dropSlugClashes only compared against what was on disk, so a clip named like an
entity the model extracted wrote both pages and neither could be updated again.
In-batch winners: source over entity over concept. Against the brain, only
entities/concepts are skipped; the source's slug was already chosen free or its
own by resolveSourceSlug, so an existing bad pair no longer locks the source out.
Mutation-checked.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- brain_promote/brain_reject take an optional queue (inbox | needs-review); a
name present in both is refused instead of silently acting on the top-level
note; brain_pending labels every entry with its queue
- parked pdf/txt files are listed (they were invisible); only markdown can be
promoted, the others are moved or rejected
- promote refuses sources/concepts/entities/wiki as a wing
- frontmatterWing reads only a top-level key inside a properly fenced block
- docs/raw-inbox.md: the contract, the three env vars, the known limits
Each behaviour has a test that fails without it (mutation-checked).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- client pattern is matched against the text and against an NFKC-folded,
format-character-free, dash-normalised, whitespace-free form
(non-breaking hyphen, zero-width characters, soft hyphen, line splits,
fullwidth letters). x/text/unicode/norm was already in the module graph as an
indirect dependency; this makes it direct, no new download.
- folder/frontmatter wing names must be ASCII (strings.ToLower folds the Kelvin
sign to 'k') and must not be sources/concepts/entities/wiki; such names are
never proposed as a wing, and a reserved default is a startup error
Mutation-checked.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- a source whose slug is taken by anything but this wing's own page gets a
wing-qualified, stable slug (talk-homelab) instead of failing the ingest
- an extracted entity/concept whose slug belongs to a different page is skipped
with a warning, so one page can no longer lock a source out of re-ingest
- in a wing only source/concept/entity are valid page types; the model (and so
the clip) can no longer choose wiki/<type>/ directories
Mutation-checked; stable-slug choice pinned directly because title resolution
masks it at the Run level.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
FindSlug returned the first hit, so a wing's own page of a name masked another
page of the same name (the retired dirs are searched before the wings, and wings
alphabetically). FindSlugs returns every match and callers refuse any that is
not their own. Found by the legacy-move planner's own test.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Removes the LLM classifier, the substring 'BM25' scorer, the agreement gate,
the backtest and the old migrate/wingmigrate tooling: measured at 2 of 55
pages placed (#208), and the migration was built for the per-wing facts layout
this branch replaces. wing now holds the decision types, the Router and
KnownWings. INGEST_WING_ROUTING=1 builds the Router (default wing
INGEST_DEFAULT_WING=web-clip, client pattern from INGEST_CLIENT_BLOCK or
CLAUDE_INGEST_CLIENT_BLOCK); a bad default or pattern is fatal at startup.
Watcher tests rewritten for the Router, including a deliberate subfolder into
any existing wing, typo'd folders parked, forged-key and edit-while-placing
cases.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- a marker that blocks a note is warned about once, not silently forever
- a marker dated more than 10 minutes ahead (clock skew, a peer) is not
believed and is treated as stale
- every stale marker is removed, not only the last; removal errors are logged
Each has a test that fails without it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
brain_pending lists a parked note under its bare filename, but promote/reject
only read raw/<name>, so the documented approval path did not exist. They now
fall back to raw/needs-review/ (top level wins on a name clash; a symlinked
directory is refused) and promote drops the resolver's proposal keys.
Mutation-checked.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
From the independent review of #210:
- F1: processed/ and failed/ copies are created O_EXCL with -2,-3 suffixes via
createUnique; same-named notes from different folders no longer replace one
another's only surviving copy
- F2: extract.TextInMax / readAllMax cap the bytes actually read (a note can
grow after the Lstat); extract.TextIn gets a 64 MiB default cap
- F3: a regular marker older than its note describes an earlier note of the
same name; it no longer hides the new one (warn + remove); non-regular
markers still count as done (no writes through links)
- F4: the planted-marker-link test is renamed to what it observes
Each has a test that fails without it; the read-cap test uses an endless reader
that errors at 8 MiB so a regression fails instead of exhausting memory.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A clip dropped in raw/ with no other action lands in web-clip, its entity and
concept are the shared pages, raw/ is empty and nothing waits for a human;
client-marked text and a typo'd wing name park; a deliberate subfolder places.
Real pipeline and file layout, model stubbed.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
With Config.Wing a source goes to wiki/<wing>/sources/ (wing/hall frontmatter);
concepts and entities stay in the shared wiki/concepts and wiki/entities, so an
entity cited from two wings is ONE page with both sources (the per-wing facts/
layout duplicated it). A source slug already used anywhere but its own wing's
sources/ is refused (slug is the graph identity). wiki.FindSlug added.
