scan — Enterprise Intelligence, Cloud, self-host license, threads/persistence tier, Inspector, premium UI/Angular SDK, analytics, security bundle, Slack/Teams), applies the free-vs-paid classifier to decide if an issue belongs here, surfaces cross-page product contradictions, and tracks enterprise questions/complaints, community-sourced prospects (with a sales owner), and enterprise reporters with prior-week trend. Elevated to the top of the weekly report. Runs standalone ("who's at enterprise this week") or invoked by weekly-report. Triggers on "enterprise report", "enterprise signals", "enterprise status", "who at enterprise this week".
中文简介正在准备中。当前显示原文说明。
P0 classification rules for community reports. Six categories that auto-headline regardless of reporter count. Reference doc — invoked by weekly-report when classifying.
中文简介正在准备中。当前显示原文说明。