Review round 2 on PR #194: with the hook made a no-op the test still
passed (no swap, the rollback wrote the real note, victim trivially
untouched). It now asserts the error says the edit could NOT be rolled
back and that the note path is still the swapped-in symlink; red when
the hook does not fire.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review round 1 on PR #194:
- The edit lock did not exclude other processes (also on main): unlock
removes the lock file, so a waiter could wake holding a flock on the
orphaned inode while the next caller locked a fresh file. flockFile now
re-checks after flock that the locked file is still the one at the path
(brainfs LockIsCurrent) and retries otherwise; OpenLock reports a lock
replaced mid-open as ErrLockChanged, which is retried instead of
failing the edit. New cross-process test (3 re-executed processes x 400
rounds, O_EXCL marker under the lock): 0 overlaps; red without the
re-check.
- OpenLock unit test: relative dangling link (only the Lstat pre-check
stops it; os.Root refuses absolute links on its own), live link, FIFO,
directory, behind a symlinked dir, unclean path, positive control.
- Rollback test: a symlink swapped in for the note before the rollback
(test seam beforeStampHook) is refused, target untouched in and outside
the brain.
- The in-brain dangling-lock case uses a relative link and asserts the
refusal names the lock. OpenLock reuses #189's RealParent; RemoveLock is
dropped for brainfs.Remove. The doc comment names the swap residual.
Rebased onto main after #189.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
brain_edit did its own I/O by path around notes.Edit:
- the pre-edit snapshot used os.ReadFile, which hung on a FIFO note held
open by a writer;
- the rollback (when the provenance stamp fails) used os.WriteFile;
- the edit lock "<note>.lock" was opened with O_CREATE by path, so a
planted dangling .lock symlink made a successful edit create a file at
the link's target, in the brain or outside it.
Snapshot is brainfs ReadRegular, rollback WriteRegular, and the lock is
opened and removed through new brainfs OpenLock/RemoveLock: clean path,
real parent directories, an existing lock must be regular, and the opened
handle must be the file at the path (SameFile).
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>
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>
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>
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>
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>
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.
llm.Router was primary-then-fallback (berget -> iguana ollama). Both
have failed together at least once in production (berget 5xx outage +
a cold fallback model exceeding the shared timeout budget), leaving
brain_answer/brain_classify fully down.
Router gains an optional Tertiary *Client, tried only if both Primary
and Fallback fail. Nil-safe — existing 2-tier deployments (no
BRAIN_LLM_TERTIARY_URL) behave exactly as before.
main.go wires it from BRAIN_LLM_TERTIARY_URL/MODEL + GEMINI_API_KEY,
pointed at Google's OpenAI-compatible Gemini endpoint
(generativelanguage.googleapis.com/v1beta/openai/) — llm.Client already
speaks OpenAI-shaped chat-completions, so no new client type needed.
Tests: TestRouter_FallsBackToTertiaryWhenPrimaryAndFallbackFail,
TestRouter_TertiaryNotCalledWhenFallbackSucceeds, TestRouter_AllThreeFail,
TestRouter_NilTertiary (2-tier backward-compat). Full suite green
(both modules, -race), golangci-lint clean.
Deployment env vars (infra repo, separate commit — GitOps, TBD there
too): BRAIN_LLM_TERTIARY_URL, BRAIN_LLM_TERTIARY_MODEL, GEMINI_API_KEY
secretRef.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
First half of brain#21: half_life (volatile|seasonal|durable), applies
(local|homelab|universal), claim_provenance (asserted|inferred).
writeHallNote already passed unknown frontmatter keys through verbatim, so
preservation was free — a test now pins it so a refactor of the merge loop
cannot silently drop the fields. What was missing is validation, and it
fails closed: a value outside the enum is rejected (a lint skips what it
cannot parse, so the note would read as classified while nothing acts on
it); an envelope field without claim_provenance is rejected (a value a
model inferred must never be indistinguishable from one a human asserted,
or a single classifier pass launders guesses into apparent judgement
across the corpus); provenance alone is rejected too.
Validation runs before MkdirAll, so a rejected write leaves nothing behind.
The field is `applies`, not `scope`: scope: has been in use since brain#5
for glossary terms (homelab-core, project:<repo>), a disjoint value space
on the same key. The shipped schema keeps the name.