Mutation-checked: collision guard, shared inventory, per-wing source path.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
No model. A subfolder/frontmatter wing must be a plain name of an existing wing
and then places; conflicting signals and client-marker text park; everything
else goes to the default wing. Wings are read from disk per decision.
Mutation-checked.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
MaxNoteBytes (default 64 MiB) is checked on the Lstat size before extraction;
an oversize note stays in raw/, is reported once and not retried until it
changes. The #199 case (marker name over NAME_MAX) is covered by the
remember-on-any-failure retire path and now has a test that fails without it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Reconciled with v0.25.0 semantics:
- markers are only the fallback when the original cannot be removed, and are
now written through the brain root; the refused set (size+mtime keyed) replaces
the path-only 'stuck' set
- #195's tests updated for inbox zero (archive written, original retired, no
marker), and the planted-marker-link test keeps its point: nothing is ever
written through the link, and the note is processed once
- the 'cannot retire' test now counts archive copies; it counted LLM calls,
which pipeline.Run's source-hash check hides (mutation: removing the
remember step left it green)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
From the second independent review of #207:
- parkForReview now skips the delete when the note changed since it was read
(an LLM call sits in that window; Syncthing can deliver a newer version)
- parked copies are created via brainfs (real dirs only, O_EXCL, no symlink
following), so a planted raw/needs-review symlink cannot redirect or hide a
note; this also removes the need for the guard allowlist entry added in
1e14c56, which is reverted
- a '----' line no longer closes the proposal block
Each has a test that failed first and a mutant that fails it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The previous commit was pushed with this guard red: the gate was run with ';'
instead of '&&'. Allowlisting is the guard's own sanctioned mechanism and is
called out for review.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
From the independent review of #207:
- parked notes are created O_EXCL with -2,-3 suffixes (same-named notes from
different subfolders were overwritten and their originals deleted)
- proposed_wing is emitted only if a plain wing name; a note's own proposal_*
keys are dropped so they cannot outrank the resolver's (first-wins reads)
- CRLF notes: wing: key honoured, one valid frontmatter block on parking
- KnownWings skips wiki dirs that are not canonical lower-case names
- adds the wingmigrate dry-run/apply command
Each fix has a test that failed first and a mutant that fails it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Sources are placed by the wing Resolver; concepts/entities inherit the strict
majority wing of the sources that cite them. Collisions, junk (auto-tunnel
logs) and anything unplaceable are Exceptions left in place. Apply writes the
destination O_EXCL then removes the source, only for Auto items.
Mutation-checked (junk guard, majority, slug identity, status gate,
no-overwrite, destination-exists).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The first version fed the text through a path that never carried it into the
file, so it passed under the mutant. A classifier error does carry it; the
test now fails with %q replaced by %s.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Approval is moving the note into raw/<wing>/ (the watcher ingests it there);
a dedicated approve tool is tracked separately. Proposal reason is written
Go-quoted so LLM text cannot start a frontmatter key; pinned by a
mutation-checked test.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Off by default (retired layout unchanged). On: eligibility from
classification.yaml, local ingest LLM as classifier, BM25 as cross-check,
INGEST_WING_THRESHOLD (default 0.9, rejected outside [0,1]). Build failure
is fatal when explicitly enabled.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
With Config.Resolver set, an explicit subfolder/frontmatter wing or an agreeing
classifier+scorer routes the note into that wing; anything else is parked with
its proposal in frontmatter and never extracted. needs-review/ is skipped by
the walk like rejected/. Nil Resolver keeps the retired layout.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Sources -> wiki/<wing>/sources, concepts/entities -> wiki/<wing>/facts with
kind: frontmatter; inventory, title resolution and back-references are
confined to the target wing. Empty Wing keeps the retired layout unchanged.
Mutation-checked: widening the inventory beyond the wing fails the
cross-wing back-reference test; a non-plain wing name is rejected.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A first backtest showed BM25 summed over a wing's top hits favours big wings
(0/4 agreement). Best-page scoring is volume-robust; pinned by a test that
failed before the change.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Explicit subfolder/frontmatter wins; otherwise a classifier and an
independent scorer must agree above a threshold on a known, eligible wing.
Anything else is an Exception for a human. Never creates a wing, never
assigns an ineligible (client/untagged) wing. Mutation-checked: dropping the
eligibility, known-wing, scorer, threshold or conflict check fails a test.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The archive holds the content read at the start; if Syncthing replaced the
note during the (minutes-long) LLM pass, removing the original would drop the
newer content. Compare size+mtime before removing; leave it for the next poll.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
When both Remove and the marker write fail (read-only raw/), remember the
path in memory so later polls skip it instead of writing another archive
copy every tick. Adds a test that a failed archive copy leaves the original
in raw/, mutation-checked against removing the early return.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
After a note is archived into raw/processed/ or raw/failed/, remove the
original from raw/ so raw/ only holds unprocessed input. The .processed /
.failed marker is now only a fallback when the removal itself fails.
