cut-release
Skill18 starsChanged 4 months agoApache-2.0View on GitHub
Cut a release end-to-end — promote CHANGELOG [Unreleased] into a dated version block (auto-drafting any gap entries via /changelog), open the prepare-release PR, watch for its merge, then push the vX.Y.Z tag that triggers the release pipeline. The human's one checkpoint is reviewing/merging the changelog PR; the tag fires automatically once it merges. Triggers on "cut a release" / "cut vX.Y.Z".
- Reads credentials
- Commits and pushes
36 other projects use a version of this.
Adopted by
Most recently taken up by
- ItalyPaleAle/skillstheir own version, first seen Nov 16, 2026
- ToolJet/ToolJettheir own version, first seen Nov 12, 2026
- fim-ai/fim-onetheir own version, first seen Nov 11, 2026
- GLinnik21/plx-nativetheir own version, first seen Nov 8, 2026
- mediajunkie/piper-morgan-producttheir own version, first seen Nov 8, 2026
- delexw/motion-graphics-agenttheir own version, first seen Nov 8, 2026
Used with
Projects with it also use quota-axi, iterate-pr and 6 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
layer5ioSkillquota-axi
Report local Claude, Codex, Cursor, GitHub Copilot, and Grok quota windows via the quota-axi CLI - remaining percentages, reset times, and provider status read from local auth sources, with no routing, recommendation, or provider mutation. Use before deciding whether it is safe to keep spending a provider's quota, when the user asks about usage, rate limits, or remaining quota, or when comparing local provider headroom.
Also in 4 projects that use this
SentrySkilliterate-pr
Iterate on a PR until actionable CI passes and high/medium review feedback is addressed. Use for PR CI failures, review feedback, or green-check loops; do not wait for human approval, draft status, or merge gates.
Also in 4 projects that use this
layer5ioSkillgh-axi
Operate GitHub through the gh-axi CLI - issues, pull requests, workflow runs, workflows, releases, repositories, labels, Projects (v2), Actions secrets and variables, search, and raw API access. Use whenever a task touches GitHub: listing or filing issues, reviewing or merging PRs, checking CI runs, triggering workflows, cutting releases, managing Projects boards, or managing Actions secrets/variables.
Also in 4 projects that use this
kunchenguidSkillchrome-devtools-axi
Control a Chrome browser session through the chrome-devtools-axi CLI - navigate, snapshot, click, fill forms, run JavaScript, inspect console and network, take screenshots, audit performance. Use whenever a task needs a real browser: opening or testing a web page, clicking through a flow, extracting page content, or debugging a website.
Also in 4 projects that use this
kunchenguidSkilllavish
Turn complex or visual agent responses into rich, reviewable HTML artifacts (HTML files) the user can annotate and send feedback on, using the lavish-axi CLI. Use when about to give a plan, comparison, diagram, table, code diff, report, or anything easier to grasp visually than as prose.
Also in 4 projects that use this
ThisIsSadeghiSkillverify-ui
Verify UI implementation against Stitch design via a token audit (HTML ↔ Code, with X-component default-render checks) and an X-components compliance check.
Also in 3 projects that use this
Other versions36 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 36
- ToolJet/ToolJet★ 41ksame name
- meshery/meshery★ 12ksame name
- danielmeppiel/apm★ 4.4ksame name
- microsoft/apm★ 4.4ksame name
- Screenly/Anthias★ 3.7ksame name
- Netflix/bpftop★ 2.7ksame name
- tansu-io/tansu★ 1.9ksame name
- fim-ai/fim-agent★ 1.6ksame name
- fim-ai/fim-one★ 1.6ksame name
- libredb/libredb-studio★ 1.3ksame name
26 more not listed
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 repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- add-logsSkill
- changelogSkill
- check-logsSkill
- refresh-third-party-inventorySkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. 0 are common, 3 are distinctive, and 13 are very specific to this project.
0
Common
Most of the 36 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
13
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Run the whole release from one command.
- Very specific: line 38
Preconditions: run gh auth status up front.