Notes with none of these fields are byte-identical to before. Bulk
backfill stays out of scope.
Schema documented in brain/CLAUDE.md via brain_edit_root, which produced
the hyperguild#97 audit record for the rewrite.
Refs brain#21, brain#19 (superseded), brain#5, hyperguild#97.
Co-authored-by: Mathias <mthbqv@gmail.com>
brain_edit_root was gated but unobserved: on success it emitted a slog.Warn
and nothing else. That proves a rewrite was allowed, not that one happened —
and brain-sync commits it within 15 minutes authored `brain-sync`, so git
history cannot tell a CLAUDE.md rewrite apart from routine note traffic.
A successful brain_edit_root now emits an ntfy notification and an AuditEntry
through the same sink a capture already uses. notes.Result gains
PrevContentHash so both ends of the change are identifiable.
Path and hashes only in the notification body. Reserved at Internal, not
Confidential, because Confidential makes DegradingSink refuse to record while
loki is down rather than buffer. Observation is best-effort and never fails
the call; a failure to observe logs at ERROR.
Ordinary brain_edit on knowledge//wiki/ is deliberately not audited — git
history already carries the same before/after bytes with a diff.
Closes#97.
Co-authored-by: Mathias <mthbqv@gmail.com>
brain_read returns a note verbatim; brain_edit is str_replace with an
exactly-once match, operating on the tree brain-sync commits from.
An adversarial design review before implementation changed three things:
1. Governance files are not in the model-facing tool. Brain content is
ingested from the outside world and reaches a model's context, and that
model would hold the editor — one poisoned page could rewrite the
CLAUDE.md a .dispatch-allow agent later loads as instructions. And
brain-sync merges with --strategy-option=ours, so the live tree beats a
human's corrective push on the conflicting lines. CLAUDE.md now goes
through brain_edit_root: de-surfaced from tools/list and gated on the
local static principal. The scopes are disjoint — the privileged door
has the smaller reach.
2. Reads needed the same allowlist, and brain_get already lacked it: it
returned verbatim session transcripts under archive/ by path. A test
proved the exposure live before the fix.
3. Canonicalise before allowlisting — knowledge/../.git/hooks/post-commit
would otherwise reach a hook that brain-sync executes in a pod holding
the Gitea push credentials.
Plus overlapping match counting, UTF-8 validation both ways, refusal while
git is mid-operation, flock with the read inside the lock, atomic
same-dir rename, and reindex only when frontmatter is touched.
Refs brain#4.
Co-authored-by: Mathias <mthbqv@gmail.com>
The tickets array was never broken. Every ticket in the reported batch
targeted a client repo, and the I1 gate refuses a confidential capture
through a us-nexus origin at any length — reproduced live: length 1 and
20 both fine for an internal repo, length 1 refused for a client one.
What made it look like an array bug is that the refusal named neither the
offending target nor that a policy had said no, so with 20 tickets in
flight "the array is broken" was a reasonable conclusion. The refusal now
names the targets carrying the effective classification, and tells a
caller who declared confidential themselves exactly that.
Regression tests at length 1 and 20 pin the path that was already
correct; a third pins per-item partial success.
Closes#91 part 1.
Co-authored-by: Mathias <mthbqv@gmail.com>
Four call sites did `s[:limit] + "…"`. Both halves are byte operations:
len() counts bytes and a slice cuts bytes. On Swedish text — å ä ö are
two bytes, an em dash three — the cut lands mid-character and leaves an
orphaned lead byte, producing invalid UTF-8.
It failed loudly in exactly one place: wiki/index.md is written raw to
disk, where every strict decoder then rejects the whole file. The other
three (brain_context excerpts x2, brain_query excerpts) feed MCP
responses, where encoding/json silently substitutes U+FFFD — so those had
been quietly eating the last character of truncated excerpts, with no
error anywhere and nothing to notice.
textutil.Truncate counts runes and appends the marker only when content
was actually dropped; an ellipsis on complete text is a lie about the
content. The test is a property, not an example: it walks the limit
across every offset of a multi-byte string and asserts utf8.Valid at each
one, so a future off-by-one cannot pass by picking a lucky boundary.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BeAp5LTscnz2W7Ubt4eJ6m
mathiasandClaude Opus 5.5