Existing markers are still honoured.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Gitea will refuse direct pushes to main in this repo. Rule 5's reviewed lane was enforced on the merge tools
(tbd_ship / pr_merge, gitea-mcp v0.17.0/v0.18.0) but not on a direct commit; this overrides rule 5's "risk/low straight
to main" for this repo and says why. Derived adapter files regenerated where the repo has them.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SZkhi49hfeSqBp5aZJNXsM
Review round 2 on PR #195: the marker writes had no test, so reverting
them to by-path writes, or the refusal to log-only, stayed green. The
link is planted from inside the fake model call, between the walk's
Lstat and the marker write, for both markers and both an outside and
an in-brain relative target; red on all four flips. The marker-entry
skip is now checked for .failed too.
Also pins cleanRel's acceptance of a backslash as a filename byte in
brainfs itself, and stops the buildAndWrite control claiming writes it
does not count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An authenticated REST request to an unregistered path reached the inner
mux, which left Pattern empty, and metrics.Middleware fell back to the
raw URL path: any token holder could create one Prometheus series per
unique path. The fallback is now the fixed label "unmatched"; real routes
keep their pattern. Tests: buildMux with an authenticated random path,
and the middleware unit test; the timing test now routes through a
ServeMux as production does.
Fixes#185.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review round 1 on PR #195:
- A refused marker or archive write left no marker, so the note was
reprocessed every interval (archive copies piling up; an LLM call per
round on the failure path). The marker check now uses Lstat (any entry
at the marker path counts, so a dangling marker link skips the note
instead of looping), and a note whose archive copy or marker write is
refused is remembered (path, size, mtime) and not retried until it
changes.
- cleanRel no longer refuses a backslash: it is a legal filename byte on
Linux, not a traversal, and a raw note named with one was never
archived or marked.
- Tests now reach every converted writer: the log.md append (real
raw/failed so the log is reached, symlinked log.md), a dangling marker
link, the refused-retry loop with a positive control, a backslash name,
a buildAndWrite swap race, and the claude-session archive with in-brain
and outside symlinked archive dirs. The stale lint allowlist entry for
appendWatcherLog is removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three Watch tests flaked under -race and CPU load (3/8 of 8 parallel
x10 runs on main):
- QueueIsBounded slept 100ms before looking for the summary; it now
waits for the summary, then checks the bound.
- DeliversEveryQueuedAlert asserted FIFO order, but the outage ends
asynchronously, possibly mid-flush, so a later item can send first. It
now asserts each removal is delivered exactly once (order is not what
it is named for).
- AccountsForEveryRemoval slept 50ms after the first summary, counted
only "removed 1" and took the last summary's count. It now sums every
"removed N" and every suppression count and waits until they account
for all plants.
Stress: 0/8 failures in three rounds of 8 parallel x -test.count=10 under
-race. Flips still red: unbounded queue, dropped retries, suppressions
not counted.
Fixes#193.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every alert was titled "brain-capture audit degraded", including the
symlink tripwire's and governance-edit alerts. NtfyNotifier.WithTitle
returns a titled copy (nil stays nil); buildNotifier(getenv, title) is
called per source: capture audit (unchanged title), "brain symlink
tripwire", "brain governance edit". Tests assert the Title header.
Fixes#163.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The remaining writers created or rewrote brain files by path:
- pipeline buildAndWrite: os.MkdirAll + os.WriteFile of every page.
- BackfillRefs: os.WriteFile of each concept page; a page swapped for a
symlink after the read was written through (the race test overwrote an
in-brain target before this change).
- watcher: the raw/processed and raw/failed archive copies (os.Create),
the .processed/.failed markers and the log.md entry (os.OpenFile).
- cmd/server: the claude-session archive.
All now go through brainfs PutRegular/AppendRegular (new package-level
PutFile/AppendFile for one-off writers: the watcher's call-site lint bans
.Open calls in that file so raw/ reads stay on OpenRegular).
Stacked on #189 (needs PutRegular/AppendRegular).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New brainfs PutRegular (create or replace), CreateRegular (create only), AppendRegular, Remove and RealParent: missing parents created one component at a time through the root, every existing component a real directory, new files created with O_EXCL and confirmed before writing. brain_write, promote/reject (O_EXCL create, source kept on refusal), StampProvenanceFile, BuildWingIndex, RebuildIndex, AppendLog and the tunnel candidate log use them; provenance.ApplyFile removed.