- Very specific: line 112
Diff, don't guess: Before drafting, read the candidate PR's diff and confirm an actual name/param/output/behavior change with no …
- Very specific: line 121
Nothing to release? Stop before promoting: After gap detection and any drafting, if [Unreleased] still has no entry line (^- ), there is …
- Very specific: line 127
A3. Version: If the version was passed, use it.
- Distinctive: line 150
If any check fails, the promotion is malformed — fix before proceeding.
- Distinctive: line 160
Do not introduce facts absent from the block.
- Very specific: line 167
Commit CHANGELOG.md and the .github/release-notes/vX.Y.Z.md summary from A6 (nothing else): Prepare release vX.Y.Z, ending the message with …
- Very specific: line 218
Tag-exists guard — distinguish remote from local-only: If vX.Y.Z is on the remote (git ls-remote --tags origin "vX.Y.Z" non-empty) → this …
- Very specific: line 273
Run it backgrounded; expected ~15–25 min end-to-end, and it may sit in queued before starting.
- Very specific: line 283
Optional cleanup (offer, don't force): After a successful release the local checkout is often still on the merged release/vX.Y.Z branch and …
- Distinctive: line 288
Stop. Do not retry the run or edit the Release.
- Very specific: line 293
Never invent a SOL ticket number — halt on any SOL-????
- Very specific: line 296
Only edit the [Unreleased] heading, the new dated block, and the ## Links lines in CHANGELOG.md.
- Very specific: line 299
Never reuse or force-update a tag.
- Very specific: line 303
Do not modify the /changelog skill — invoke it unchanged for gap drafting.
“Common” is measured: kept by at least 6 in 10 of the 36 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Cut a release
Run the whole release from one command. Invoked with /cut-release <version> (or "cut a new
v0.6.0 release").
The release pipeline (.github/workflows/release.yml) triggers on a pushed v* tag and does the
heavy lifting itself — builds binaries for 4 OS/arch, runs the LLM eval suite against the tagged
commit, and does not push the multi-arch image or create the GitHub Release until that suite
passes. It extracts the tagged commit's ## [X.Y.Z] CHANGELOG block and attaches artifacts +
checksums. No GitHub-UI "draft release / auto-generate notes" step is needed. The suite starts
beside the build, not after it. A full run costs API credits (about $4.78 when measured on
2026-08-05) and does not retry. A red suite is gh run rerun <run-id> --failed. A full rerun starts the suite again. A missing or rotated key is fixed in repo settings and then that same rerun, not a new tag. There is no admin
skip. RELEASING.md has the job graph. A finished binary build is not "clear to publish."
So cutting a release reduces to: get the dated CHANGELOG block onto main, then push the tag.
The one human checkpoint is merging the changelog PR — that is where "make sure the changelog
is updated properly" happens, and it is required because main is protected and the pipeline reads
the CHANGELOG from the tagged commit. Everything before and after is automatic: the merge is
treated as the go-ahead, and the tag fires on merge (guarded — see Phase D).
Usage
/cut-release <version>— e.g./cut-release 0.6.0orv0.6.0.
Validate first: reject anything not matching ^v?[0-9]+\.[0-9]+\.[0-9]+$ (stable SemVer) before
any command uses the value. Then normalize once and derive both forms explicitly: strip a leading
v for the bare form (0.6.0) used in the ## [0.6.0] heading and wherever the Phase D awk
takes v="0.6.0"; keep the v-prefixed form (v0.6.0) for the git tag and compare-URL
segments. Never pass the raw argument through un-normalized (a v0.6.0 input must not yield
## [v0.6.0] or a vv0.6.0 tag).
Preconditions: run gh auth status up front. Phases B–E (open/watch the PR, tag, report)
require an authenticated gh — if it's unavailable, stop and have the operator run
gh auth login before those phases. Phase A2 alone can degrade to a git log fallback when gh is
missing (less precise — see A2), so preparation may proceed unauthenticated, but the release cannot
be cut without gh.
