kernel-lore-search

Kernel Lore Search

lore.kernel.org archives every kernel mailing list. Its web pages and search sit behind a bot challenge, so an agent reads lore through the public-inbox git mirrors instead: each list is a git repository in which every commit is one message.

When to Use This Skill

Use this skill when...	Use something else when...
Checking whether a problem was reported or patched before	Reading the code itself at a tag: /kernel:source-history
Verifying a quote before citing a maintainer	Searching GitHub issues of a userspace project: github-actions-plugin:github-issue-search
Confirming your own message reached the archive	Writing or sending the report: /kernel:list-report

1. Use the Mirror, Not the Web Interface

WebFetch on lore gets a 403 or a challenge page. Sending a fake user agent to get past the challenge is bot-check circumvention: Claude Code's auto-mode classifier blocks it, and it should not be worked around. Cloning the mirror with plain git is the documented way in.

2. Find the Epochs and Clone

Large lists are split into numbered epochs, newest highest. Probe until one is missing:

for e in 0 1 2 3; do printf 'epoch %s: ' "$e"; git ls-remote "https://lore.kernel.org/linux-btrfs/$e" HEAD 2>&1 | head -1 | cut -c1-12; done
git clone -q --mirror https://lore.kernel.org/linux-xfs/git/0.git linux-xfs-0.git

• Both https://lore.kernel.org/<list>/<N> and https://lore.kernel.org/<list>/git/<N>.git worked as clone URLs.
• Partial clone does not help: --filter=blob:none printed filtering not recognized by server, ignoring and fetched everything. The full linux-xfs epoch 0 holds about 174,000 messages (2006–2026).
• Update an existing clone with git fetch origin; it is fast.

3. Search

Each commit's subject is the message subject, its author the sender, and the raw RFC 822 message is the blob m in its tree.

cd linux-xfs-0.git
git log --all --format='%H %ad %s' --date=short --since=2021-01-23 --until=2021-01-26 | grep -i 'reflink/subvol'
git show <commit>:m | grep -i -m3 '^From:\|^Date:\|^Message-ID:'
git show <commit>:m | grep -i -B1 -A1 'diagnostic we use'

• Subject and date filters run on git log and are fast.
• Body search means reading m blobs: loop git show <c>:m over the candidates from a subject or date filter, or script a full pass for prior-art sweeps (the 2026-11 sweep regex-searched all 174k linux-xfs messages with a Python script; that script is not part of this skill).
• Cross-list search on lore's /all/ index is web-only. Clone the other lists that matter (linux-fsdevel, linux-btrfs, linux-mm) instead.

4. Control-Test Every Negative

"Nobody reported this" gates a report, so prove the search can find things:

• Run a query you know must hit (reflink returned 12,518 linux-xfs messages) and report the count beside the empty ones.
• Before concluding a recent message is absent, look at the newest commits: git log --all -3 --format='%ad %s' --date=iso. In 2026-11 a --since/--grep filter returned nothing for a message sent minutes earlier while git log --all -3 showed it at the top; the filter, not the archive, was wrong.

5. Verify Quotes Before Citing Them

A research agent's summary of a thread is a claim. Before a quote goes into a report: find the message in the mirror, check the From: line, and grep the exact words in its body, excluding lines that start with > (quoted text from someone else). Cite the lore URL, which is the Message-ID without angle brackets: https://lore.kernel.org/<list>/<message-id>/.

6. Confirm a Sent Message Was Archived

After git send-email returns Result: 250, the archive is the proof that the list received it:

git -C linux-xfs-0.git fetch -q origin
c=$(git -C linux-xfs-0.git log --all -1 --format=%H)
git -C linux-xfs-0.git show "$c:m" | grep -m4 -i '^Message-ID:\|^From:\|^To:\|^Cc:'

In 2026-11 it was in the mirror within six minutes of sending. Record the Message-ID and the lore URL where the investigation is tracked.

7. Kernel Bugzilla

bugzilla.kernel.org has a REST API, for example https://bugzilla.kernel.org/rest/bug?product=File%20System&component=XFS&summary=fiemap. Many subsystems, XFS among them, work on the mailing list rather than in bugzilla, so an empty result there is weak evidence. (A research agent used the API in 2026-11; the URL shape here is from its report, not re-run.)

Agentic Optimizations

Context	Command
Which epochs exist	git ls-remote https://lore.kernel.org/<list>/<N> HEAD
Messages by subject in a window	git log --all --format='%H %ad %s' --date=short --since=<d> --until=<d> | grep -i '<words>'
Headers of one message	git show <c>:m | grep -m4 -i '^From:|^Date:|^Message-ID:|^Subject:'
Newest messages (control)	git log --all -3 --format='%ad %s' --date=iso
Permanent link	https://lore.kernel.org/<list>/<message-id-without-brackets>/