Fixes#188. Refs #158. Adversarial review: BLOCKED, BLOCKED, PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
StripUnresolvedLinks logs a warning and returns the content unchanged when the link index cannot be built (e.g. an unreadable subdirectory), instead of failing every brain_write/brain_update and losing the learning. Nothing is unwrapped on a partial index.
Fixes#178.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Scan reports special files (not regular, directory or symlink) in Result.Special, so the startup check refuses with a message naming them; Watch removes them through an os.Root after re-checking the type, and reports any it cannot remove (never as removed).
Fixes#179. Adversarial review: BLOCKED, BLOCKED, PASS. Flake on main tracked in #193.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
ReadRegular, ReadRegularMax and WriteRegular require a clean, local path and a real directory at every level above the file, so an in-brain directory symlink (or a ".." after one) can no longer redirect brain_update, brain_tunnel, brain_read/get or the other readers elsewhere in the brain. session.Append opens through an os.Root on the brain dir, refuses a non-directory sessions dir, and writes only after confirming the log path is the regular file it opened.
Fixes#173. Adversarial review: BLOCKED, PASS. Follow-up: #188 (brain_write create path).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
DetectTunnels (inside every brain_write/brain_update via AutoTunnel) walks with brainfs.WalkRegular and reads titles with ReadRegular: a FIFO under wiki/ no longer hangs the write, and a symlinked note's title no longer becomes a tunnel candidate. A wiki/ that is not a real directory yields no candidates; other walk errors still abort.
Fixes#174. Refs #157 (last reader slice). Adversarial review: BLOCKED, PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Ingest pipeline readers (readSourcePage and loadSchema feed the LLM prompt; reviseOrCreate; injectSourceRefs; BackfillRefs list and reads) and classification.Load read through brainfs: symlinked or special files no longer reach the prompt, written pages or the I1 sovereignty config, and a FIFO no longer hangs ingest or startup. A symlinked/special classification.yaml is a hard error; an absent one still yields defaults.
Refs #157. Adversarial review: BLOCKED, BLOCKED, PASS. Follow-ups: #158 (write paths), #179 (FIFO tripwire).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
brain_tunnel accepts only clean wiki/.../*.md paths and reads/rewrites both notes through brainfs; new brainfs.WriteRegular never creates, never writes through a final symlink (Lstat + O_NOFOLLOW|O_NONBLOCK + os.SameFile on the handle, since os.Root follows in-root final symlinks even with O_NOFOLLOW). brain_update resolves like brain_edit (knowledge/ and wiki/ notes only) and reads/rewrites through brainfs. session_log requires a plain-name session id and opens with O_NOFOLLOW|O_NONBLOCK.
Fixes#169. Adversarial review: BLOCKED, BLOCKED, PASS (round 3 by the estate session's reviewer). Follow-ups: #173, #174.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
AssertNoLeakFrom and the search Query race test passed vacuously when
every symlink swap failed or every read errored. Both now count
successful symlink swaps and reads that returned the inside note and
fail if either is zero; the Query race reader no longer swallows
Query's error.
Verified by flipping: a swapper whose os.Symlink always fails turns
notes, wiki and search race tests red; Query on a missing brain dir
turns the search race test red.
Closes#177
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
BuildWingIndex/BuildAllWingIndexes list wings and halls with the new brainfs.ReadDirDirs and ReadDirRegular and read titles with ReadRegular: a symlinked note no longer reaches _index.md, a symlinked wing is refused (nothing written into its target), a symlinked hall is skipped. linkcheck entryIndex walks with WalkRegular, so a symlinked .md cannot make a link resolve, and a linked tree that is not a real directory is skipped with a warning (brain_write still lands).
Refs #157. Adversarial review: BLOCKED, PASS. Follow-up: #178.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
govulncheck 2026-10-04 found called vulns in ingestion (brain-mcp, public),
reached via audit.LokiCentral.Push: golang.org/x/net v0.26.0
(GO-2026-4918 HTTP/2 infinite loop, GO-2026-5026) and golang.org/x/text
v0.29.0 (GO-2026-5970). x/net -> v0.55.0, x/text -> v0.39.0; tidy also
moved x/sync, x/sys, x/term. govulncheck ./... now: 0 called findings
(19 in required modules, none reachable).
Refs: govulncheck 2026-10-04
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The retry landed in 9a232ad but was tested only through startupretry with
fake funcs; nothing exercised chassisauth.NewJWTValidator's real 502 path.
Extracts the main() closure into buildJWTValidator (behaviour unchanged)
and tests it against an httptest issuer: 3x502 then 200 succeeds with 4
discoveries; a persistent 502 stops at exactly the 8-attempt budget with
the last error; cancelling ctx mid-backoff returns context.Canceled.
