When Cloudflare answers idx.co.id with a challenge, curl fails with only
`curl: (22)`, which exits 4 like any other HTTP failure, and every profile
is tried twice. A challenge needs a browser, so no profile can solve it. The
source is monthly, so the publisher only needs one run per month to get
through: treat a challenge as "retry tomorrow" and alert on stale data.
- Add IdxError::CloudflareChallenge (CLOUDFLARECHALLENGE, exit 6), detected
from the cf-mitigated header. curl reports it on stderr through --write-out,
because --fail drops the response.
- A URL with a challenge gets one round of profiles; other failures keep two.
This also covers `stocks flow --source idx`, which now exits 6 on a challenge.
- -v logs which profile fetched each URL.
- Discovery keeps error types, so a challenge survives an announcement miss.
A source answering with something unparseable still is not called a challenge.
- `ownership discover --require-data-page` fails when the data page cannot be
read. Default discover is unchanged: it warns and lists older announcements.
- The publisher and builder use that flag, so a failed data page can no longer
end in a false "up-to-date". A challenge ends in `RESULT: blocked stage=<stage>`,
exit 6; every other failure stays `RESULT: FAILED`.
- The systemd example treats exit 6 as success and runs the freshness check, so
the unit fails on FAILED or on stale data.
- live-smoke reports a challenge as CHALLENGED, a failure and never a pass.
Golden change: completions-bash.golden gains `--require-data-page` in the
discover options, one word on one line. It is the direct result of the new flag;
no other golden file changes.
Live check on 2026-11-05 against release/v0.5.0: 19 passed, 2 failed. Both
failures are real challenges (flow-idx, discover), each exit 6 with empty stdout.
The import, releases and ticker cases did not run, because discover did not pass.
Refs #83
Findings from a Codex (gpt-6.1-sol, xhigh) review of v0.2.4..HEAD, each
verified in the code before fixing:
- high: ticker symbols reached cache file paths unchecked, so a symbol
like `../../REPORT` read (and deleted, if unparseable) .json files
outside the cache, reachable through MCP get_quotes. resolve_symbol now
accepts only ticker characters, and cache keys are encoded as single
safe path components.
- high: sync decided before downloading and replaced the database
without rechecking; an import that landed meanwhile was erased. It now
rechecks the local state before installing, and idx-mcp runs
ownership_sync exclusively with ownership reads.
- replaces_sqlite_sha256 never downgrades history: the client requires
the rebuild to keep at least the local release count, and the
publisher lists a replaced build only then.
- cache writes are atomic (temp file + rename), so concurrent readers
never see, and delete, a partial entry.
- idx-mcp caps symbols (20, deduplicated) and limits, and runs at most 4
tool calls at once.
- each Yahoo client gets its own cookie jar, paired with its crumb.
- ZIP entries (XLSX, KSEI archives) are read with a 256 MiB bound;
non-ASCII archive dates and huge percentages are parse errors instead
of panics or overflow.
- agent-smoke --keep deletes every copied agent config, not only auth
files.
CLI goldens unchanged; the MCP tool golden changes only to state the
new caps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A snapshot republished for the same as-of date (the TRUE ticker fix)
kept the date, latest release, and release count, so sync saw nothing
to do and existing installs kept the bad data without --force.
The manifest gains an optional snapshot.replaces_sqlite_sha256 listing
earlier builds for that date. Queries never write to a synced file, so
a local database byte-identical to a listed build is replaced; one that
matches nothing (local imports or edits) still needs --force. The
publisher fills the list on a same-date rebuild. Older clients ignore
the field.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On 2026-09-25 a snapshot built with --history 5 (15.3 MB) was
published, and `idx ownership sync` failed for every client:
"the response body is larger than request limit: 10485760" (ureq's
default body cap). It was republished by hand with --history 2 (8.2 MB).
- snapshot download: raise the ureq body limit to 1 GiB; the file is
still verified against the manifest's size and SHA-256. The PDF/XLSX
report downloads get a 256 MiB limit for the same reason.
- CLI test: sync a >10 MiB snapshot over HTTP (fails without the fix
with the exact production error).
- build-latest-ownership-snapshot.sh: --max-snapshot-bytes (default
10485760, the limit of released clients <= v0.2.3); an oversized
snapshot fails the build, so the published one stays untouched.
- publish-ownership-snapshot.sh passes --max-snapshot-bytes through;
publisher tests cover the failure and that the release is unchanged.
- docs: size table and guidance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017a5Cv5Z1NzdYKgKnTiAF4E
Addresses review on #29:
- drop errtrace: a failure inside $(...) printed two RESULT lines
- explicit failures go through fail(), which always prints
RESULT: FAILED stage=...; idx check gets its own `preflight` stage;
the never-failing published-manifest read is no longer a stage
- validate the built latest_as_of_date (`// empty` + date regex) so a
corrupt manifest cannot compare as "newer"
- freshness: validate --now (exit 2) and latest_as_of_date (exit 3)
before calling date
- flake: package an explicit script list instead of scripts/*.sh
- docs: GH_TOKEN prerequisite for the packaged publisher
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017a5Cv5Z1NzdYKgKnTiAF4E
The self-hosted publish timer has not produced an upload since it was
enabled (manifest last updated 2026-07-14, a manual run), and failures
were invisible. Make the job safe to run daily and easy to deploy:
- publish-ownership-snapshot.sh: skip the upload when the published
manifest already has the latest as-of date (pre-check via discover
for XLSX sources, post-build check for legacy PDF); --force to
override. Always ends with one line: RESULT: published|up-to-date
<as-of> or RESULT: FAILED stage=<stage> (exit N). gh release create
no longer requires a git checkout.
- check-ownership-snapshot-freshness.sh: exit 1 when the published
latest_as_of_date is older than --max-age-days (default 40).
- flake: packages.ownership-publisher bundles the scripts with a
pinned idx (curl-impersonate + mupdf) and jq/gh/sqlite/curl as
bin/idx-ownership-publish and bin/idx-ownership-freshness, so hosts
need no checkout, nix develop or cargo build at run time. Built in CI.
- contrib/systemd + docs: daily 06:00 UTC timer, run semantics table,
freshness alarm, packaged usage.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017a5Cv5Z1NzdYKgKnTiAF4E
- build-latest-ownership-snapshot.sh: discover with a wider limit and
add --history <n> to import the n previous supported above-1% XLSX
months (one per as-of date, oldest first) before the latest, so
`idx ownership changes` works straight from a synced snapshot.
- publish-ownership-snapshot.sh: pass --history through (for the
clan-private systemd timer via extraArgs).
- publish-ownership-snapshot.yml: `history` input (default 5).
- Docs: publish/self-hosted flow now describes the XLSX source,
observed upload days (2nd-3rd of the month), and --history.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017a5Cv5Z1NzdYKgKnTiAF4E