Phase A — Prepare the changelog
A1. Sync and preconditions. First git fetch origin main --tags so every step runs against
fresh remote state. Warn if the working tree is dirty. The promotion must be based on
origin/main, not the local checkout — git fetch updates the origin/main ref but not your
local main/working tree, so a local main that's behind origin would make A4 promote a stale
CHANGELOG.md and silently drop [Unreleased] entries merged since. The release branch is
therefore created from the fetched remote tip (see the clean-start path below), and A2–A4 run on
that branch.
vPREV (for the compare range) is the highest stable tag strictly below the target version —
not the global newest tag. Filter out pre-release tags (those containing -) and any tag ≥ the
target, then take the max. Plain git tag --sort=-v:refname | head -1 is wrong for a backport/patch
on an older line (it would pick a newer minor) and floats a -beta tag above stable. If no stable
tag below the target exists, this is the first release — stop and confirm the bootstrap range with
the operator.
Idempotency / resume — decide from REMOTE state, not the working tree. A resumed run (fresh
clone, or main checked out after a dropped session) must not re-promote or open a duplicate PR.
Check, in order:
vX.Y.Zalready on the remote (git ls-remote --tags origin "vX.Y.Z") → release already cut; go to Phase E (reporting only).## [X.Y.Z]block already onorigin/main(via the Phase D extraction) but no remote tag → the PR merged; jump straight to Phase D and cut the tag.- An open PR from
release/vX.Y.Z(gh pr list --head "release/vX.Y.Z" --state open) → resume the Phase C watch on it. - A
release/vX.Y.Zbranch exists (local or remote) but no PR → push it and open the PR (Phase B). - None of the above → create the release branch from the fresh remote tip
(
git switch -c release/vX.Y.Z origin/main) so promotion runs on the latestmain, then start at A2.
A2. Gap detection + auto-draft. Enumerate PRs merged since vPREV:
gh pr list --base main --state merged --limit 500 \
--json number,title,mergedAt,headRefName,filesThe explicit --limit 500 is required — gh's default is 30, and a release window wider than
that would silently drop older merged PRs and miss their gaps (the exact failure this phase exists to
catch). Keep those merged after vPREV's commit time, compared in UTC — gh's mergedAt is
UTC (…Z) but git --format=%cI carries a local offset (-04:00), so a raw lexical compare across
the two is wrong. Get the cutoff as a UTC instant with
TZ=UTC0 git log -1 --date=format-local:%Y-%m-%dT%H:%M:%SZ --format=%cd vPREV (portable — avoids
BSD-vs-GNU date(1) parsing differences), then string-compare the two …Z timestamps. If gh is
unavailable, fall back to git log vPREV..origin/main --format=%s%n%b — this degrades gracefully but
is less precise, so prefer the authenticated gh path (see Preconditions). Keep only PRs hitting production surface, filtered
with the single source of truth shared by the reminder hook and CI gate — source it, never re-embed
the pattern:
source .github/scripts/production-surface.sh
grep -E "$SURFACE_RE" | grep -v "$SURFACE_TEST_EXCLUDE"Confirm a real gap before drafting — do not draft from SOL presence alone. A surface-regex
match is a candidate, not a gap; applied literally the naive "one /changelog per unmatched SOL"
rule over-fires (on v0.6.0 it flagged 11 candidates for 1 true gap, and auto-drafting all 11 would
have produced duplicate/garbage entries and 10 wasted /changelog runs). A true gap is a
production-surface PR whose user-facing contract change has no representation in [Unreleased].
For each candidate, in order, and skip it (no /changelog) as soon as one applies:
- Umbrella entries. A PR's
SOL-XXXXXabsent from theTracked under SOL-…trailers is not a gap if its effect is already captured under an umbrella entry (oneAdded/Changedbullet that rolls up several sub-tickets of one feature and lists their SOLs). Match against the delivered contract change, not a 1:1 SOL-to-bullet mapping. - Delivered ticket, not follow-ups. Match the ticket the PR delivers. SOLs also appear as
inline future-work references (e.g.
