changelog-watch
Skill44 starsChanged 3 months agoMITView on GitHub
Use when checking whether new Claude Code or Codex releases affect Mainframe's adapters, or when auditing adapter drift after a CLI upgrade. Fetches the upstream delta since the last reviewed version, classifies every entry against the consumed-surface checklists, and produces a triage report separating compatibility risks from adoption opportunities.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- prestarius/jet-skills-stacktheir own version, first seen Oct 11, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- prestarius/jet-skills-stack★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- assistant-uiSkill
- claude-protocol-debuggerSkill
- codex-protocol-debuggerSkill
- mainframe-design-systemSkill
- radix-ui-design-systemSkill
- release-downloadsSkill
- shadcnSkill
- tauri-v2Skill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 94
Before acting on a high risk, confirm it against the live CLI with .claude/skills/claude-protocol-debugger/ or …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Changelog Watch
One question per run: what changed upstream since Mainframe last looked, and what does it mean for the adapters? All paths below are relative to the repo root. Analysis only — never fix drift in the same run; findings become todos.
Steps
- Read
.claude/skills/changelog-watch/state.json. Pick the tool(s) to run —claude,codex, or both.lastReviewedRefis the anchor: entries up to and including it are already triaged. It is whatever the upstream list calls that entry — a bare version in changelog mode, a release tag in releases mode — and is never parsed, only matched. Fetch the delta. Per tool:
node .claude/skills/changelog-watch/scripts/fetch-delta.mjs --tool claude node .claude/skills/changelog-watch/scripts/fetch-delta.mjs --tool codexDone when the CLI names the delta file it wrote under
.claude/skills/changelog-watch/reports/, or it printedno changes: <tool> is current at <ref>— in which case stop here and report exactly that.--jsonswaps the human summary for the same result as fields (head,count,truncated,nextAnchor,out); it does not suppress the delta file, which is written whenever the delta is non-empty. The fetcher refuses to overwrite an existing delta file — pass a different--out, or--forceto replace it.- Load the tool's checklist —
docs/research/adapters/claude/CONSUMED-SURFACE.mdordocs/research/adapters/codex/CONSUMED-SURFACE.md(schema indocs/research/adapters/README.md). Done when every checklist ID is in context. - Classify every delta entry per
.claude/skills/changelog-watch/classification.md: relevance filter first (drop by default), then risk vs opportunity, severity, and the routing table. Done when every entry is either mapped to a checklist ID or explicitly dropped, and the dropped count is recorded. An unclassified entry means this step is not finished. - Write the report to
.claude/skills/changelog-watch/reports/<ISO-date>-<tool>-since-<anchor>-report.mdusing the template inclassification.md— the anchor in the name keeps a second pass on the same day from overwriting the first. Print the summary: risks (each naming its checklist ID, Rust consumer, and recommended regression test), opportunities, relevant-no-action entries, dropped count. Advance state — only after the report exists. Rerun the same fetch, adding
--commit-state <newest entry reviewed>and the--outpath the pass wrote, then commitstate.jsontogether with the pass's report. The report is the only record of its findings:node .claude/skills/changelog-watch/scripts/fetch-delta.mjs --tool claude \ --since 2.1.176 --max 5 --out <pass delta> --commit-state 2.1.181The ref must be one the refetched delta contains, and that delta file must still be on disk, or the CLI refuses: state may only walk over entries a run actually fetched and reported. Committing the current anchor is the allowed no-op — it records "checked, nothing new".
- Offer the todo drafts (template in
classification.md) for filing perdocs/guides/issue-tracker.md. Filing is a separate, explicit action — never automatic.
Range too wide
Claude ships roughly a release a day; a stale anchor can mean dozens of versions. Walk forward in passes instead of classifying everything at once:
node .claude/skills/changelog-watch/scripts/fetch-delta.mjs --tool claude --max 5--max returns the oldest unreviewed versions first and prints
nextAnchor plus the exact command to continue, --out included. Classify
the pass, report, --commit-state the pass's newest version, repeat. Each
pass leaves state consistent, so an interrupted walk never skips entries.
Give every pass its own --out: the default path carries the pass's anchor
for that reason, and deltas are gitignored, so a delta overwritten by the
next pass is gone.
Verifying a suspected change
The report classifies changelog prose; it does not prove runtime behaviour.
Before acting on a high risk, confirm it against the live CLI with
.claude/skills/claude-protocol-debugger/ or
.claude/skills/codex-protocol-debugger/.
Adding a tool
A new tools entry in state.json (repo, mode: changelog|releases,
changelogPath or includePrerelease, seed lastReviewedRef) plus a
docs/research/adapters/<tool>/CONSUMED-SURFACE.md checklist. No code change,
provided the repo publishes either a changelog file or GitHub releases.
Failure modes
unknown anchor: <ref> — absent from the complete upstream list— the ref is genuinely gone (upstream rewrote history) or is a typo in--since. Investigate; never work around it by picking "everything".unknown anchor: <ref> — absent from the N newest releases fetched— the walk stopped before reaching it. The releases fetch pages until it sees the anchor, so this means the anchor is older than the page cap; raise the cap infetch-delta.mjsrather than reseeding state, which would skip everything in between.ghmissing — the fetcher falls back to anonymous HTTPS per page; if both fail it exits non-zero. An empty delta is only trusted when the CLI saysno changes, never when a fetch failed. Transientgh5xx and stream errors surface verbatim; rerun.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