Flip-verified: Attempts=1 turns all three red.
Lane: risk/medium issue, committed to main because the behaviour change
already shipped in 9a232ad; this is a test plus a pure extraction.
Closes#83
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
graphsync, wiki inventory/index read through brainfs; brainfs refuses an in-brain file swapped in behind a symlink (os.SameFile). Gatekeeper round 2 PASS on d90ce66, CI run 9180 green.
Refs #157
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
vectorstore.Sync walks and reads via brainfs (regular files only, handle-checked), so symlinked notes are no longer embedded and their old chunks are purged. brainfs.WalkRegular requires every component of its root to be a real directory (fixes in-brain root aliasing in Sync and search Wing/Wing+Hall). Walk errors, including a symlinked scan root, abort Sync before the delete pass. brainfs gains Lstat. Directory-swap race recorded as accepted residual (#167).
Refs #157. Adversarial review: BLOCKED, BLOCKED, PASS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
readExcerpt, ReadNote and ReadScoped now read through brainfs (os.Root, regular files only, type checked on the opened handle), so a symlink swapped in after a check is not followed and a FIFO does not hang the reader. ReadRegularMax bounds the read itself with io.LimitReader(max+1). notes.Target carries the brain root.
Refs #157. Adversarial review round 2: PASS (round 1 BLOCKED: unbounded read, rule-9 test names).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Closes#144 (audit line per tools/call) and #145 (origin/written_by stamped on every model-reachable write, surfaced in brain_query/brain_get).
Passed pr-gatekeeper after four rounds of adversarial review, verified by execution: #147
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
The scrubber had a rule for `postgres://user:pass@` and none for the
general `scheme://user:secret@host` form — which is exactly what a git
credential store holds:
https://<user>:<token>@git.d-ma.be
Verified against all 13 DefaultRules with an invented value: it matched
none of them. Not authorization-header, not bearer-token, and not
homelab-env-token, which needs a `_TOKEN=`-style prefix a bare URL does
not have. So such a turn was ingested rather than dropped, against the
stated intent in this file's own doc comment:
Fail-closed by design: we'd rather lose signal than ingest
credentials into a public-readable brain.
It failed closed for the shapes it named and open for this one.
How it surfaced: a test suite in mathias/infra executed a script that
configures git while inheriting the operator's real $HOME, read koala's
live ~/.git-credentials, and printed two live values. ~/.claude/projects
is this watcher's own input, so the scrubber was the last line of defence
for that mistake. A sweep of that tree found 12 occurrences of 2 distinct
credential-shaped URLs across 4 transcripts in 3 sessions — so the shape
had been passing for a while, not just that once. Both credentials are
being rotated (infra#466).
The generalisable point is not "do not read the credential file". That
mistake will recur. It is that this shape must fail closed, because the
scrubber is what decides whether such a mistake is survivable or
permanent.
Ordering: the new rule sits after postgres-uri-with-password so a DSN
still reports the more specific name, the same convention as
authorization-header before bearer-token.
Both directions are tested, because a rule that dropped every URL with a
colon would discard most of the signal this watcher exists to keep. The
{8,} floor on the secret half keeps `https://host:443/path` and
`https://user:@host` out; the `[^:\s/@]+` user half stops a host:port
pair being read as userinfo. Must-not-scrub cases include a host with a
port, a password-less DSN, a colon in a query string, and the literal
printf format string `https://oauth2:%s@%s`.
Closes#142
Refs mathias/infra#466
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
infra's scripts/context-sync.sh is the estate canonical, and it moved in infra
262369b (a SIGPIPE fix on the placeholder-reporting path: `printf | grep -o |
head -1` became `head -1 < <(grep -o ...)`). Every other repo still carried the
old line, so ContextSyncScriptDrift and ContextAdapterDrift were both firing.
Also picks up the canonical AGENT.md change from local-dev a01ee80 — rule 5's
GitOps review lane and the new precedence rule — which every repo's adapters
need regenerated for regardless of the script.
Refs mathias/infra#470
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ContextAdapterDrift was firing for 13 repos. Picks up behaviour rule 7
("Finish what's closable"), which landed in ~/dev/.context/AGENT.md and never
reached any repo's derived files.
Where mcp.json changes, it also picks up root MCP config the repo was stale on:
brain/gitea/infra move from public hostnames to tailnet Serve URLs, and gitea's
bearer changes from GITEA_MCP_TOKEN to DMABE_GITEA_API_TOKEN — gitea-mcp
authenticates by pass-through and 401s on the former, so a harness syncing from
the old file had a dead gitea server and no signal saying so.