[Unreleased]mentions a follow-up SOL) — a body-mention match is a false signal; use the PR's own tracked ticket. tools.yamlprose-only edits. A candidate that matchesSURFACE_REonly throughinternal/composite/definitions/tools.yamlneeds a second gate: inspect the diff for a changed tool name / param / output / step-key. If it changed only description or example prose (no contract change), it is not a gap — skip it (it belongs tono-changelog).- Diff, don't guess. Before drafting, read the candidate PR's diff and confirm an actual
name/param/output/behavior change with no representation in
[Unreleased]. Decide from the diff, never from SOL presence alone.
For each candidate that survives all four checks, invoke the /changelog skill (Skill tool)
with that PR's range to draft the missing entry into [Unreleased]. Halt the whole flow if any
draft left a SOL-???? placeholder — surface it for the human to resolve before anything is
promoted or pushed.
Nothing to release? Stop before promoting. After gap detection and any drafting, if [Unreleased]
still has no entry line (^- ), there is nothing to ship since vPREV. Do not promote an empty block
(it would only fail A5 and the release gate downstream). Report "nothing to release since vPREV" and
stop — and because A1's clean-start path already created release/vX.Y.Z (with no commits yet), leave
no litter behind: git switch - && git branch -D release/vX.Y.Z.
A3. Version. If the version was passed, use it. Otherwise suggest a bump from the [Unreleased]
contents per RELEASING.md (from 1.0: any - **BREAKING**: → MAJOR; new Added → MINOR; only
Fixed/Security → PATCH) and confirm with the human before continuing.
A4. Promote (deterministic). Edit CHANGELOG.md (today = date +%F):
- Rename
## [Unreleased]→## [X.Y.Z] - <today>. - Insert a fresh
## [Unreleased]heading above it (heading only, no empty category stubs). - In
## Links: repoint[Unreleased]to…/compare/vX.Y.Z...HEADand add[X.Y.Z]: …/compare/vPREV...vX.Y.Zbeneath it, matching the existing URL shape. - Touch nothing else — no existing bullets, and not the bottom
## Release Process/## Versioningsections.
A5. Verify the promotion. The release gate now refuses both a missing section and a
heading-only one (no - entry), but it does not check the [Unreleased]/## Links structure — so
assert the rest by hand before proceeding:
.github/scripts/extract-release-notes.sh vX.Y.Z /tmp/release-notes-dryrun.mdsucceeds (it fails fast if the block is missing or entry-less) and the extracted notes contain at least one real entry, not just the## [X.Y.Z]heading.- A fresh
## [Unreleased]heading still exists above the dated block (dropping it silently breaks the next cycle's gap detection and the changelog hook). - The
## Linksblock has a well-formed[Unreleased]: …/compare/vX.Y.Z...HEADand a new[X.Y.Z]: …/compare/vPREV...vX.Y.Zline.
If any check fails, the promotion is malformed — fix before proceeding.
A6. Draft the human-readable release-notes summary. The published GitHub Release should be
skimmable, not the verbatim CHANGELOG block. The pipeline runs on the tag
with no LLM available, so the summary must be authored here and committed in the release PR:
.github/scripts/extract-release-notes.sh publishes .github/release-notes/vX.Y.Z.md when present
and otherwise falls back to the raw block. Write that file:
- Draft from the promoted
## [X.Y.Z]block only — condense each entry to 1–2 sentences, grouped by the same categories (Added/Changed/Fixed/Security/…), dropping rationale and implementation detail. The CHANGELOG remains the source of truth. - Do not introduce facts absent from the block; do not restate the release gate — the script still enforces "the CHANGELOG section exists and has real entries" independently of this file.
Phase B — Open the prepare-release PR
- The
release/vX.Y.Zbranch was already created fromorigin/mainin A1 (never commit release changes onmainitself) — confirm you're on it with the promotion staged. - Commit
CHANGELOG.mdand the.github/release-notes/vX.Y.Z.mdsummary from A6 (nothing else):Prepare release vX.Y.Z, ending the message with theCo-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>trailer. - Push and open the PR with
gh pr create, basemain, body ending with the🤖 Generated with [Claude Code](https://claude.com/claude-code)trailer. Summarize the promotion (version, categories, any gap entries drafted), and state explicitly that merging this PR automatically tags and publishesvX.Y.Z— an immutable public GitHub Release +ghcr.ioimage — so the reviewer knows the merge is the release go-ahead, not just a docs edit. THIRD_PARTY_LICENSES.mdreminder — only when deps actually changed. Detect withgit diff --stat vPREV..HEAD -- go.mod go.sum; if it reports changes, include the reminder to regenerate viago-licenses report ./cmd/serverand name the changed modules (from thego.moddiff). Ifgo.mod/go.sumare untouched, stay silent — do not emit the reminder.- Print the PR URL and tell the human: review + merge to make the changelog official; the release fires automatically on merge.
Phase C — Watch for the merge
This wait is gated on a human merging the PR, so it is unbounded from the skill's side — do not
burn it in a foreground ~30s poll loop (each cycle reloads full context and is the bulk of the
skill's own churn/latency). Instead launch one blocking watch loop as a background command — via
your runner's background execution (e.g. the Bash tool's background mode), not a shell & inside the
snippet — so it sleeps between checks and wakes the model once, on a terminal state, not every cycle:
Extract the state with --jq (don't grep the raw JSON — gh only emits compact JSON when piped,
so a literal "state":"MERGED" match is fragile), and bound the loop mechanically (~30 min at 30s):
tries=0
while :; do
state=$(gh pr view <n> --json state --jq '.state')
if [ "$state" = "MERGED" ]; then
MERGE_SHA=$(gh pr view <n> --json mergeCommit --jq '.mergeCommit.oid'); break
fi
[ "$state" = "CLOSED" ] && { echo "closed unmerged — release cancelled"; break; }
tries=$((tries + 1))
[ "$tries" -ge 60 ] && { echo "PR #<n> still open after ~30 min — stop; re-run to resume"; break; }
sleep 30
done- On
MERGED→MERGE_SHAholds themergeCommitoid; carry it into Phase D. That exact commit is what gets tagged — do not re-derive it from the branch tip. - On
CLOSED(unmerged) → stop; the release is cancelled. - On the ~30 min bound → stop and report: "PR #N is still open at
<url>— nothing was tagged; re-run/cut-release <version>to resume." A dropped session is safe: the A1 remote-state check resumes cleanly (merged-but-untagged → Phase D).
Phase D — Cut the release (guarded auto-tag)
The merge is the go-ahead, but still verify the hard gates before pushing. MERGE_SHA is the
mergeCommit captured in Phase C (on a resumed run with none in hand, get it from
gh pr view <n> --json mergeCommit).
git fetch origin main --tags.- Tag-exists guard — distinguish remote from local-only. If
vX.Y.Zis on the remote (git ls-remote --tags origin "vX.Y.Z"non-empty) → this version is already released; tags are immutable, so do not re-tag — go to Phase E and report the existing release (matching A1's resume behavior). Shipping further changes needs a new version, i.e. a fresh/cut-release. If the tag exists locally only (git tag --listmatches but the remote does not) → a previous push failed after tagging; re-rungit push origin vX.Y.Zto finish, then go to Phase E. - Verify
origin/maincarries the dated block — extract from the remote branch and refuse if the block is empty or contains no entry line (^-). The CI release gate now rejects an entry-less block too, but that fires only after the tag is pushed; catching it here — before an immutable tag exists — avoids a doomed run, and an empty/entry-less block usually means the promotion did not actually merge. Substitute the bare version forv=(e.g.v="0.6.0", neverv0.6.0). The awk whole-record reference is written$(0)— identical to$0in awk, but with no literal$0token for a positional-argument substitution to turn into the version string, and without the\$0escaping that awk itself rejects as a syntax error. Keep the parentheses:git show origin/main:CHANGELOG.md | awk -v v="0.6.0" ' BEGIN { gsub(/\./, "\\.", v) } $(0) ~ ("^## \\[" v "\\]") { p = 1; print; next } p && /^## \[/ { exit } p { print } ' - Confirm the tagged commit is releasable — scoped to the commit, not the branch. Verify