No literal secrets: mcp.json carries ${VAR} placeholders only.
CLAUDE.md deliberately unchanged — context-sync writes it from PROJECT.md alone,
since Claude Code walks up to ~/dev/CLAUDE.md for root context.
Refs mathias/infra ContextAdapterDrift
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat(server): say at startup when alerting is unconfigured (#107)
WARN once at startup when BRAIN_NTFY_URL is unset, naming what will not fire (brain_edit_root governance rewrites, capture audit degradation). Rebase of PR #108 (never CI-tested) onto current main, no functional change to alertingWarnings.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
7 review rounds closed 8 distinct bypasses across two mechanism families: fence-detection (field-name substring, block-containment, fence-stripping, CRLF, UTF-8 BOM) and value-extraction (folded YAML block scalar hiding an invalid value). A known v1 gap remains in the block-scalar fold's handling of blank-line-interrupted continuations — documented on checkClaimEnvelope, accepted deliberately (narrow, requires an already-privileged caller deliberately constructing a specific malformed-YAML shape) rather than chased past diminishing returns. Real fix (shared, tested, encoding-normalized frontmatter parser) tracked separately as hyperguild#138.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Closes#125. Skips updated_at/supersedes/supersede_reason in the caller fragment merge loop so they cannot be set via the fragment, only via the intended entry points.
delete_branch_after_merge=true
Co-authored-by: Mathias <mthbqv@gmail.com>
Adds brain_reject (RejectNote + MCP tool) and fixes the watcher to skip raw/rejected/, preventing declined notes from being silently re-ingested and republished.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Removes brain_write legacy domain/type/knowledge-path params (ADR-0012) and migrates the previously-missed cmd/hyperguild CLI caller (hyperguild brain write) from <type> <slug> to <wing> <hall> <slug>, fixing a 100%% regression the first review round caught. Adds TestWrite_CLIPayloadRoundTrips exercising the real Handler.Write over HTTP to pin the wire contract.
Closes#79.
Co-authored-by: Mathias <mthbqv@gmail.com>
Textual-diff conflict detection only (semantic contradiction detection via brain-gardener Judge deferred, confirmed by Mathias, comment 12594). Includes canonicalBlock whitespace-run normalization fix for the disagreement comparison.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
fix(wiki): give the preamble its own field instead of overloading heading=="" (#103)
Round 3 review: verified structurally that the preamble is no longer in the sections slice or the idx map, so it cannot collide with a degenerate `## ` heading. Flip-tested by reverting to the heading=="" model and confirming TestMerge_DegenerateEmptyHeadingDoesNotCollideWithPreamble goes red. Adversarial pass (multiple degenerate headings, frontmatter-only page, unicode around the ## boundary, 3x self-merge idempotency with preamble+degenerate) all held. CI green on 09aeaa9 (run 7820).
Co-authored-by: Mathias <mthbqv@gmail.com>
Two runs of the full build+deploy fired for every push, and the second died:
nothing to commit, working tree clean
❌ Failure - Main Update infra repo
The deploy was fine both times — the second run had nothing left to do. The
cause is the pair of triggers meeting: ci.yml ran on `push` to main AND on a
`v*` tag, and cd.yml keys off "CI succeeded", so `git push origin main` followed
by `git push origin refs/tags/vX` produced two CI runs for one sha and two
deploys behind them. The tag run carries the same github.sha, so it rebuilt the
same image and then found the infra manifest already pointing at it.
Nothing consumed the tag-triggered run: there is no release workflow, and cd.yml
reads github.sha rather than the tag. Dropping `tags:` removes the duplicate
build outright.
The idempotency guard stays anyway, because a manual re-run has the same shape
and should not ring the CI-red alarm for work that is already done — infra#366
turns a red run into a tracked issue, so a false red costs a real triage.
Closes#118
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Removes supervisor:dev, supervisor:build, supervisor:test:smoke, smoke:routing and scripts/smoke-routing.sh -- all pointed at cmd/supervisor (deleted c117634, 2026-05-12) or cmd/routing (iceboxed fb59c39, 2026-07-01). Repoints `build` at hyperguild:build + ingestion:build and adds `build` to `check`, so a task target naming a deleted package fails the gate instead of staying invisible to lint/test/vet.
Refs #120. Item 2 of #120 (internal/config has no non-test callers) is deliberately NOT addressed -- #120 stays open for it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Mathias <mthbqv@gmail.com>
Closes#116.
Deletes config/models.yaml, internal/config/models.go (LoadModels, ModelFor),
its test, the ModelsFile field on Config, and SUPERVISOR_MODELS_FILE from
.env.example. Option A from the issue.