MERGE_SHAis contained inorigin/main(git merge-base --is-ancestor <MERGE_SHA> origin/main), then capture thebuild-and-testrun id for that exact commit and block on it once (don't foreground-poll):
Right after a merge the run is normally stillrun_id=$(gh run list --workflow=build-and-test.yml --commit <MERGE_SHA> \ --json databaseId --jq '.[0].databaseId') gh run watch "$run_id" --exit-status # blocks to completion; exit 0 = success, non-zero = failurein_progress/queued— that is not a failure;gh run watchblocks until it finishes (run it backgrounded), expected ~5–10 min. Proceed only on a zero exit; stop and surface only on an actual failure. - Tag the exact merged commit (not the branch tip —
origin/mainmay have advanced since the merge) and push:git tag vX.Y.Z <MERGE_SHA> git push origin vX.Y.Z
Phase E — Report and hand off
- Confirm the release run actually started for this tag — don't assume. Resolve the run for
vX.Y.Z(gh run list --workflow=release.ymlfiltered to the tag /MERGE_SHA). Distinguish two states so the check doesn't false-alarm: a run record that exists but sits inqueuedis normal — it is waiting on a runner/environment, not swallowed, so wait on it. Only when no run record appears at all within ~2 min of the push should you flag it — tag protection or a disabled workflow can swallow the trigger. - Once a run exists, capture its id and block on it once, rather than a foreground poll loop:
Run it backgrounded; expected ~15–25 min end-to-end, and it may sit inrun_id=$(gh run list --workflow=release.yml --commit <MERGE_SHA> --json databaseId --jq '.[0].databaseId') gh run watch "$run_id" --exit-statusqueuedbefore starting. - Report the run URL and its conclusion. On failure, point the operator at
RELEASING.mdRollback and warn that the container image and its moving pointers (:latest,{major}.{minor}) may already be published onghcr.ioeven though a later job failed — recovery is roll-forward to the next PATCH, never a retag. A red LLM eval job is the earlier case: the image was not pushed and no GitHub Release exists. Re-run the failed jobs withgh run rerun <run-id> --failed. A missing or rotatedLLM_SERVICE_API_KEY, or an empty endpoint or model variable, is fixed in repo settings and then that same rerun. A suite that is wrong, rather than a flake, is a fix onmainand the next patch tag. - On success, print the
RELEASING.md"After pushing the tag" checklist: verifygh release view vX.Y.Zshows five archives +checksums-sha256.txt+ the curated notes, spot-check a binary's--version, and announce once verified. - Optional cleanup (offer, don't force). After a successful release the local checkout is often
still on the merged
release/vX.Y.Zbranch and the remote branch lingers. Offer to tidy up:git switch main && git pull --ff-only origin main, then delete the merged branch local + remote (git branch -d release/vX.Y.Z;git push origin --delete release/vX.Y.Z). Skip silently if the branch was already deleted on merge. - Stop. Do not retry the run or edit the Release.
Rules
- One human gate only: reviewing/merging the changelog PR. Never merge the PR yourself.
- Never invent a SOL ticket number — halt on any
SOL-????placeholder (Phase A2). - Source the production-surface pattern from
.github/scripts/production-surface.sh— the single source of truth shared with the reminder hook and CI gate. Never re-embed the regex. - Only edit the
[Unreleased]heading, the new dated block, and the## Linkslines inCHANGELOG.md. Never rewrite existing bullets or the bottom## Release Process/## Versioningsections. - Never reuse or force-update a tag; never tag when the tagged commit's
origin/mainblock is missing or entry-less, or itsbuild-and-testrun is notsuccess(Phase D). - Tag the exact merge commit captured in Phase C, verified to be an ancestor of
origin/main— never the branch tip or localHEAD. - Do not modify the
/changelogskill — invoke it unchanged for gap drafting.
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.