Merge-gate evidence: negative claim independently re-established (no Go, shell,
Dockerfile, k8s, CI or infra-repo caller); task check exit 0 across both
modules; CI run 7374 green on the PR head.
Co-authored-by: Mathias <mthbqv@gmail.com>
AutoTunnel matched note titles case-insensitively, so a one-word title linked
every entry that happened to use that word. Measured on the live corpus: 185
entries carry an auto-written See-also link to wiki/telos/decisions/wrong.md
(title "WRONG") and 136 to .../status.md — the two most-linked nodes in the
whole graph, and both of them artifacts of the English language rather than
references. telos's projects/problems/strategies/goals/beliefs entries account
for another ~60 between them.
That pollution is not cosmetic: inbound-link count is what the duplicate
Fixer picks survivors by, what orphan detection reads, and what brain_graph
shows.
One-word titles now need the title's own casing in the prose ("WRONG", "BERT",
"NLP" — an entity name in a sentence still links). Multi-word titles keep the
case-insensitive match: "risk tier middleware" is a real mention however it is
capitalised, and a three-word collision is not an accident.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The BrainStore is not the only door. POST /write — the HTTP path CLAUDE.md
hands to curl and scripts — calls api.WriteNote directly, so the strip
shipped one commit ago covered brain_write and capture while leaving the raw
door ungated. A gate one door wide is not a gate.
The BrainStore keeps its own call so it can still report Ref.Unlinked to the
MCP caller; running the strip twice is a no-op on content the primitive has
already cleaned.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every brain write accepted [[targets]] that resolve to nothing — skill
names, repo names, agent-local memory-file slugs — and each one became a
permanent [high] broken-link finding. The brain-audit digest grew 19 -> 63
high findings between 2026-08-26 and 2026-09-16; 53 of the 63 were exactly
this, and none are fixable by brain-gardener's BrokenLinkFixer, which only
repoints edit-distance-1 renames. There is no target to repoint to.
StripUnresolvedLinks resolves each link against knowledge/ + wiki/ using
brain-gardener's own resolver semantics (bare slug or wing/hall/slug path
form, code spans excluded) and unwraps the misses to plain text, reporting
them as Ref.Unlinked / the MCP `unlinked` field.
Unwrap rather than reject: a refused write on claudewatcher's unattended
ingest path is a lost learning, and the author's words are worth keeping
even when the link is not. An empty corpus is a no-op, so a misconfigured
brainDir cannot strip every link in a note.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
CLAUDE_INGEST_SKIP_PROJECTS takes filepath.Match globs tested against
the project directory NAME, and a matching directory is skipped during
the directory WALK — so its files are never opened and their content
never enters the process. That is a stronger property than filtering
after reading, and the difference is observable: without SkipDir,
WalkDir lists the directory's entries, and an unreadable one aborts the
entire walk, silently stopping ingestion for every other project.
Globs rather than exact names because a project directory is derived
from the session's working directory, so a git worktree yields an
ephemeral name no exact-match list could keep up with.
An unparseable pattern refuses to run. filepath.Match returns an error
ALONGSIDE a false match, so a bad pattern checked per-file would ingest
the directory while the log complained — the failure nobody notices.
Validated once, up front, in both entry points.
This complements the client-name scrubber rather than replacing it. The
scrubber drops individual turns whose content matches a client
identity; this excludes whole workspaces whose content is confidential
in ways no pattern reliably recognises. Learnings from excluded
sessions are captured deliberately through the session-close ritual
instead.
Recorded caveat: the project directory reflects the session's working
directory, NOT the repositories it touches. A session run from ~/dev
that reads an excluded repo is not excluded by this. The scrubber
remains the control for content; this is a control for workspaces.
A mutation test earned the strongest assertion here: the original tests
could not distinguish SkipDir from filtering after the walk, because
neither opens files. The unreadable-directory test can.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adopting it turned CI red, and my pre-push verification missed why:
`golangci-lint run ./...` does NOT traverse into nested Go modules. This repo
has two — ./go.mod and ./ingestion/go.mod. I linted the root (0 issues, which
CI reproduces exactly) and never linted ingestion/, which has 3 findings:
ingestion/internal/audit/degrading.go:81 missing case capture.AuditCentral (exhaustive)
ingestion/internal/capture/entities.go:37 missing case capture.ZoneUnknown (exhaustive)
ingestion/internal/mcp/handlers.go:29 var int_ should be int (revive)
Those are behavioural questions and a rename, not formatting — the same class
of finding that kept cad, dispatch, ingestion-svc, dma-cli and
email-triage-agent out of the earlier waves. Consistent with that, this repo
goes back in the queue rather than being force-fitted now.
The gofumpt formatting from the same commit is kept: it is independently
correct and already green.
Verification method corrected going forward: a repo with nested modules must
be linted per module, not once from the root.
Canonical lives at infra ci-templates/golangci.yml, distributed by the
infra#309 pattern: byte-identical copies, checksum drift gate in infra. Do
not edit this file here -- edit the canonical and run `task ci:sync`.
Verified locally before pushing, not pushed-and-hoped: golangci-lint v2.11.4
against a Go 1.26.1 toolchain (matching CI) reports 0 issues here.
Also applies gofumpt, which this config enables and the golangci-lint v2
default set does not check at all. That gap is why the drift was invisible:
earlier passes ran gofmt, which is weaker. Mechanical, deterministic,
build+vet verified.
Estate-wide consolidation pass. golangci-lint v2 default set enables NO
formatter, so gofmt drift accumulated silently across the estate -- 8 repos
were carrying it, this one included. Same hole that let two files rot in
dma-cli until they were found by hand (dma-cli#12).
Mechanical and deterministic: `gofmt -w`, nothing hand-edited. Verified
before commit -- gofmt -l now reports 0, go build and go vet both pass.
Clears this repo to adopt the canonical .golangci.yml (infra
ci-templates/golangci.yml, infra#309 pattern) in a later wave.
hyperguild's Taskfile ran context:sync unconditionally in 'check', with no
guard for the CI runner having no ancestor .context/AGENT.md. That made it
silently regenerate AGENTS.md and .context/system-prompt.txt from project
context alone (82 lines vs. the committed, correct 430) and then fail its
own drift assertion against that wrong baseline every run -- a false
positive, not real drift. Verified locally: syncing with the real root
context reproduces the committed 430-line files byte for byte; syncing
without it (as CI does) produces the shorter fallback that triggered this.
Same guard dma-cli and gitea-mcp already carry, same rationale: drift is
now caught either on a developer machine or by infra's estate-wide
context-sync-drift CronJob (ADR-0019), which does have canonical access.
The previous commit's regeneration produced AGENTS.md byte-identical to
CLAUDE.md (82 lines, project-only) instead of the expected root+project
concatenation (430 lines) -- root_block() rendered empty despite the run's
own log claiming "Root context: .../AGENT.md" and "AGENTS.md (root +
project...)". Not reproduced: re-running the identical command
(ROOT_CONTEXT=.../AGENT.md bash scripts/context-sync.sh) immediately after
produces the correct 430-line output every time. Whatever caused the
original miss was transient and is not diagnosed here -- flagging rather
than asserting a cause I have not confirmed.
Caught by the estate-wide context-sync-drift gate's own verification run,
not by trusting the fix commit. Regenerated fresh and confirmed root
content is present before committing this time.
Caught by the new estate-wide context-sync-drift gate (infra, ADR-0019).
This repo's copy had diverged from the canonical established in dma-cli/
infra -- synced to it, dead Cursor/Aider artifacts removed (2026-09-01
harness decision), Taskfile's drift assertion narrowed to match, adapters
regenerated against the current canonical AGENT.md.
Changed:
M .context/system-prompt.txt
M AGENTS.md
M scripts/context-sync.sh
Refs: ADR-0019 (infra)
infra#344. The removed block suppressed
docs/superpowers/plans/2026-04-22-phase4-attempt-wiring.md, and its description
stated why:
"Literal placeholder word REDACTED matched as if it were a token
(verified 2026-08-04: extracted Secret == 'REDACTED' exactly, gitleaks'
curl-auth-header rule matched the placeholder text itself, not a real
credential)"
That is not what the file contained. Line 1054 carried a real 40-character hex
Gitea PAT in an `Authorization: token ...` header — not the eight-character word
REDACTED. Whether the file changed after the check or the check was wrong, the
justification has been guarding the wrong thing since.
Confirmed dead by side-effect before touching it, never by printing it: `GET
/api/v1/user` with that token returns 401. So there is nothing to rotate. It is
redacted here anyway, because a dead credential in a doc is still a working
example of how to write a live one, and git history keeps the original either
way.
Two independent failures had to line up, which is the same pair as infra#256:
1. The `paths` clause made the block far broader than its description. gitleaks
ORs global allowlist clauses and ignores `condition = "AND"`, so the ENTIRE
file was immune — a live token pasted into that doc tomorrow would have been
suppressed on the strength of a note about a different value.
2. The justification was never re-checked against the file it justifies.
With the token redacted the allowlist has nothing left to do, so it is deleted
rather than narrowed. Repo scans clean, and a planted live-shaped credential in
that same file is now caught.
Refs: mathias/infra#344, mathias/infra#256, mathias/infra#317
mathiasandClaude Sonnet 5.5